Intelligent Systems Reference Library 78

Konstantina Chrysafiadi
Maria Virvou

Advances in
Personalized
Web-Based
Education
Intelligent Systems Reference Library

Volume 78

Series editors
Janusz Kacprzyk, Polish Academy of Sciences, Warsaw, Poland
e-mail: [email protected]

Lakhmi C. Jain, University of Canberra, Canberra, Australia, and

University of South Australia, Adelaide, Australia
e-mail: [email protected]
About this Series

The aim of this series is to publish a Reference Library, including novel advances and
developments in all aspects of Intelligent Systems in an easily accessible and well
structured form. The series includes reference works, handbooks, compendia, text-
books, well-structured monographs, dictionaries, and encyclopedias. It contains well
integrated knowledge and current information in the field of Intelligent Systems. The
series covers the theory, applications, and design methods of Intelligent Systems.
Virtually all disciplines such as engineering, computer science, avionics, business,
e-commerce, environment, healthcare, physics and life science are included.

More information about this series at http://www.springer.com/series/8578

Konstantina Chrysafiadi · Maria Virvou

Advances in Personalized
Web-Based Education

13
Konstantina Chrysafiadi
Maria Virvou
Department of Informatics
University of Piraeus
Piraeus
Greece

ISSN 1868-4394 ISSN 1868-4408 (electronic)

ISBN 978-3-319-12894-8 ISBN 978-3-319-12895-5 (eBook)
DOI 10.1007/978-3-319-12895-5

Library of Congress Control Number: 2014954596

Springer Cham Heidelberg New York Dordrecht London

© Springer International Publishing Switzerland 2015

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed. Exempted from this legal reservation are brief excerpts
in connection with reviews or scholarly analysis or material supplied specifically for the purpose of
being entered and executed on a computer system, for exclusive use by the purchaser of the work.
Duplication of this publication or parts thereof is permitted only under the provisions of the Copyright
Law of the Publisher’s location, in its current version, and permission for use must always be obtained
from Springer. Permissions for use may be obtained through RightsLink at the Copyright Clearance
Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Preface

This book aims to provide important information to researchers, educators,

and software developers of computer-based educational software ranging from
e-learning and mobile learning systems to educational games, including stand-
alone educational applications and intelligent tutoring systems. In particular, this
book explains how fuzzy logic can be used to automatically model the learning or
forgetting process of a student. Also, it describes an innovative module, which is
responsible for tracking cognitive state transitions of learners with respect to their
progress or non-progress. Therefore, this book shows how personalized tutoring
modeling may be achieved by taking into account either how a student is making
progress in learning or how the student’s knowledge can decrease. In order to make
the student modeling process clear, a review of the literature concerning student
modeling during the past decade is included in a special chapter. This chapter aims
to answer the three basic questions on student modeling: what to model, how and
why. It presents comparative tables that are the results of a 10-year review study
in student modeling. So the particular chapter can be, also, used as a guide for
making decisions about the techniques that should be adopted when designing a
student model for an adaptive tutoring system. However, the work presented in this
book is not limited to adaptive instruction, but can also be used in other systems
with changeable user states, such as e-shops, where consumers’ preferences change
over time and affect one another. Thereby, this book can provide important infor-
mation not only to those interested in educational systems and student modeling,
but also to all researchers and software developers who are interested in user modeling
in any adaptive and/or personalized system.

v
Contents

1 Student Modeling for Personalized Education:

A Review of the Literature. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Student Modeling Techniques and Methods . . . . . . . . . . . . . . . . . . . 3
1.2.1 The Overlay Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.2 User Stereotypes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.3 Models for Misconceptions and Erroneous Knowledge . . . . 5
1.2.4 Machine Learning Techniques. . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.5 Cognitive Theories. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.6 Modeling the Uncertainty of Learning. . . . . . . . . . . . . . . . . . 8
1.2.7 Ontology-Based Student Modeling . . . . . . . . . . . . . . . . . . . . 10
1.3 Student’s Characteristics to Model. . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.1 Knowledge Level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.2 Errors/Misconceptions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.3 Cognitive Features Other Than Knowledge Level. . . . . . . . . 16
1.3.4 Affective Features. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3.5 Meta-Cognitive Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2 Fuzzy Logic in Student Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 An Overview of Fuzzy Logic. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.1 Type-1 Fuzzy Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.2 Interval Type-2 Fuzzy Sets. . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.3 Rule-Based Fuzzy Logic System. . . . . . . . . . . . . . . . . . . . . . 29
2.2.4 Applications of Fuzzy Logic . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3 Fuzzy Logic for Knowledge Representation. . . . . . . . . . . . . . . . . . . 32
2.3.1 Knowledge Domain Representation Using
a Fuzzy Related-Concept Network. . . . . . . . . . . . . . . . . . . . . 36
2.4 A Novel Rule-Based Fuzzy Logic System for Modeling
Automatically the Learning or Forgetting Process of a Student . . . . 45

vii
viii Contents

2.4.1 Integration of the Fuzzy Rules. . . . . . . . . . . . . . . . . . . . . . . . 49

2.4.2 Application of the Presented Rule-Based Fuzzy
Logic System in a Programming Tutoring System . . . . . . . . 50
2.5 Conclusions and Discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3 A Novel Hybrid Student Model for Personalized Education . . . . . . . . 61

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.3 The F.O.S. Hybrid Student Model. . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.3.1 Fuzzy Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.3.2 Overlay Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.3.3 Stereotypes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.4 Operation of F.O.S.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.5 Application of F.O.S. in a Programming Tutoring System . . . . . . . . 72
3.5.1 Fuzzy Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.5.2 Overlay Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.5.3 Stereotypes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.5.4 Cognitive State Transitions of Learners
of the Programming Tutoring System . . . . . . . . . . . . . . . . . . 79
3.6 Examples of Operations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

4 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2 The Evaluation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2.1 The Evaluation Framework PERSIVA. . . . . . . . . . . . . . . . . . 93
4.2.2 The Evaluation Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2.3 The Evaluation Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.2.4 The Evaluation Population. . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.3.1 Learners’ General Satisfaction. . . . . . . . . . . . . . . . . . . . . . . . 97
4.3.2 Learners’ Performance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.3.3 Changes on Learners’ Behavior and Thoughts
About Computer Programming . . . . . . . . . . . . . . . . . . . . . . . 99
4.3.4 Changes on Learners’ Behavior and Thoughts
About E-Learning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.3.5 Results on Learners’ Further Studies. . . . . . . . . . . . . . . . . . . 100
4.3.6 Learners’ Satisfaction About the System’s
Adaptive Responses to Their Needs. . . . . . . . . . . . . . . . . . . . 108
4.3.7 The Validity of the Conclusions Drawn by the
Student Model Concerning the Aspects
of the Students’ Characteristics . . . . . . . . . . . . . . . . . . . . . . . 109
4.3.8 The Validity of the Adaptation Decision
Making of the Student Model. . . . . . . . . . . . . . . . . . . . . . . . . 112
4.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Contents ix

Conclusions and Discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

Appendix A: The Matrixes of the System’s FR-CN. . . . . . . . . . . . . . . . . . . 119

Appendix B: Questionnaires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

Appendix C: Screenshots. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
Abstract

The rapid advances of computer technology and Internet have led to an enormous
growth of interest in the field of e-learning applications. However, e-learning
systems have several shortcomings, which concern adaptivity problems, when
compared with real-classroom education. Therefore, this book aims to provide
important information about adaptivity in computer-based and/or web-based edu-
cational systems. Initially, a literature review on student modeling techniques and
approaches during the past decade is presented. Then, a novel student modeling
approach, which maximizes the effectiveness of learning and adaptivity, is presented.
This book presents how fuzzy logic can be used for offering adaptation and
increasing learning effectiveness in Intelligent Tutoring Systems. In particular,
it presents a hybrid student model, which incorporates a rule-based mechanism
that allows each individual learner to complete the training program in her/his
own learning pace and abilities. The presented student model combines an over-
lay model and stereotypes with fuzzy sets and fuzzy rules. It is responsible for
identifying and updating the learner’s knowledge level for all the domain con-
cepts of the learning material each time. Particularly, each time the learner’s
knowledge level on a domain concept changes, the system has to infer how the
learner’s knowledge level on the related concepts also changes. In this way, the
system ­discovers if the student learns or not, if s/he forgets, if s/he has difficul-
ties in understanding, if s/he assimilates the knowledge. Therefore, the presented
approach models either how learning progresses or how the student’s knowledge
can be decreased. As a result, the system adapts the delivery of the learning mate-
rial to each individual learner’s need and pace. The operation of the presented
approach is based on a Fuzzy Network of Related-Concepts (FNR-C), which is a
combination of a network of concepts and fuzzy logic. It is used to represent the
organization and structure of the learning material as the knowledge dependencies
that exist between the domain concepts of the learning material.
The presented novel approach was fully implemented and evaluated. It was
integrated in a programming tutoring system for the programming language ‘C’.
Students of a postgraduate program in the field of Informatics on the University
of Piraeus, Greece, used the particular system to learn how to program with the

xi
xii Abstract

p­ rogramming language ‘C’. The evaluation results were very encouraging. They
demonstrated that the presented student modeling approach had a positive impact
on the learners’ performance and on the learning process. Furthermore, they
showed that the system made valid and meaningful adaptation decisions. The
gain of the presented approach is that it allows the system to model the student’s
knowledge level and the learning process in a more realistic way. Furthermore, the
particular approach constitutes a novel generic tool, which is able to model the
changeable use states (e.g., knowledge level, preferences, emotions).
Introduction

During the last decades there has been an ever-growing increased interest
in e-learning applications. The reason is either the easy access of e-learning appli-
cations by a large and heterogeneous group of learners at any time and place,
or the challenge to develop an adaptive e-learning system. The goal of each
­web-based educational system is to maximize the effectiveness of learning and
introduce the learning and teaching process of real-classroom education to the
web. However, in real classrooms, human teachers can readjust the instructional
process and their teaching strategy, each time they think that the learning outcomes
of their students fall short of their teaching expectation. Consequently, the chal-
lenge is to develop Web-based educational systems that adapt dynamically to each
individual student for effective delivery of knowledge domain to heterogeneous
student populations.
Learners of web-based educational systems have not only different needs,
but also different learning characteristics. They have different knowledge level,
­cognitive and meta-cognitive abilities, preferences, learning styles, emotions, reac-
tions, etc. An adaptive educational system has to identify each individual student’s
needs and learning characteristics and react accordingly offering effective person-
alization. Therefore, an adaptive educational system has to identify the student’s
learning characteristics, to infer her/his needs and preferences, to deliver the appro-
priate learning material adapted to the student’s needs, to advise the learner and to
provide personalized feedback. In this way, the system facilitates and maximizes
the effectiveness of the learning process.
Each learner has her/his own learning pace, and consequently educational envi-
ronments have to adapt to this. In fact, it is pedagogically ineffective to deliver
the same learning material and provide the same instructional conditions to all
the learners without considering their learning needs and characteristics. Not
all learners should be told to read the same material in the same order. Instead,
learning material should be delivered with respect to students’ knowledge level
and ­personal needs. Furthermore, the developers of personalized and/or adaptive
educational systems have to consider that the learner’s knowledge of a domain

xiii
xiv Introduction

concept is subject to change. Hence web-based educational systems have to rec-

ognize the individual learner’s knowledge level and how this changes, and then
they should provide effective instruction, adapting the delivery of the knowledge
domain to the learner’s learning needs and pace.
A solution to the above problem is the technology of Intelligent Tutoring
Systems (ITSs). ITSs are computer-based tutoring systems, which incorporate arti-
ficial intelligence and thus they can adapt dynamically the content and instruction
to the individual student’s needs and preferences offering a highly personalized
learning experience. The success of adaptation is based on the four modules of the
Intelligent Tutoring System (ITS): the interface module; the knowledge domain
module; the student model; and the tutoring module.
The knowledge domain representation is an important aspect that has to be
specified for offering adaptation. The knowledge domain representation is a
description of expertise in the subject-matter domain of the ITS. The most ­popular
techniques used for knowledge domain representation are: hierarchies, ­network
of concepts, linkage graphs, and concept maps. These techniques are used either
to represent the order in which each domain concept has to be taught or the
­knowledge dependencies that exist between the domain concepts of the learning
material. However, the representation of the relations between concepts is, mainly,
restricted to “part-of”, “is-a”, and prerequisite relations. Yet, there is the need to
represent how the student’s knowledge level on a domain concept is affected by
her/his knowledge level on other related domain concepts. In such cases, the rep-
resentation of this kind of relations of the learning material’s domain concepts
is performed using fuzzy techniques. The combination of a network of concepts
and fuzzy logic creates a Fuzzy Related-Concepts Network (FR-CN). FR-CN is
a network of concepts, which also depicts the knowledge dependencies that exist
between the domain concepts of the learning material. However, the knowledge
domain representation has to be combined with a well-designed student model,
which is responsible for how the system uses the knowledge domain module to
make the right decisions to offer personalized instruction and support to the
learner.
The ability of an ITS to provide adaptivity is based, mainly, on the technology
of student modeling (Devedzic 2006). Student modeling has been introduced in
ITSs, but its use has been extended to most current educational software appli-
cations that aim to be adaptive and personalized. Student modeling allows the
system to identify the students’ needs and leads it to make adaptive instructional
decisions. This means that the system generates hypotheses about students’ needs
based on evidence that has been previously collected silently during the learner’s
interaction with the system. In return, it provides personalized tutoring to each
individual student.
There are a variety of techniques for student modeling. The most widely known
techniques are overlay models and user stereotypes. Other techniques for student
modeling are: perturbation, machine learning techniques, cognitive theories, con-
straint-based model, fuzzy logic techniques, Bayesian networks, ontologies. Each
student modeling technique is appropriate for modeling some particular students’
Introduction xv

characteristics. For example, the overlay model is, usually, used for representing
the student’s knowledge level; stereotypes are preferred to model the student’s
learning styles and preferences; cognitive theories are used for modeling the affec-
tive features of students, etc. Many researchers have used a combination of the
above techniques to model more than one features of the students.
Frequently, student modeling deals with uncertainty. Learning is a complicated
process. It cannot be accurately hypothesized that a learner knows or does not
know a domain concept. For example, a new domain concept may be completely
unknown to the learner but in other circumstances it may be partly known due to
previous related knowledge of the learner. On the other hand, domain concepts,
which were previously known by the learner, may be completely or partly forgot-
ten. Hence, currently they may be partly known or completely unknown. In this
sense, the level of knowing cannot be accurately represented. Finally, the teaching
process itself changes the status of knowledge of a user. This happens due to the
fact that a student learns new concepts while being taught.
In view of the above, the representation of the learner’s knowledge is a moving
target. A solution to this problem is the use of fuzzy logic. Fuzzy logic allows the
system to model either the increase or the decrease on the student’s knowledge
level. In particular, fuzzy logic techniques can be used to model how the learn-
er’s knowledge level on a domain concept of the learning material is affected by
changes in her/his knowledge level on another related concept. This means that
fuzzy logic technique can model the uncertain and inaccurate states of learning
and forgetting.
In view of the above, a novel approach in ITS, which includes fuzzy logic tech-
niques, is presented in the particular book. More specifically, it includes a rule-
based fuzzy logic mechanism in combination with an overlay model and user
stereotypes for providing personalized tutoring to each learner. This mechanism
identifies either the domain concepts that the learner has forgot or the concepts that
s/he has learned. Therefore, the presented fuzzy student model reveals if a student
learns or not, if s/he forgets or if s/he assimilates the learning material and allows
the system to adapt the instruction to each individual student’s learning pace.
The presented novel fuzzy system was fully implemented in a web-based pro-
gramming tutoring system that teaches the programming language ‘C’. The reason
for the selection of the particular knowledge domain is the fact that the need for
adaptivity is crucial in the programming tutoring system. In the domain of com-
puter programming, there are many different programming languages and learners
have different backgrounds and characteristics.
Programming language learners can vary from novice programmers, to more
experienced programmers who know programming languages other than that being
taught. Obviously, while learning a new programming language a novice programmer
has to learn many more domain concepts than does a more experienced programmer,
who already knows the principles and the basic structures of computer programming.
Furthermore, if a learner already knows an algorithm (e.g., calculating the sum of
integers in a ‘for’ loop), there is no need for her/him to learn another similar algo-
rithm (e.g., counting in a ‘for’ loop). Similarly, if a learner knows a programming
xvi Introduction

structure (e.g., one-dimensional arrays), it is ­easier to understand another program-

ming structure (e.g., multidimensional arrays), so this new structure should not be
considered as being completely unknown to the learner. On another occasion, if a
learner’s performance on a domain concept is poor, this suggests that she/he has for-
gotten another relevant domain concept. For example, if a learner has difficulties in
calculating a sum in a ‘while’ loop, her/his knowledge of the previous domain con-
cept of “calculating a sum in a ‘for’ loop” has eroded. In view of these problems,
the presented web-based programming tutoring system incorporates a student model
responsible for identifying and updating the student’s knowledge level, taking the dif-
ferent pace of learning of each individual learner into account.
In particular, the presented fuzzy system retains static information about each
student, such as her/his previous experience on computer programming and the
programming languages that she/he already knows. It also retains dynamic infor-
mation such as errors, misconceptions, and progress. Such kind of information is
gathered during the learner’s interaction with the system. In each learning session,
the system recognizes the learner’s knowledge level and the changes that occur in
the state of her/his knowledge of a domain concept; it then updates the student’s
overall knowledge level according to the knowledge dependencies between the
learning material’s domain concepts and the learner’s progress. The system rec-
ognizes when a new domain concept is completely unknown to the learner, or
when it is partly known due to the learner having previous related knowledge.
Furthermore, it recognizes when a previously known domain concept has been
completely or partly forgotten by the learner. Thus it models either the possible
increase or decrease of the learner’s knowledge. Furthermore, each time it checks
if the learner’s errors were due to possible confusion with features of another pre-
viously known programming language. In this case, the system responds accord-
ingly by adapting instantly the sequence of learning lessons. The personalization
achieved, allows every learner to complete the e-training course on their own pace
and ability.
The presented programming tutoring system was used by the students of a post-
graduate program in the field of Informatics in the University of Piraeus, Greece,
in order to learn how to program in programming language C. For the evalu-
ation of the fuzzy student model approach, the evaluation framework PERSIVA
(Chrysafiadi and Virvou 2013), which includes both questionnaires and observa-
tions through experiments, was used. The evaluation method assessed either the
educational impact (i.e., performance, satisfaction, change of learners’ attitudes) or
the effectiveness and validity of the educational system’s adaptivity are assessed.
The results of the evaluation were very encouraging. They demonstrated that the
system is able to adapt dynamically to each individual learner’s needs by scheduling
the sequence of lessons instantly. This personalization allows each learner to com-
plete the e-training course at their own pace and according to their ability.
The main body of this book is organized into four chapters. The first chapter
concerns a literature review of techniques and applications of student modeling for
personalized education. The second chapter presents an overview of fuzzy logic
Introduction xvii

and describes how fuzzy logic can be used for representation of the k­ nowledge
domain and in student modeling. In that chapter, the mechanism of fuzzy rules is
described. The third chapter presents a novel hybrid student model for ­personalized
education that the book’s authors have created. That student model is ­responsible
for identifying the improvements and the decay of the learner’s knowledge.
Furthermore, in that chapter, the implementation of the presented hybrid s­tudent
model in an integrated programming tutoring system for the programming l­anguage
‘C’ is described. In the last chapter, the evaluation of the novel system is presented
and discussed. Finally, the conclusions drawn from this work are presented.
Chapter 1
Student Modeling for Personalized
Education: A Review of the Literature

Abstract The rapid development of computer technology and e-learning

­reinforces the need of dynamic adaptation to the needs of each individual student.
Adaptation is performed through the student model, which is a crucial module of
an Intelligent Tutoring System. There are many student modeling techniques and
approaches. In this chapter, a review of the literature concerning student modeling
during the past decade is presented. The aim is to answer the three basic questions
on student modeling: what to model, how and why. This chapter presents compara-
tive tables that are the results of a 10-year review study in student modeling. They
reveal either the most common modeled student’s characteristic, or the student
modeling approaches that are preferred in relation to student modeling character-
istics. So, the particular chapter can be, also, used as a guide for making decisions
about the techniques that should be adopted when designing a student model for
an adaptive tutoring system.

1.1 Introduction

“Intelligent Tutoring Systems” (ITSs) are computer-based educational systems that

contain some intelligence and can be used for adaptive learning. Their goal is to maxi-
mize the effectiveness of e-learning as human tutors provide the most effective instruc-
tion at classroom level readjusting each time the instructional process and the teaching
strategy considering the student’s needs and abilities. The design and development of
an ITS is based on techniques that combine theories and models from the computer
science, cognitive science, psychology, learning science, computational linguistics,
artificial intelligence (Nwana 1990; Graesser et al. 2012) (Fig. 1.1). The typical archi-
tecture of an ITS includes the following four modules (Fig. 1.2):
• A knowledge domain model that stores the learning material that is taught to
students.
• A student model that stores information about the learner’s knowledge level,
abilities, preferences and needs.

© Springer International Publishing Switzerland 2015 1

K. Chrysafiadi and M. Virvou, Advances in Personalized Web-Based Education,
Intelligent Systems Reference Library 78, DOI 10.1007/978-3-319-12895-5_1
2 1 Student Modeling for Personalized Education …

Fig. 1.1  An intelligent tutoring system

Fig. 1.2  The typical architecture of an ITS

• A tutoring (pedagogical) model, which makes student diagnosis and controls

the tutoring process and make appropriate instructional decisions based on the
information provided by the other components of the ITS.
• A User Interface that allows the system to interact with the user—learner.
The main feature and advantage of an ITS is its ability to adapt the content and the
presentation of the learning material, the feedback and the instruction process and
strategy to the student’s needs and preferences. To fulfill this, artificial intelligence
1.1 Introduction 3

techniques are applied to the main components of the ITS’ architecture. In other
words, artificial intelligence techniques are applied to the knowledge domain rep-
resentation, to the student model and to the pedagogical model (Badaracco and
Martinez 2013). Researches in ITS include researches in techniques that will make
an ITS to ‘behave’ in a more intelligent way (Conati 2009), which means to diag-
nose the student’s learning status and needs in a more effective way and manage
the instructional and pedagogical strategy as a real domain expert. Many of these
researches have been extended to researches in student model. This has happened
due to the fact that the student model is the base for personalization in computer-
based educational applications.
As a consequence, a crucial factor for designing an effective ITS and/or an
adaptive educational system is the construction of an effective student model. In
order to construct a student model, it has to be considered what information and
data about a student should be gathered, how it will update in order to keep it up-
to-date, and how it will be used in order to provide adaptation (Millán et al. 2010;
Nguyen and Do 2009). In fact, when a student model is constructed, the following
three questions have to be answered:
(i) What are the characteristics of the user we want to model?
(ii) How we model them?
(iii) How we use the user model?
The target of this chapter is to present the student’s characteristics that are usually
modeled. Furthermore, the student modeling techniques that are used in the literature
in relation to each student’s characteristic are presented.

1.2 Student Modeling Techniques and Methods

1.2.1 The Overlay Method

One of the most popular and common used student models is the overlay model.
It was invented by Stansfield et al. (1976) and has been used in many systems ever
since. The reason for its extensive use is the fact that the overlay model can rep-
resent independently the user knowledge for each concept. According to the over-
lay modeling, the student model is a subset of the domain model (Martins et al.
2008; Vélez et al. 2008), which reflects the expert-level knowledge of the subject
(Brusilovsky and Millán 2007; Liu and Wang 2007) (Fig. 1.3). Therefore, the stu-
dent’s knowledge is represented as incomplete but no as incorrect. The incomplete
student’s knowledge is defined by the differences between her/his and the expert’s
set of knowledge (Bontcheva and Wilks 2005; Michaud and McCoy 2004; Staff
2001; Nguyen and Do 2008). According to the overlay student modeling approach,
the knowledge domain is decomposed into individual topics and concepts that
are called elements. Usually, each element is characterized as known or unknown
for the student. However, there are overlay models, in which each element is
4 1 Student Modeling for Personalized Education …

Fig. 1.3  Overlay model

characterized by a qualitative measure (good-average-poor) or a qualitative measure

such as the probability that the student knows the concept (Brusilovsky and Millán
2007). This kind of representation informs the system about the degree to which the
learner knows each domain element.

1.2.2 User Stereotypes

Stereotypes were introduced to user modeling by Rich (1979). The main idea of
stereotyping is to create groups of students with common characteristics. Such
groups are called stereotypes. A learner will be assigned into a related stereo-
type if some of his/her characteristics match the ones contained in the stereotype.
For example, a stereotype model can present the knowledge lever (Fig. 1.4) or the
learning style of a student. In these cases the stereotypes could be {novice, beginner,
knowledgeable, advanced, expert} and {visual, verbal} accordingly. Each stereo-
types has a set of trigger conditions, which activate the stereotype if they are true,
and a set of retraction conditions, which deactivate the stereotype if they are true
to Kay (2000). The stereotype is a particularly important form of reasoning about
users and also student modeling with stereotypes is often a solution for the prob-
lem of initializing the student model by assigning the student to a certain group of
1.2 Student Modeling Techniques and Methods 5

Fig. 1.4  Stereotypes of the learner’s knowledge level

students (Tsiriga and Virvou 2002). An appealing property of the stereotype is that
it should enable a system to get started quickly on its customized interaction with
the user Kay (2000). However, stereotype approach is quite inflexible and error-
prone due to the fact that stereotypes are constructed in a hand-crafted way before
real users have interacted with the system and they are updated only by the sys-
tem’s designer or developer (Kass 1991; Tsiriga and Virvou 2002).

1.2.3 Models for Misconceptions and Erroneous Knowledge

1.2.3.1 Perturbation

A perturbation student model is an extension of the overlay model. It represents

student’s knowledge as a subset of the expert’s knowledge along with her/his
misconceptions (Mayo 2001; Nguyen and Do 2008) (Fig. 1.5). The perturbation
student model is useful for diagnostic reasoning. It allows the system to identify
the student’s erroneous knowledge and wrong rules that s/he has applied and led
her/him to wrong answer (Martins et al. 2008). Thus, the system remediate the
student’s misconceptions providing her/him the appropriate learning material,
advices and feedback.

Fig. 1.5  Perturbation model

6 1 Student Modeling for Personalized Education …

The perturbation model has a collection of mistakes, which is usually called

bug library. The bug library and can be built either by empirical analysis of mis-
takes (enumerative technique) or by generating mistakes from a set of common
misconceptions (generative technique). In enumerative technique, the designers
and analysts of the system determine the possible errors that a student can make
(Smith 1998). In generative modeling the system uses a cognitive model, which
considers students’ behavior, to detect students’ errors (Clancey 1988).

1.2.3.2 Constraint-Based Model

The Constraint-Based Model (CBM) uses constraints to represent both domain and
student knowledge. The knowledge domain is represented as set of constraints and
the student model is the set of constraints that have been violated (Fig. 1.6). A con-
straint has a satisfaction clause and a relevance condition. If the satisfaction clause
becomes false for the relevance condition, then the learner has made an error (Martin
1999). The particular model is based on Ohlsson’s theory of learning from errors
(Ohlsson 1996). According to this theory a learner often makes mistakes when per-
forming a task, even when s/he has been taught the correct way to do it. According to
Mitrovic et al. (2001), the most important advantages of CBM are: its computational
simplicity, the fact that it does not require a runnable expert module, and the fact that
it does not require extensive studies of student bugs as in enumerative modeling.

Fig. 1.6  Constraint-based model

1.2 Student Modeling Techniques and Methods 7

1.2.4 Machine Learning Techniques

The student model is responsible for the identification of the student’s knowledge
level, misconceptions, needs and preferences. This kind of information is obtained
by observing the student’s behavior and action during her/his interaction with the
adaptive and/or personalized tutoring system. The processes of the student’s behavior
observation and reasoning should be made automated by the system. This is achieved
by machine learning techniques. Machine learning concerns the formation and
study of models that allow the system to learn from observation’s data and make
automatically inferences (Webb 1998). Machine learning have so far been used
either to induce a single, consistent student model from multiple observed stu-
dent behaviors, or for the purpose of automatically extending or constructing from
scratch the bug library of student modelers (Sison and Shimura 1998). Therefore,
machine-learning techniques can be used to predict future actions (Webb et al.
2001) and make the system able to adapt the instruction and learning processes to
the student’s needs. An approach of machine learning is the use of artificial neural
networks. They are computational systems inspired by the biological nervous system
of the brain. Artificial neural networks are presented as interconnected networks of
“neurons” that can learn through experience via algorithms.

1.2.5 Cognitive Theories

The adaptive and/or personalized tutoring systems have to integrate pedagogical

and psychological theories, except of artificial intelligence, to be effective. Indeed,
many researchers (e.g. Salomon 1990; Welch and Brownell 2000) have pointed
out that technology is effective when developers thoughtfully consider the merit
and limitations of a particular application while employing effective pedagogi-
cal practices to achieve a specific objective. Pedagogical practices can be inte-
grated in a student model by using cognitive theories, which attempt to explain
human behavior during the learning process. Cognitive theories can model either
the student’s cognitive characteristics like knowledge, attention, ability to learn
and understand and memory or the student’s emotional states and motivation.
Therefore, they contribute significantly to the student’s reasoning trying to under-
stand human’s processes of thinking and understanding.
There are a variety of cognitive theories. Some cognitive theories that have
been used in student modeling are: the Human Plausible Reasoning (HPR)
theory (Collins and Michalski 1989), which is a domain-independent theory that
categorizes plausible inferences in terms of a set of frequently recurring infer-
ence patterns and a set of transformations on those patterns (Burstein and Collins
1988; Burstein et al. 1991); the Ortony et al. (1988) (OCC) theory, which allows
modeling possible emotional states of students, and the Control-Value theory
(Pekrun et al. 2007), which is an integrative framework that employs diverse
factors, e.g. cognitive, motivational and psychological, to determine the existence
8 1 Student Modeling for Personalized Education …

of achievement emotions. The use of cognitive theories in student modeling adds

more “human” reasoning to the computer.

1.2.6 Modeling the Uncertainty of Learning

The processes of learning and student’s diagnosis are complex. They are defined by
many factors and are depended on tasks and facts that are uncertain and, usually,
unmeasured. The determination of the student’s knowledge, mental state and behavior
is not a straightforward task, but it is based on uncertain observations, measurements,
assumptions and inferences. The presence of uncertainty in student’s diagnosis is
increased in an adaptive/personalized tutoring system due to either the indirect inter-
action between the learner and the teacher, or the technical difficulties (Grigoriadou
et al. 2002). The most common used techniques to encounter this kind of uncertainty
are fuzzy logic and Bayesian Networks.

1.2.6.1 Fuzzy Student Modeling

Fuzzy logic was introduced by Zadeh (1965) as a methodology for computing with
words. It is able to handle the uncertainty of learning and student’s diagnosis, which
is based on imprecise data and human decisions, since it encounters the uncer-
tainty problems that are caused by incomplete data and human subjectivity (Drigas
et al. 2009). The core of the fuzzy logic theory is the fuzzy sets, which are used to
describe an element (characteristic, thing, fact or state) and have no concrete limits
(Fig. 1.7). An element can belong to two adjacent fuzzy sets at the same time, but
with different membership degree. For example, a student can be 85 % advanced
(membership degree: 0.85) and 15 % expert (membership degree: 0.15) or 30 %
novice (membership degree: 0.3) and 70 % beginner (membership degree: 0.7)
(Fig. 1.8).
Fuzzy logic can help to improve the adaptation of an intelligent tutoring system.
Fuzzy logic can help the system to decide what is the appropriate instruction model

Fig. 1.7  Fuzzy sets of the students knowledge level

1.2 Student Modeling Techniques and Methods 9

Fig. 1.8  A Bayesian network

each time considering a set of criteria and model specifications (Shakouri and
Menhaj 2008). Chrysafiadi and Virvou (2012) have showed that the integration of
fuzzy logic into the student model of an ITS can increase learners’ satisfaction and
performance, improve the system’s adaptivity and help the system to make more
valid and reliable decisions. Therefore, several researchers have incorporated fuzzy
logic techniques in student modeling.

1.2.6.2 Bayesian Networks

Bayesian Networks is another well-established tool for representing and reasoning

about uncertainty in student models (Conati et al. 2002). BN’s graphical repre-
sentation, sound mathematical foundations and ability to represent uncertainty
using probabilities make them attractive to many researchers (Jameson 1996;
Liu 2008; Desmarais and Baker 2012). Indeed, the presence of capable and
robust Bayesian libraries (e.g. SMILE), which can be easily integrated into the
existing or new student modeling applications, facilitates the adoption of BNs in
student modeling (Millán et al. 2010). A Bayesian Network (BN) is a directed
acyclic graph in which nodes represent variables and arcs represent probabilistic
dependence or causal relationships among variables (Pearl 1988) (Fig. 1.8). The
causal information encoded in BN facilitates the analysis of action sequences,
observations, consequences and expected utility (Pearl 1996). In student
modeling nodes of a BN can represent the different components/dimensions of a
student such as knowledge, misconceptions, emotions, learning styles, motivation,
goals etc.
10 1 Student Modeling for Personalized Education …

1.2.7 Ontology-Based Student Modeling

Recently a lot of research has been done on the crossroad of user modeling and
web ontologies. Due to the fact that the adaptive and/or personalized tutoring sys-
tems attempt to model the teaching and learning processes in real world and the
most of them are web-based applications, they can be combined with web ontolo-
gies. Ontologies support the representation of abstract enough concepts and prop-
erties and make them reused and extended in different application (Clemente et al.
2011). These characteristics of ontologies can help student modeling. The main
advantages of ontology-based student models are: formal semantics, easy reuse,
easy probability, availability of effective design tools, and automatic serialization
into a format compatible with popular logical inference engines (Winter et al.
2005).

1.3 Student’s Characteristics to Model

A significant initial stage of constructing a student model is the selection of appro-

priate students’ characteristics that should be considered and represented. The per-
sonalization is accomplished efficiently by modeling either the domain dependent
student’s characteristics or the domain independent domain student’s character-
istics (Yang et al. 2010). For example, domain dependent student’s characteris-
tics are the knowledge level, the misconceptions, and the prior knowledge. Some
domain independent student’s characteristics are learning style, memory, concen-
tration, and self-assessment. The student’s characteristics are, also, categorized
into static characteristics (like email, age, native language) and dynamic character-
istics (like knowledge level, errors). The static characteristics are set by the student
at the beginning of the learning process, usually through questionnaires, while the
dynamic characteristics are defined and updated each time the student interacts
with the system.
Therefore, the challenge is to define the dynamic student’s characteristics that
constitute the base for the system’s adaptation to each individual student’s needs.
These characteristics include knowledge and skills, errors and misconceptions,
learning styles and preferences, affective and cognitive factors, meta-cognitive
factors (Fig. 1.9).
The student’s characteristics that are usually modeled are:
1. Knowledge level
2. Errors and misconceptions
3. Cognitive features other than knowledge level
4. Affective features
5. Meta-cognitive features
These students’ characteristics are described in detail in the following subsections.
1.3 Student’s Characteristics to Model 11

Fig. 1.9  Dynamic characteristics of a student

1.3.1 Knowledge Level

Knowledge level is the most commonly modeled student’s characteristic.

Knowledge refers to the prior knowledge of a student on the knowledge domain as
well as her/his current knowledge level. The student’s knowledge level is usually
assessed through questionnaires and tests that the student has to complete during
the learning process.
The most commonly used approach for representing the student’s knowledge
level is the overlay student model. During last years many adaptive and/or per-
sonalized tutoring systems have used overlay to represent the student’s knowledge
level. Surjono and Maltby (2003) have modeled the learner’s knowledge using the
overlay approach. Similarly, Kassim et al. (2004) used an overlay student model
in a web-based intelligent learning environment for digital systems (WILEDS) in
order to represent dynamically the emerging knowledge and skills of each student.
Furthermore, in MEDEA (Carmona and Conejo 2004) an estimation of the student
knowledge level for each domain concept is stored in an overlay model. InfoMap
has used an overlay technique to model the knowledge level of children to basic
arithmetic (Lu et al. 2005, 2007). Similarly, an overlay approach has been used
in TANGOW (Alfonseca et al. 2006) for modeling the student’s knowledge.
Also, Kumar (2006a, b) has used an overlay model to represent the knowledge of
students in programming tutors. Glushkova (2008) has applied a qualitative over-
lay student model to represent learners’ knowledge level to DeLC system. LS-Plan
(Limongelli et al. 2009) is another adaptive learning system that has used a quali-
tative overlay student model. An overlay model has been used, also, by Mahnane
et al. (2012). In addition, PDinamet (Gaudioso et al. 2012), which is a web-based
adaptive learning system for the teaching of physics in secondary education, have
12 1 Student Modeling for Personalized Education …

incorporated an overlay student model to represent the student’s knowledge level

in order to provide effective and personalized selection of the appropriate learning
resources.
Another student modeling technique that is usually used to model the learner’s
knowledge level is stereotyping. Examples of adaptive and/or personalized tutoring
systems that have used stereotypes for modeling the student’s knowledge lever are the
following. AUTO-COLLEAGUE (Tourtoglou and Virvou 2008, 2012), which is an
adaptive and collaborative learning environment for UML, represents the level of
students’ expertise through a stereotype-based modeling technique. Furthermore,
Chrysafiadi and Virvou (2008) have developed a stereotyping approach to model
the knowledge level of learners in the programming language Pascal in order to
adapt the system’s responses to each individual student dynamically. Also, a stere-
otype-like approach for modeling the student’s knowledge level is used in Wayang
Outpost, which is a software tutor that helps students learn to solve standardized-
test type of questions, in particular for a math test called Scholastic Aptitude Test,
and other state-based exams taken at the end of high school in the USA, in order to
discern factors that affect student behavior beyond cognition (Arroyo et al. 2010).
Moreover, Durrani and Durrani (2010) have used stereotypes for modeling the
student’s knowledge the adaptive C++ tutor CLT. Finally, Grubišić et al. (2013)
have defined knowledge stereotypes based to model the student’s proficiency in
an adaptive e-learning system called Adaptive Courseware Tutor (AC-ware Tutor).
Another technique of modeling the learner’s knowledge is the Constraint-
Based Model (CBM). Mitrovic (2003) have used the CBM approach to model
the student’s knowledge of a web-enabled intelligent tutoring system that teaches
the SQL database language. Another system that uses CMB for modeling the
student’s knowledge is COLLECT-UML, which is an ITS that teaches object-
oriented design using Unified Modeling Language (Baghaei et al. 2005). Also,
Weerasinghe and Mitrovic (2011) have applied CBM to model the student’s
knowledge in EER-Tutor, which is an ITS that teaches conceptual database design.
Furthermore, BNs have been used for the representation of the student’s
­knowledge. For example, Bunt and Conati (2003) used Bayesian Networks to
detect when the learner is having difficulties in an intelligent exploratory learning
environment for the domain of mathematic functions. A Bayesian student model
was applied in English ABLE for modeling the student’s knowledge in English
grammar (Zapata-Rivera 2007). Furthermore, in TELEOS a Bayesian network
based student model was used in order to explicitly diagnose the student’s knowl-
edge level (Chieu et al. 2010). Similarly, AdaptErrEx has used BNs to model
learners’ skills (Goguadze et al. 2011a, b). Also, INQPRO system predicts the
acquisition of scientific inquiry skills by modeling students’ characteristics with
Bayesian networks (Ting and Phon-Amnuaisuk 2012).
Several student models for learners’ knowledge representation have been built
based on ontologies. For example, MAEVIF (Clemente et al. 2011) and SoNITS
(Nguyen et al. 2011) have used ontologies to model the student’s knowledge.
Also, Peña and Sossa (2010) have adopted a semantic representation and manage-
ment of student models with ontologies in order to represent learners’ knowledge.
1.3 Student’s Characteristics to Model 13

Moreover, Pramitasari et al. (2009) have developed a student model ontology

based on student performance.
However, many adaptive and/or personalized tutoring systems perform mod-
eling of the student’s knowledge by combining different student modeling tech-
niques. Thereby, there are systems, like TADV (Kosba et al. 2003, 2005), which
combine an overlay model with fuzzy techniques to represent the knowledge of
individual students. Stathacopoulou et al. (2005) have used fuzzy techniques to
represent the knowledge and abilities of students to help them to construct the con-
cepts of vectors in physics and mathematics. Another combination of the overlay
technique is with stereotypes. ICICLE (Michaud and McCoy 2004) is an adaptive
tutoring system that attempts to capture the user’s mastery of various grammatical
units and to predict the grammar rules s/he is most likely using when producing
language by combining overlay with stereotypes. A similar combination of stu-
dent modeling techniques has been performed in ELaC (Chrysafiadi and Virvou
2013c), which is a web-based educational system that teaches the programming
language ‘C’.
Also, there are adaptive educational systems that have combines overlay and
stereotypes with fuzzy techniques to model the learner’s knowledge. Examples of
such systems are: INSPIRE (Grigoriadou et al. 2002; Papanikolaou et al. 2003),
which is an intelligent system for personalized instruction in a remote environ-
ment, that models knowledge on a topic classifying it to one of the four levels of
proficiency (insufficient, rather insufficient, rather sufficient, sufficient); DEPTHS
(Jeremić et al. 2012), which is an intelligent tutoring system for learning soft-
ware design patterns, models the student’s mastery; and FuzKSD (Chrysafiadi and
Virvou 2014) that is an e-learning environment for the computer programming.
Other adaptive educational systems that used hybrid student models for rep-
resenting the student’s knowledge level are: KERMIT (Suraweera and Mitrovic
2004), which maintains a constraint-based model and an overlay model;
InterMediActor (Kavčič 2004a) that models the student’s knowledge using over-
lay in combination with ontologies; F-SMILE (Virvou and Kabassi 2002) that
uses a novel combination of the cognitive theory Human Plausible Reasoning
(Collins and Michalski 1989) and a stereotype-based mechanism; and AMPLIA
(Viccari et al. 2009) that models the learner’s knowledge by combining Bayesian
networks with cognitive theories. Furthermore, OPAL (Cheung et al. 2010) and
IWT (Albano 2011) used a combination of overlay and ontologies to model the
learner’s knowledge level.
In addition, a variety of adaptive tutoring systems, like SimStudent (Li et al.
2011) and AIWBES (Homsi et al. 2008), used machine-learning techniques to
observe the student’s behavior and make inferences about her/his knowledge
automatically. Baker et al. (2010) have used a combination of machine learning
technique with Bayesian networks in order to observe students’ reactions and
adjust the instruction automatically to each individual learner. Furthermore, Web-
EasyMath (Tsiriga and Virvou 2002, 2003c), which is a Web-based Algebra Tutor,
uses a combination of stereotypes with the machine learning technique of the dis-
tance weighted k-nearest neighbor algorithm, in order to initialize the model of a
14 1 Student Modeling for Personalized Education …

Table 1.1  Student modeling approaches in relation to knowledge level

Overlay Stereotypes Constraint- Machine Cognitive Fuzzy Bayesian Ontologies
based learning theories techniques networks
model
Knowledge 42.55 % 29.79 % 8.51 % 14.89 % 4.26 % 10.64 % 14.89 % 14.89 %
level

new student. The student is first assigned to a stereotype category concerning her/
his knowledge level and then the system initializes all aspects of the student model
using the distance weighted k-nearest neighbor algorithm among the students
that belong to the same stereotype category with the new student. A combina-
tion of stereotypes with machine learning techniques has been, also, used in Web-
PTV (Tsiriga and Virvou 2003a, b) and GIAS (Castillo et al. 2009) to model the
learner’s knowledge. Moreover, Al-Hmouz et al. (2010, 2011) have applied a
hybrid student model, which combines machine-learning techniques with stereo-
types, to predict the student knowledge.
Therefore, there are a variety of student modeling techniques that can be
used or combined to model the learner’s knowledge. Each one is preferred
in relation with the system’s characteristics and the researchers needs. In
Table 1.1, the percentages of preferences for each one of the student modeling
techniques for modeling the student’s knowledge are presented considering the
above literature review. The information that is derived from the particular table
is the number of the adaptive educational systems that incorporate a particular
student modeling technique for the representation of the learner’s knowledge
level in a set of one hundred adaptive educational systems. For example, if we
have a hundred adaptive educational systems 42.55 of them will use overlay,
29.79 will use stereotypes etc. A system can integrate more than one student
modeling techniques.

1.3.2 Errors/Misconceptions

Knowledge level is not the only the common student’s characteristic that is, usu-
ally, detected and measured through questionnaires and tests. The educational sys-
tem can, also, identify the student’s misconceptions and errors through these tests
as well as observing student’s actions during the learning process. A student’s mis-
conception is an erroneous belief, idea, thought. It is a misunderstanding that is
usually caused by incorrect thinking or faulty facts.
Many researchers of intelligent and/or adaptive educational systems have
modeled the student’s errors and misconceptions in order to provide each indi-
vidual learner with personalized feedback and support. The most commonly used
approach for modeling the learner’s errors and misconceptions is the perturbation
model. Many adaptive educational systems have used the particular student modeling
1.3 Student’s Characteristics to Model 15

technique perturbation to model the student’s errors and misconceptions. In par-

ticular, Surjono and Maltby (2003) have used a perturbation student model to per-
form a better remediation of student mistakes. Furthermore, LeCo-EAD (Faraco
et al. 2004) and InfoMap (Lu et al. 2005) have modeled students’ misconceptions
by using a perturbation model. InfoMap’s perturbation student model involves 31
types of addition errors and 51 types of subtraction errors (Lu et al. 2005). The stu-
dent model of both systems allows the reasoning of students’ errors and helps the
system to expand the explanation during the feedback to the students. Moreover,
Baschera and Gross (2010) have represented through the perturbation approach
the student’s strength and weaknesses, in order to allow for appropriate remedia-
tion actions to adapt to students’ needs. A perturbation student model for detecting
the student’s errors has been used in AUTO-COLLEGE (Tourtoglou and Virvou
2012).
Furthermore, there are adaptive tutoring systems that have used other techniques
than perturbation to model the student’s errors and misconceptions. In particular,
Virvou and Kabassi (2002) have added more “human” reasoning to F-SMILE by
using stereotypes and cognitive theory of Human Plausible Reasoning (HPR)
(Collins and Michalski 1989). F-SMILE reacts accordingly trying to find out the
cause of the problematic situation in which the user is involved when s/he learns
how to manipulate file store of her/his computer. Goel et al. (2012) used a fuzzy
model for student reasoning based on imprecise information coming from the stu-
dent-computer interaction and performed the prediction of the degree of error a stu-
dent makes in the next attempt to a problem. Also, Chrysafiadi and Virvou (2008)
have modeled the type of programming errors that a student can make during
her/his interaction with a web-based educational application that teaches the pro-
gramming language Pascal (Web_Tutor_Pas) using stereotypes. Furthermore, so
KERMIT (Suraweera and Mitrovic 2004) that teaches conceptual database design,
as J-LATTE (Holland et al. 2009) and INCOM (Le and Menzel 2009), which teach
language programming, use the CBM approach to diagnose the student’s errors. In
addition, AdaptErrEx has used BNs to model learners’ misconceptions (Goguadze
et al. 2011a, b). BNs have been, also, used for modeling student’s errors in Andes
(Shapiro 2005). Moreover, Pérez-de-la-cruz (2002) has modeled the student’s
misconceptions applying BNs in combination with cognitive theories.
Therefore, there are a variety of student modeling techniques that can be
used to model the learner’s errors and misconceptions. In Table 1.2, the percent-
ages of preferences for each one of the student modeling techniques for mod-
eling the student’s errors and misconceptions are presented considering the above
literature review. The information that is derived from the particular table is the
number of the adaptive educational systems that incorporate a particular student
modeling technique for the modeling of the learner’s errors and misconceptions
in a set of one hundred adaptive educational systems. For example, if we have
a hundred adaptive educational systems 35.71 of them will use the perturbation
approach, 21.43 will use the Constraint-based model, 21.43 % will use Bayesian
Networks etc.
16 1 Student Modeling for Personalized Education …

Table 1.2  Student modeling approaches in relation to student’s errors and misconceptions

Perturbation Constraint- Stereotypes Cognitive Fuzzy Bayesian
based model theories techniques networks
Errors/ 35.71 % 21.43 % 14.29 % 14.29 % 7.14 % 21.43 %
misconceptions

1.3.3 Cognitive Features Other Than Knowledge Level

Student’s cognitive features are among the most sophisticated student character-
istics that are described in a student model. These features refer to aspects such
as attention, ability to learn and understand, memory, perception, concentration,
collaborative skills, abilities to solve problems and making decisions, analyzing
abilities, critical thinking, learning style and preferences.
Learning style refers to individual skills and preferences that affect how a stu-
dent perceives, gathers and processes learning materials (Jonassen and Grabowski
1993). Some learners prefer graphical representations, others prefer audio materials
and others prefer text representation of the learning material, some students prefer
to work in groups and others learn better alone Popescu (2009). Adapting courses
to the learning preferences of the students has a positive effect on the learning pro-
cess, leading to an increased efficiency, effectiveness and/or learner satisfaction
(Popescu et al 2010). A proposal for modeling learning styles, which are adopted
by many ITSs, is the Felder-Silverman learning style (FSLSM). FSLSM classifies
students in four dimensions: active/reflective, sensing/intuitive, visual/verbal, and
sequential/global (Felder and Silverman 1988; Felder and Soloman 2003). Another
method for modeling learning styles is the Myers-Briggs Type Indicator (MBTI)
(Bishop and Wheeler 1994), which identifies the following eight categories of
learning styles: extrovert, introvert, sensing, intuitive, thinking, feeling, judging,
perceiving.
Many researchers have modeled the student’s learning style and preferences.
Most of them have been used stereotypes for modeling the particular cogni-
tive features. For example, the stereotypes of the student model of INSPIRE
(Grigoriadou et al. 2002; Papanikolaou et al. 2003) provides information about
the learning style of the learner. Furthermore, Surjono and Maltby (2003) have
used stereotypes to model the student’s preferences (i.e. font, colour, illustration)
and learning styles (i.e. competitive, collaborative, avoidant, participant, depend-
ent, independent). Also, Glushkova (2008) has modeled the student’s preferences,
habits and behaviors during the learning process by using stereotypes. Moreover,
Carmona et al. (2008) have used a student model that classifies students in four
stereotypes according to their learning styles. In WELSA (Popescu et al. 2009)
the courses are adapted to the learning preferences of each student applying
stereotyping.
In addition, Salim and Haron (2006) used a combination of fuzzy logic
with a stereotype-like mechanism to model the student’s personality factor
MBTI (Myers-Briggs Type Indicator). AHA! (Stash et al. 2006) and TANGOW
1.3 Student’s Characteristics to Model 17

(Alfonseca et al. 2006) have modeled the student’s learning styles using an overlay
approach. Fuzzy techniques have been, also, used by Stathacopoulou et al. (2005)
for modeling the student’s learning style. They have applied a student model to a
discovery-learning environment that aimed to help students to construct the con-
cepts of vectors in physics and mathematics, which drive pedagogical decisions
depending on the student learning style. Furthermore, Crockett et al. (2013) have
tried to predict learning styles in a conversational intelligent tutoring system using
fuzzy logic. Similarly, Oscal CITS adapts to the student’s learning styles incorpo-
rated a fuzzy mechanism (Latham et al. 2014). Also, TADV (Kosba et al. 2003,
2005) includes a student model, which combines overlay with fuzzy logic, to
represent communication styles of individual students, except of their knowledge.
Moreover, in GIAS (Castillo et al. 2009) the appropriate selection of the
course’s topics and learning resources are based not only on the student’s goals
and knowledge level but also on the student’s learning style that is modeled
using stereotypes and machine learning techniques. In addition, many research-
ers, like Bunt and Conati (2003), Parvez and Blank (2008), Schiaffino et al.
(2008), and Bachari et al. (2011) have been used Bayesian Networks to detect a
student’s learning style and/or preferences automatically. To perform the same
goal, Hernández et al. (2010) have combined Bayesian Networks with cognitive
theories. In addition, Lo et al. (2012) as well as Zatarain-Cabada et al. (2010)
have used artificial neural networks (learning machine) to identify the student’s
cognitive and learning styles correspondingly. Finally, the student’s preferences in
Personal Reader (Dolog et al. 2004) and in the tutoring system of Pramitasari et al.
(2009) have been modeled by using ontologies.
Many attempts to model other cognitive characteristics of students except of
learning styles have, also, made. Conati et al. (2002) have tried to model in Andes
cognitive aspects like long-term knowledge assessment, plan recognition, ability to
solve problems and reading latency using Bayesian Networks. In Web-PTV (Tsiriga
and Virvou 2003a, b), which teaches the domain of the passive voice of the
English language, the carefulness of the student while solving exercises is esti-
mated through a hybrid student model, which combines stereotypes with machine
learning techniques. Furthermore, in F-CBR-DHTS (Tsaganoua et al. 2003) the
diagnosis of students’ cognitive profiles of historical text comprehension was done
with fuzzy techniques and a stereotype-like mechanism. In TELEOS the student’s
cognitive behavior has been explicitly diagnosed through Bayesian Networks
(Chieu et al. 2010). AUTO-COLLEAGUE (Tourtoglou and Virvou 2008, 2012)
uses stereotypes to model the personality of students. Durrani and Durrani (2010)
have considered the student’s cognitive abilities in the adaptive C++ tutor CLT
using stereotypes, also. Jia et al. (2010) have designed an adaptive learning system,
which is based on fuzzy logic and helps learners to memory the content and improve
their comprehension. Peña and Sossa (2010) have used an ontology-based student
model to represent learners’ knowledge, personality, learning preferences and con-
tent, and to deliver the appropriate option of lecture to students.
Furthermore, DEPTHS (Jeremić et al. 2012), which is an intelligent tutoring
system for learning software design patterns, models, except of the student’s
18 1 Student Modeling for Personalized Education …

Table 1.3  Student modeling approaches in relation to student’s learning style and preferences
Overlay Stereotypes Fuzzy Cognitive Machine Bayesian Ontologies
techniques theories learning networks
techniques
Learning 13.64 % 31.82 % 22.73 % 4.55 % 13.64 % 22.73 % 4.55 %
styles and
preferences

mastery, her/his cognitive characteristics by combining overlay with stereotypes

and fuzzy techniques. Also, Mahnane et al. (2012) have used stereotypes to inte-
grate thinking style (AHS-TS) in an adaptive hypermedia system. In addition,
Wang et al. (2009) have built a student model, which is based on machine learning
techniques and represents the learner’s language competence, cognitive character-
istics and learning preferences, in order to assist students in successfully mastering
the English language. Other researchers that have modeled the cognitive char-
acteristics of students are: Jurado et al. (2008), who have used machine learning
techniques in combination with fuzzy techniques; Al-Hmouz et al. (2010, 2011),
who have combined stereotypes with machine learning techniques, and Viccari
et al. (2008), who have built a student model based on cognitive theories and
Bayesian networks.
Therefore, there are a variety of student modeling techniques that can be
used to model the learner’s cognitive features. In Table 1.3, the percentages of
preferences for each one of the student modeling techniques for modeling the
student’s learning styles and preferences are presented considering the above
literature review. Furthermore, in Table 1.4, the percentages of preferences for
each one of the student modeling techniques for modeling the student’s general
cognitive features other than knowledge (including learning styles and prefer-
ences) are presented considering the above literature review. The information
that is derived from the above tables is the number of the adaptive educational
systems that incorporate a particular student modeling technique for modeling
the student’s learning style, preferences and other cognitive features in a set
of one hundred adaptive educational systems. From the data on the tables, it is
concluded that stereotypes is the most popular student modeling technique for
representing the student’s learning styles and other cognitive features (other than
knowledge).

Table 1.4  Student modeling approaches in relation to student’s cognitive features other than
knowledge
Overlay Stereotypes Fuzzy Cognitive Machine Bayesian Ontologies
techniques theories learning networks
techniques
Cognitive 8.33 % 38.89 % 22.22 % 5.56 % 19.44 % 22.22 % 8.33 %
f­eatures other
than knowledge
1.3 Student’s Characteristics to Model 19

1.3.4 Affective Features

The emotional state of a student affects the learning process and the student’s
performance and progress. The emotional state can have a negative or positive
effect on learning. That is the reason why in real classroom settings, experienced
teachers and professors observe and react accordingly to the emotional state of the
students in order to motivate them and improve their learning process (Johnson
et al. 2000; Lehman et al. 2008). Therefore, adaptive and/or personalized educa-
tional systems should detect the emotional state of students and adapt its behavior
to their needs, giving an appropriate response for those emotions (Katsionis and
Virvou 2004).
These emotional factors that influence learning are called affective factors. The
affective states can be the following: happiness, sadness, anger, anxiety, interest,
fear, boredom, frustration, distraction, confusion, tiredness, indifference, concen-
tration and enthusiasm. Some of these emotions, like happiness and concentration,
have positive effect on the learning process. However, other emotions, like bore-
dom, tiredness and distraction, have negative effect on the learning process and
lead students to an off-task behavior (Rodrigo et al. 2007), which are associated,
usually, with deep motivational problems (Baker 2007). Off-task behavior means
that students’ attention becomes lost and they engage in activities that have any-
thing to do with the learning process and aim (Cetintas et al. 2010), like surfing
the web, devoting time to off-topic readings, talking with order students without
any learning aims (Baker et al. 2004). Therefore, the affective factors should be
considered when a student model is built.
Many researchers have used cognitive, pedagogical and psychological theories
in combination with student modeling techniques in order to identify and model
the emotional states of students. In particular, Conati and Zhou (2002) have used
the OCC cognitive theory of emotions (Ortony et al. 1988) for recognizing user
emotions for their educational game prime climb. The same theory has also been used
in a Mobile Medical Tutor (MMT) for modeling possible states that a tutoring agent
may use for educational purposes (Alepis and Virvou 2011). The same researchers
have constructed user stereotypes concerning the users emotional behavior while
they interact with computers (Alepis and Virvou 2006). VIRGE is another ITS-
game, which has adopted OCC theory in order to provide important evidence
about students’ emotions while they learn (Katsionis and Virvou 2004; Virvou
et al. 2005).
A significant attempt to recognize and convey emotions in order to enhance
students’ learning and engagement have been done by Muñoz et al. (2010, 2011) in
PlayPhysics, which is an emotional game-based learning environment for teaching
physics. They have used Bayesian networks in combination with the Control-Value
theory (Pekrun et al. 2007), which is an integrative framework that employs
diverse factors, e.g. cognitive, motivational and psychological, to determine
the existence of achievement emotions. Furthermore, Alepis et al. (2008) have
described a novel mobile educational system that incorporates bimodal emotion
20 1 Student Modeling for Personalized Education …

Table 1.5  Student modeling approaches in relation to student’s affective features

Stereotypes Machine learning Cognitive Bayesian Ontologies
techniques theories networks
Affective 6.67 % 40 % 40 % 33.33 % 6.67 %
features

recognition through a multi-criteria theory. Also, Conati and Mclaren (2009)

developed a probabilistic model of user affect, which recognizes a variety of user
emotions by combining information on both the causes and effects of emotional
reactions. Moreover, Moridis and Economides (2009) have developed a neural net-
work method (machine learning technique) to recognize a learner’s affective state.
Also, Baker (2007) have constructed a machine learning based model that can
automatically detect when a student using an intelligent tutoring system is off-
task, i.e. engaged in behavior, which does not involve the system or a learning
task. Similarly, Cetintas et al. (2010) have performed the automatic detection of
off-task behaviors in intelligent tutoring systems using machine-learning tech-
niques. Furthermore, Balakrishnan (2011) build a student model upon ontology
of machine learning strategies in order to model the effect of affect on learning.
Machine learning techniques have been also used for predicting the emotions of
boredom and curiosity in an Intelligent Tutoring System that is called MetaTutor
(Jaques et al. 2014). Also, Hernández et al. (2010) have applied an affective
student model combining the OCC theory with Bayesian Networks. Inventado
et al. (2010) have used a combination of Bayesian Networks and machine learn-
ing techniques to model the student’s affective features in POOLE III. Finally,
Crystal Island, which is a game-based learning environment, uses Bayesian
Networks to model and predict student affect for improving the learning process
and motivation.
Therefore, there are a variety of student modeling techniques that can be
used to model the learner’s affective features. In Table 1.5, the percentages of
preferences for each one of the student modeling techniques for modeling the
student’s affective features are presented considering the above literature review.
The information that is derived from the particular table is the number of the
adaptive educational systems that incorporate a particular student modeling
technique for modeling the learner’s emotions in a set of one hundred adaptive
educational systems. For example, if we have a hundred adaptive educational
systems 40 of them will use cognitive theories and machine learning techniques,
33.33 will use Bayesian networks etc.

1.3.5 Meta-Cognitive Features

Meta-cognitive features allow the student to be aware of her/his knowledge

and abilities and make her/him able to monitor and direct her/his own learn-
ing processes. In other words, meta-cognition concerns to the active monitoring,
1.3 Student’s Characteristics to Model 21

controlling, regulation and orchestration (Flavell 1976). For example, a student

has meta-cognitive features when s/he is aware of and controls their own thinking;
s/he is able to select her/his own learning goal; s/he can use properly the obtained
and prior knowledge; s/he can choose the appropriate each time problem-solving
strategy (Mitrovic and Martin 2006; Barak 2010). Some meta-cognitive skills are
reflection, self-awareness, self-monitoring, self-regulation, self-explanation, self-
assessment, and self-management (Peña and Kayashima 2011). Metacognitive
features allow the students to participate more actively to her/his own learning
processes. Thereby, adaptive and/or personalized tutoring systems must consider
students’ meta-cognitive skills.
Although the field of meta-cognitive students’ features is a new field of
research, a number of attempts to model the meta-cognitive feature of students
have been made. Conati et al. (2002) have tried to monitor and encourage self-
explanation of students that learn Newtonian physics using Bayesian networks.
Wayang Outpost, which is a software tutor that helps students learn to solve
standardized-test type of questions, have tried to recognize students’ behaviors
related to meta-cognitive factors (Arroyo et al. 2010). Furthermore, Albano (2011)
has presented a model that allows the students to build up their competence in
mathematics concerning meta-cognitive factors. Moreover, Ting et al. (2013)
have modeled student engagement in a computer-based scientific inquiry learning
environment using a Bayesian Networks model. Also, Liaw and Huang (2013) and
Cho and Kin (2013) have investigated learner self-regulation in e-learning envi-
ronment. They have tried to define the learner’s characteristics that affect the
self-regulation.
Meta-cognitive features are very sophisticated and constitute a new field of
research. Their identification is difficult and their definition is complex. There
is not adequate literature review to be able to draw conclusions about either the
learner’s characteristics that determine a meta-cognitive factor or the technique
that is suitable for modeling a particular meta-cognitive feature. That is the reason
for the absence of a comparative discussion about the meta-cognitive features in
relation with student modeling techniques from this subsection.

1.4 Discussion

Student modeling is a research field of the Intelligent Tutoring Systems (ITSs)

that has attracted the interest of many researchers. Although, student modeling
has been introduced in ITSs, its use has been extended to most current educa-
tional software applications that aim to be adaptive and personalized. Therefore,
an attempt to model the student’s characteristics has been made in many adaptive
educational systems. The aim of each adaptive educational system is to model the
most appropriate student’s characteristics in order to carry out the personaliza-
tion efficiently. Hence, the student’s characteristics that are usually modeled are:
the student’s knowledge and misconceptions, her/his preferences and cognitive
22 1 Student Modeling for Personalized Education …

Table 1.6  Combination for compound student model

Overlay Stereotypes Erroneous Machine Cognitive Uncertainty Ontology-
knowledge learning theories models based
models models
Overlay x x x x
Stereotypes x x x x x
Erroneous x x
knowledge
models
Machine x x x x
learning
Cognitive x x x
theories
Uncertainty x x x
models
Ontology- x x x
based
models

features, affective and meta-cognitive factors. The developers of the student model
select the most appropriate each time student-modeling technique or the most
appropriate combination of such techniques to model the above student’s char-
acteristics. The prevailing student modeling techniques that are presented in the
literature are: overlay, stereotypes, perturbation model, constraint-based model,
machine learning techniques, neural networks, cognitive theories, fuzzy logic
techniques, Bayesian networks and ontologies. Many adaptive and/or personal-
ized tutoring systems perform student modeling combing different modeling
techniques to bring together features of different techniques of user modeling.
A compound student model allows the tutoring system to carry out the person-
alization efficiently. Table 1.6 presents the most common combination of student
modeling techniques. In that table the category of erroneous knowledge models
include the perturbation and the Constraint-based model. Also, the category of
uncertainty models includes fuzzy logic techniques and Bayesian networks.
Two questions come of the literature review: (i) “Which student’s characteristics
are the most common-modeled?” (ii) “Which student modeling approaches are pre-
ferred in relation to student modeling characteristics?”. A thorough study and com-
parison of the adaptive and/or personalized tutoring systems that were presented
in this chapter give answers to the above two questions. The presented adaptive
and/or personalized tutoring systems have been developed from 2002 up to now
(2014). Mostly of them (96 %) are results of Scopus, which is the world’s largest
abstract and citation database of peer-reviewed literature. Scopus is considered as
one of the most valid search engine for research papers. Furthermore, a respect-
able number of these systems have been evaluated. The rest systems, which have
not been evaluated, are trends that have not been established yet.
The results of this literature review are very interesting and useful for the
researchers, designers and developers of educational systems and student models.
1.4 Discussion 23

Table 1.7  Which student’s characteristics are preferred for modeling

Knowledge level Error/misconceptions Cognitive features other than knowledge
52.81 % 15.73 % 40.45 %
Affective features Meta-cognitive features
16.85 % 6.74 %

According to that, the most common-modeled student’s characteristic is the

knowledge level and the least common-modeled student’s characteristic is her/
his meta-cognitive features (Table 1.7). The sum of the percentages of Table 1.7
is not 100 %. The reason for this is the fact that a system can model more than
one different student characteristics. Also, many researchers have interested in
modeling student’s cognitive aspects other than knowledge. Furthermore, the
answer to the question “Which student modeling approaches are preferred in rela-
tion to student modeling characteristics?” is given in Table 1.8. The sum of the
percentages of a line of Table 1.8 is not 100 %. The reason for this is the fact
that two different student-modeling techniques can be combined and used in
the same system. For example, a system can combine stereotypes with machine
learning techniques to model the student’s learning style. The information that
is derived from each line of the particular table is the answer to the question: “if
there are one hundred adaptive educational systems how many of them will incor-
porate a particular student modeling technique to model the learner’s character-
istic that corresponds to the table’s line?”. The results of the research (Table 1.8)
demonstrated that: (i) the most common used student modeling technique for the
representation of the student’s knowledge level is the overlay approach; (ii) the
perturbation and constraint-based model (erroneous knowledge models) are pre-
ferred for representing the student’s misconceptions and errors; (iii) uncertainty
models (like fuzzy logic techniques and Bayesian networks) and stereotypes are
preferred for modeling student’s cognitive aspects other than knowledge; (iv) the
uncertainty models are, also, chosen to represent the affective and meta-cognitive
features of the student; (v) the student’s emotions and affective features are very

Table 1.8  Student modeling approaches in relation to student modeling characteristics

Overlay Stereotypes Erroneous Machine Cognitive Uncertainty Ontology-based
(%) (%) knowledge learning theories models (%) models (%)
models (%) (%) (%)
Knowledge 42.55 29.79 8.51 14.89 4.26 25.53 14.89
level
Errors/mis- 0 14.29 57.14 0 14.29 28.57 0
conceptions
Cognitive 8.33 38.89 0 19.44 5.56 44.44 8.33
features
other than
knowledge
Affective 0 6.67 0 40 40 33.33 6.67
features
24 1 Student Modeling for Personalized Education …

often modeled with machine-learning techniques, also. Meta-cognitive features of

students are not presented in Table 1.8, due to the fact that student modeling of
these features is a new field of research and there are not adequate related ref-
erences in literature. Moreover, the ontology-based student model has not been
developed enough, since the research interest in the particular student modeling
approach has started to arise recently.
Chapter 2
Fuzzy Logic in Student Modeling

Abstract The significant development of the e-learning systems has changed the
ways of teaching and learning. In nowadays, everyone can have access to e­ -learning
systems from everywhere. Therefore, the e-learning systems have to adapt the
­learning material and processes to the needs of each individual learner. However,
learning and student’s diagnosis are complex processes, which deal with uncertainty.
A solution to this is the use of fuzzy logic, which is able to deal with uncertainty
and inaccurate data. This chapter explains how fuzzy logic can be used to automati-
cally model the learning or forgetting process of a student, offering adaptation and
increasing the learning effectiveness in Intelligent Tutoring Systems. In particular,
it presents a novel rule-based fuzzy logic system, which models the cognitive state
transitions of learners, such as forgetting, learning or assimilating. The operation of
the presented approach is based on a Fuzzy Network of Related-Concepts (FNR-C),
which is a combination of a network of concepts and fuzzy logic. It is used to rep-
resent so the organization and structure of the learning material as the knowledge
dependencies that exist between the domain concepts of the learning material.

2.1 Introduction

Over the past decade, the rapid development of computer and Internet technolo-
gies has affect a variety of fields of the human’s everyday life. Such a field is the
education. The ways of teaching and learning have been changed and the e-learn-
ing systems and processes have been developed significantly. E-learning systems
offer easy access to knowledge domains and learning processes from everywhere
for everybody at any time. As a result, users of web-based educational systems are
of varying backgrounds, abilities and needs. Therefore, the e-learning systems and
applications have to offer dynamic adaptation to each individual student.
Adaptation is performed through the student model. In particular, the student
model is a core component in any intelligent or adaptive tutoring system that is
responsible for identifying and reasoning the student’s knowledge level, miscon-
ceptions, abilities, preferences and needs. The student model represents many

© Springer International Publishing Switzerland 2015 25

K. Chrysafiadi and M. Virvou, Advances in Personalized Web-Based Education,
Intelligent Systems Reference Library 78, DOI 10.1007/978-3-319-12895-5_2
26 2 Fuzzy Logic in Student Modeling

of the student’s features, such as knowledge and individual traits, so as to be

accessible for offering adaptation (Brusilovsky and Millán 2007). The adaptive
and/or personalized educational system consults the student model and delivers
the learning material to each individual learner with respect to her/his personal
characteristics.
However, student modeling in many cases deals with uncertainty. Learning
and student’s diagnosis are complex. They are defined by many factors and are
depended on tasks and facts that are uncertain and, usually, unmeasured. One pos-
sible approach to deal with this is fuzzy logic, which was introduced by Zaheh
(1965) as a methodology for computing with words in order to handle uncer-
tainty. It encounters the uncertainty problems that are caused by incomplete data
and human subjectivity (Drigas et al. 2009). Chrysafiadi and Virvou (2012) have
showed that the integration of fuzzy logic into the student model of an ITS can
increase learners’ satisfaction and performance, improve the system’s adaptivity
and help the system to make more valid and reliable decisions. Consequently, fuzzy
logic techniques are able to analyze the students’ knowledge level, needs and
behavior and to make the right decision about the instructional model that has to
be applied for each individual learner.
The issue of fuzzy logic and how it can be used in student modeling are pre-
sented in the remainder of this chapter. In particular, an overview of the fuzzy logic
theory and fuzzy sets are described. Also, applications of fuzzy logic in student
modeling are presented. Furthermore, the use of fuzzy logic in the representation
of the knowledge domain of an adaptive and/or personalized tutoring system is
described. In addition, a novel rule-based fuzzy logic system for modeling auto-
matically the learning or forgetting process of a student is presented. Finally, a
brief discussion and the conclusions drawn from this work are presented.

2.2 An Overview of Fuzzy Logic

Fuzzy logic was introduced by Zadeh (1965) to encounter imprecision and uncer-
tainty. It deals with reasoning that is approximate rather than fixed and exact. It is
a precise logic of imprecision and approximate reasoning (Zadeh 1975, 1979). In
other words, fuzzy logic is able to reason and make rational decisions in circum-
stances of imprecision, uncertainty, human subjectivity, incomplete information
and deficient computations (Zadeh 2001).
The basic element of the fuzzy logic theory is the fuzzy set. A fuzzy set
describes a characteristic, thing, fact or state. For example, ‘novice’ is a fuzzy set
that describes the student’s knowledge level, ‘young’ is a fuzzy set that describes
the person’s age, ‘cold’ is a fuzzy set that describes the environment’s tempera-
ture, ‘tall’ is a fuzzy set that describes the person’s height, ‘loud’ is a fuzzy set that
describes the sound’s intensity, ‘close’ is a fuzzy set that describes the distance
between two objects. The fuzzy sets that describe an element have no concrete
limits (Fig. 2.1).
2.2 An Overview of Fuzzy Logic 27

Fig. 2.1  Fuzzy sets and their partitions. a Fuzzy sets for age; b Fuzzy sets for knowledge level;
c Fuzzy sets for height

Fuzzy logic variables have a truth-value that ranges in degree between 0 and 1.
That value declares the degree in which the particular variable belongs to a fuzzy
set. For example, if x is a fuzzy logic variable that describes the student’s knowl-
edge level and its value is 0.6 for the fuzzy set ‘novice’, then it means that the
particular student is considered to be 60 % novice. This value is called degree of
membership or membership value and is symbolized with μ. A fuzzy logic ele-
ment can belong to two adjacent fuzzy sets at the same time, but with different
membership degrees. For example, if a person’s height is 1.72 cm, then according
to the fuzzy sets that are depicted in Fig. 2.1c, the particular person is considered
to be 80 % tall (the membership degree for the fuzzy set ‘tall’ is 0.8) and 20 %
medium (the membership degree for the fuzzy set ‘medium’ is 0.2).
Taking into account the above, the definition of a fuzzy set follows (Fig. 2.2).
Let S be a set of values that represent an element (i.e. S = {1.20, …, 2.10} for

Fig. 2.2  Definition of fuzzy set

28 2 Fuzzy Logic in Student Modeling

height; S = {1, 2, 3, …, 120} for age; S = {0, 1, 2, …, 100} for grades) and x ∈
S. In other words, x is a particular value that belongs to the set S. A fuzzy set FS is
a pair (x, μ(x)), where x ∈ S and μ(x): S → [0, 1]. In other words, for each x ∈ S,
there is a value μ(x) between 0 and 1, which declares the membership degree of x
to the fuzzy set FS.
• If μ(x) = 0, then x is not included in FS
• If μ(x) = 1, then x is fully included in FS
• If 0 < μ(x) < 1, then x is partially included in FS

2.2.1 Type-1 Fuzzy Sets

This first approach of fuzzy sets theory, which points that the value of the mem-
bership function of a fuzzy set can range between 0 and 1, is called type-1 fuzzy
sets. Two common examples of a membership function of type-1 fuzzy sets are
depicted in Fig. 2.3. Type-1 fuzzy sets have been criticized about their ability to
handle uncertainty. It has been advocated that it is not reasonable to use an accu-
rate membership function for something uncertain. Type-1 fuzzy sets used in con-
ventional fuzzy systems cannot fully handle the uncertainties that are present in
intelligent systems (Castillo and Melin 2008). To handle these uncertainties, Lotfi
Zadeh (1975) proposed a more sophisticated kind of fuzzy sets theory that is
called type-2 fuzzy sets (Mizumoto and Tanaka 1976; Mendel 2001).

2.2.2 Interval Type-2 Fuzzy Sets

The concept of a type-2 fuzzy set was introduced first by Zadeh (1975) as an extension
of the type-1 fuzzy set. In particular, the membership function of a general type-2 fuzzy
set is three-dimensional (Fig. 2.4):

Fig. 2.3  Examples of type-1 fuzzy sets

2.2 An Overview of Fuzzy Logic 29

Fig. 2.4  The membership function of a general type-2 fuzzy set

• 1st dimension: the primary variable x (e.g. age, height, grade, temperature)
• 2nd dimension: the primary membership function (PMF), which is a function
and not just a value between 0 and 1.
• 3rd dimension: the secondary membership function (SMF), which is the value
of the membership function at each point on its two-dimensional domain that
is called its footprint of uncertainty (FOU). The value of SMF is, also, range
between 0 and 1.
Using type-2 fuzzy logic can reduce the amount of uncertainty in a system. This is hap-
pened due to the fact that type-2 fuzzy logic offers better capabilities to handle linguis-
tic uncertainties by modeling vagueness and unreliability of information (Liang and
Mendel 2000). Such sets are useful in circumstances where it is difficult to determine
the exact membership function for a fuzzy set, as in modeling a word by a fuzzy set.
When the value of the third dimension is the same (e.g. 1) everywhere, then
the type-2 fuzzy set is called interval type-2 fuzzy set. For an interval type-2 set
the SMF is ignored and only the FOU is used to describe it. The more (less) area
in the FOU the more (less) is the uncertainty (Mendel 2001). The FOU represents
the blurring of a type-1 membership function. It is completely described by its two
bounding functions (Fig. 2.5): (i) a lower membership function (LMF) and (ii) an
upper membership function (UMF).

2.2.3 Rule-Based Fuzzy Logic System

Type-2 fuzzy sets are finding very wide applicability in rule-based fuzzy logic
systems (FLSs). The operation of FLSs is based on rules. The rules are expressed
as a collection of IF-THEN statements (e.g. If George’s grade at mathematics is
65/100, then he is classified to moderate students). Fuzzy sets are associated with
the terms that appear in the antecedents (IF-part) or consequents (THEN-part) of
rules. For example in the example “if George’s grade at mathematics is 65/100,
then he is classified to moderate students”, the fuzzy set ‘moderate’ appears in
30 2 Fuzzy Logic in Student Modeling

Fig. 2.5  The membership function of an interval type-2 fuzzy set

the consequents, while in the example “if the temperature indicates cold, then
the heater must be switched on”, the fuzzy set ‘cold’ appears in the antecedents.
Membership functions are used to describe these fuzzy sets.
Experts construct the rules of a FLS considering their experience or data that
have been extracted from experiments or surveys. Therefore, the knowledge and
data that are used to construct the rules of a FLS are uncertain. This uncertainty
leads to rules that have uncertain antecedents and/or consequents, which in turn
translates into uncertain corresponding membership functions (Karnik et al. 1999).
This uncertainty can be handled using type-2 fuzzy sets.
A type-2 FLS is depicted in Fig. 2.6. Two steps are required to go from an
interval type-2 fuzzy set to a number:
• Type-reduction: in this step an interval type-2 fuzzy set is reduced to an interval-
valued type-1 fuzzy set. This is achieved using particular algorithms. There are a
comparable number of type-reduction methods (Mendel 2001).
• Defuzzification: In this step the centroid of the type-reduced set is computed.
In particular, the average of the two end-points of the finite interval of numbers,
which has been come off the process of type-reduction, is calculated. In other
words, defuzzification maps the type-1 FS that came of the type-reduction step.

Fig. 2.6  A type-2 rule-based fuzzy logic system

2.2 An Overview of Fuzzy Logic 31

2.2.4 Applications of Fuzzy Logic

The ability of fuzzy logic to handle the uncertainty, imprecise and incomplete
data, and information that is characterized by human subjectivity makes it useful
in many human-centric fields. Mendel (2007) has categorizes the applications of
fuzzy logic in: approximation; clustering; control; databases; decision making;
embedded agents; health care; hidden Markov models; neural networks; noise
cancellation; pattern classification; quality control; spatial query; wireless com-
munications. In addition, fuzzy set theory has been applied in education and edu-
cational systems. The applications of fuzzy logic in the educational field can be
categorized into:
• Grading systems: Fuzzy logic is used to define the grade (as a letter, as a num-
ber, or as a percentage) that characterized the student’s level of achievement.
Examples of fuzzy applications in grading systems are the researches of (Bai
and Chen 2006a, 2006b, 2008; Biswas 1995; Cheng and Yang 1998; Echauz and
Vachtsevanos 1995; Law 1996; Wang and Chen 2006; Wilson et al. 1998).
• Student’s evaluation: It includes an overall assessment of the student’s learn-
ing. In particular, it is a complex process that includes student’s performance,
abilities, skills and learning characteristics. Some of the fuzzy logic applications
in the process of the student’s evaluation, which appear in the literature, are
the following: (Chang and Sun 1993; Chen and Lee 1999; Ma and Zhou 2000;
Nykänen 2006; Weon and Kim 2001).
• Learning adaptation: Learning and teaching are complex processes that have
to consider each individual student’s characteristics and abilities in order to be
effective. The educational systems have to adapt dynamically to each individual
learner’s needs and abilities. Many researchers (Alves et al. 2008; Jili et al. 2009;
Jurado et al. 2008; Kosba et al. 2003; Suarez-Cansino and Hernandez-Gomez
2008) have used fuzzy logic for providing learning and teaching adaptation.

2.2.4.1 Applications of Fuzzy Logic in Student Modeling

The aim of the adaptive and/or personalized tutoring systems is to readjust each
time the instructional process and the teaching strategy considering the student’s
needs and abilities. This operation is based on human subjectivity and concep-
tualizations. That is the reason for the need of fuzzy logic. Therefore, there are
many researchers that have used fuzzy logic techniques in student modeling to
deal with uncertainty in the student’s diagnose. For example, Xu et al. (2002) have
used fuzzy models to represent a student profile in order to provide personalized
learning materials, quiz and advices to each student. Furthermore, Kavčič (2004a)
have succeeded to provide personalization of navigation in the educational con-
tent of InterMediActor system through the construction of a navigation graph and
the adoption of fuzzy logic into student reasoning. A fuzzy-based student model
has been applied, also, by Stathacopoulou et al. (2005) to a discovery-learning
32 2 Fuzzy Logic in Student Modeling

environment that aimed to help students to construct the concepts of vectors in

physics and mathematics. The particular fuzzy-based student model allows the
diagnostic model to some extent imitate teachers in diagnostic students’ charac-
teristics, and equips the intelligent learning environment with reasoning capa-
bilities that can be further used to drive pedagogical decisions depending on the
student learning style. Moreover, Jia et al. (2010) have applied fuzzy set theory
to the design of an adaptive learning system in order to help learners to memory
the content and improve their comprehension. Also, Goel et al. (2012) have used a
fuzzy student model for facilitating the student reasoning process, which is based
on imprecise information coming from the student-computer interaction, and pre-
dicting the degree of error that a student is possible to make in the next attempt
to a problem. In addition, Salim and Haron (2006) have provided a personalized
learning environment that exploit pedagogical model and fuzzy logic techniques.
Other educational systems that have incorporated fuzzy logic techniques into the
student model are: F-CBR-DHTS (Tsaganoua et al. 2003); TADV (Kosba et al.
2003, 2005) and DEPTHS (Jeremić et al. 2012).

2.3 Fuzzy Logic for Knowledge Representation

The knowledge domain module is one of the most major modules of an Intelligent
Tutoring System (ITS). The knowledge domain representation is the base for the
representation of the learner’s knowledge, which is usually performed as a subset
of the knowledge domain. It contains a description of the knowledge or behaviors
that represent expertise in the subject-matter domain the ITS is teaching. In other
words, the knowledge domain module is responsible for the representation of the
subject matter taking into account the course modules, which involve domain
concepts. The particular module has been introduced in ITS but its use has been
extended to most current educational software applications that aim to be adaptive
and/or personalized.
To enable communication between system and learner at content level, the
domain model of the system has to be adequate with respect to inferences and
relations of domain entities with the mental domain of a human expert (Peylo
et al. 2000). Therefore, the knowledge domain representation in an adaptive and/
or personalized tutoring system is an important factor for providing adaptivity. The
appropriate approach for knowledge representation makes easier the selection of
the appropriate educational material satisfying the student’s learning needs. The
most common used techniques of knowledge domain representation in adaptive
tutoring systems are hierarchies and networks of concepts.
A hierarchical knowledge representation is usually used in order to specify
the order in which the domain concepts of the learning material have to be taught
(Chen and Shen 2011; Siddara and Manjunath 2007; Vasandani and Govindury
1995), and can be implemented through trees (Kumar 2005; Geng et al. 2011).
For example, in INMA, which is a knowledge-based authoring tool for music
2.3 Fuzzy Logic for Knowledge Representation 33

education, the knowledge domain is described in terms of hierarchies (Virvou

et al. 2006). Also, Siddappa et al. (2009) have developed a multilevel hierarchical
model for the representation of knowledge domain of an intelligent tutoring system
for numerical method (ITNM). This multilevel hierarchical model was based on
various aptitude levels of students. An example of hierarchical representation is
depicted in Fig. 2.7.
Hierarchies give information about the order in which the learning material
should be taught, but they do not clearly depict the relations among the domain
concepts. The network of concepts gives this kind of information. In a network of
concepts, nodes represent concepts and arcs represent relations between ­concepts
(Fig. 2.8). Many adaptive tutoring systems, such as Web-PTV (Tsiriga and Virvou
2003a, 2003b), DEPTHS (Jeremić et al. 2009) and IDEAL (Khamis 2011) use
a network of concepts for representing the knowledge domain. However, in a
­network of concepts the relations between concepts are restricted to “part-of”,
­“is-a” and prerequisite relations. They do not depict how the knowledge of a
domain concept may be affected by the knowledge of another concept. They do
not give answers to the questions: “If a student learns the concept Ci, which will
be her/his knowledge level of the depended domain concept Cj?”; “If the student’s
knowledge of concepts Ci improves, how will be affected her/his knowledge of the

Fig. 2.7  A hierarhical tree

34 2 Fuzzy Logic in Student Modeling

Fig. 2.8  A network of

concepts

depended concept Cj?”; “If the student has misconceptions on the domain concept
Ci, how will be affected her/his knowledge level of the depended concept Ci?”.
The domain concepts that constitute the learning material are not independent
from each other. The student’s knowledge level of a domain concept usually is
affected by her/his knowledge level of other related domain concepts. For example,
a new domain concept may be completely unknown to the learner but in other
circumstances it may be partly known due to previous related knowledge of the
learner. On the other hand, domain concepts, which were previously known by
the learner, may be completely or partly forgotten. Hence, currently they may be
partly known or completely unknown. Therefore, the knowledge representation
approach has to allow the system to recognize either the domain concepts that
are already partly or completely known for a learner, or the domain concepts that
s/he has forgot, taking into account the learner’s knowledge level of the related
concepts. Therefore, the representation of dependencies between the domain con-
cepts of the learning material includes imprecise and uncertain information. As a
result an effective solution for handling this uncertainty is to use fuzzy logic tech-
niques in the representation of the knowledge domain.
A fuzzy logic application, which is used to model the behavior of complex sys-
tems (Leon et al. 2011) and emphasizes the connections and dependencies between
the system’s elements, is the Fuzzy Cognitive Map (FCM). Fuzzy Cognitive
Maps (FCMs) constitute a way to represent real-world dynamic systems; in a
form that corresponds closely to the way humans perceive it (Papageorgiou 2011;
Papageorgiou and Iakovidis 2013). They are able to incorporate experts’ knowl-
edge (Papageorgiou and Salmeron 2012; Salmeron 2009; Salmeron et al. 2012)
and approach representation of knowledge by emphasizing the connections and
the structure (Lin 2007). A FCM illustrates the whole system as a combination of
2.3 Fuzzy Logic for Knowledge Representation 35

concepts and the various relations that exist between its ­concepts (Azadeh et al.
2012; Song et al. 2011; Stula et al. 2010). They are inference networks, using
cyclic directed graphs, for knowledge representation and reasoning (Fig. 2.9). In
particular, A FCM consists of nodes (N1, N2, … Nn), which represent the impor-
tant elements of the mapped system, and directed arcs, which represent the causal
relationships between two nodes (Ni, Nj). The directed arcs are labeled with fuzzy
values (fij) in the interval [−1, 1] that show the “strength of impact” of node Ni
on node Nj. If fij has a positive value, then it indicates that node Ni affects posi-
tively node Nj. In other words, the positive value on the directed arc that connects
Ni with Nj, means that the increase of the value of Ni leads to the increase of the
value of Nj, or the decrease of the value of Ni leads to the decrease of the value
of Nj. Otherwise, If fij has a negative value, then it indicates that node Ni affects
negatively node Nj. In other words, the negative value on the directed arc that con-
nects Ni with Nj, means that the increase of the value of Ni leads to the decrease of
the value of Nj, or the decrease of the value of Ni leads to the increase of the value
of Nj. Therefore, a FCM is a cognitive map whose relations between the nodes
can be used to compute the “strength of impact” of these elements. This property
of FCM makes it able to predict, to make decisions, to generate a more accurate
description of a difficult situation and to explain behaviors, actions and situations
(Codara 1998). That is the reason of their extensive use in a wide range of appli-
cations (Craiger et al. 1996; Kosko 1999; Miao and Liu 2000; Rodriguez-Repiso
et al. 2007; Stylios and Groumpos 2004). Furthermore, according to Papageorgiou
(2011), in the past decade, FCMs have gained considerable research interest
and are widely used to analyze causal systems such as system control, decision-
making, management, risk analysis, text categorization, prediction etc. However,
the contribution of FCMs to the knowledge representation of an adaptive tutoring
system has not been discussed before.
Taking into account the above, there is the need to represent the knowledge
dependency relations between the individual domain concepts of the domain
knowledge. In particular, the knowledge dependencies that exist between the
domain concepts of the learning material, as well as their “strength of impact” on
each other have to be represented. A solution to this is to use a combination of
a network of concepts with Fuzzy Cognitive Maps. In this way, a new approach
of domain knowledge representation derives. That new approach is called Fuzzy
Related-Concept Network (FR-CN).

Fig. 2.9  A fuzzy cognitive

map
36 2 Fuzzy Logic in Student Modeling

2.3.1 Knowledge Domain Representation Using a Fuzzy

Related-Concept Network

A Fuzzy Related-Concepts Network is a network of concepts, which depicts,

also, the knowledge dependencies that exist between the domain concepts of the
learning material. Therefore, it illustrates so the structure of the learning mate-
rial, as the concepts’ knowledge dependencies. Particularly, it represents the
fact that the knowledge level of a domain concept is increased when the knowl-
edge level of a related topic improves, as well as the fact that the knowledge
level of a domain concept is decreased when the knowledge level of a depended
topic is not satisfactory. The Fuzzy Related-Concepts Network (Fig. 2.10) con-
sists of: nodes, which depict the domain concepts of the learning material, and
directed arcs, which represent relations between the concepts of the learning
material.
The relations that exist between the concepts of the learning material depict
so the order in which the domain concepts have to be delivered and the struc-
ture of the learning material, as the knowledge dependencies. In particular, there
are three type of relations between the concepts: “precedes” that declares the
order in which each domain concept of the learning material has to be taught

Fig. 2.10  A fuzzy related-concepts network

2.3 Fuzzy Logic for Knowledge Representation 37

(for example, in Fig. 2.10 the domain concept C3 is delivered to the learner
before the domain concept C5); “part-of” that declares that a concept belongs to
another concept (for example, in Fig. 2.10 the domain concept C2 in includes the
domain concepts C10, C11 and C12); the dependence relation that declares that the
knowledge level of a domain concept is affected by the learner’s knowledge level
on another related concept (For example, in Fig. 2.10 the knowledge level of the
domain concepts C14, C8 and C9 is affected by the learner’s knowledge level on
the concept C15).
The dependence relations allow the tutoring system to identify how the knowl-
edge level of a concept is affected by the learner’s knowledge level on other
related concepts. A dependence relation is characterized by the symbol ‘+’ or the
symbol ‘−’ and a number (strength of impact). The symbol depicts the order in
which the two related concepts are delivered to the learner. If the symbol ‘+’ is
labeled on the arc that connects Ci with Cj with direction from Ci to Cj (Ci → Cj),
then it denotes that Ci is taught before Cj. Otherwise, if the symbol that is labeled
on the particular directed arc is the symbol ‘−’, then it denoted that that Cj is
taught before Ci. The numbers that are labeled on the directed arcs depict the
degree at which the knowledge level of a domain concept is affected regarding
the knowledge level of its related domain concepts. In other words, they depict
the “strength of impact” of a domain concept on a related concept. The particu-
lar numbers are only positive. This is happened due to the fact that the increase
of the knowledge level of a domain concept leads to the increase of the knowl-
edge level of a depended domain concept, and the decrease of the knowledge level
of a domain concept leads to the decrease of the knowledge level of a depended
domain concept. Therefore, the numbers of the directed arcs that depict the
knowledge dependencies belong to the interval (0, 1]. For example, in Fig. 2.10,
the value ‘+0.8’ that is labeled on the directed arc, which connects C10 with
C13 (C10 → C13), denotes that the concept C10 is delivered to the learner before
the concept C13 and the “strength of impact” of C10 on C13 is 0.8. Similarly, the
value ‘−0.72’ that is labeled on the directed arc, which connects C15 with C14
(C15 → C14), denotes that the concept C15 is delivered to the learner after the
concept C14 and the “strength of impact” of C15 to C14 is 0.72.
The arcs in the FR-CN, which represent the domain concepts’ dependencies
of the knowledge domain, are bidirectional. Furthermore, the value of the arc
Ci → Cj is not essentially equal to the value of the arc Cj → Ci. This is hap-
pened due to the fact that changes on the knowledge level of Ci may affect the
knowledge level of Cj in a different degree than changes on the knowledge level
of Cj affect the knowledge level of Ci. It has to be clear that the value 1 on the
directed arc that connects two dependent domain concepts does not mean that
the two dependent concepts are the same. It implies that if a learner knows a
domain concept of a section, s/he may know a related concept of another sec-
tion at the same degree. The percentage of increase or decrease of the knowledge
level of a domain concept that occurs due to changes on the knowledge level of
another concept related with this domain concept is defined by experts of the
knowledge domain.
38 2 Fuzzy Logic in Student Modeling

Therefore, a FR-CN that is used to represent the knowledge domain of the

learning material is a 6-tuple (C, ORD, PART, IMPACT, KL, f), where:
• C = {C1, C2, … Cn} is the set of concepts of the knowledge domain.
• ORD: (Ci, Cj) → {0, 1} is a matrix, which denotes that the concept Ci is
delivered to the learner before the concept Cj (the value of the corresponding
matrix’s cell—line i, column j—is 1). If the value of the corresponding matrix’s
cell is 0, then it denotes that there is no “precedes” relation between the two
domain concepts.
• PART: (Ci, Cj) → {0, 1} is a matrix, which denotes that the concept Ci is part-
of the concept Cj (the value of the corresponding matrix’s cell—line i, column
j—is 1). If the value of the corresponding matrix’s cell is 0, then it denotes that
there is no “part-of” relation between the two domain concepts.
• IMPACT: (Ci, Cj) → wij is a matrix, where wij is a weight of the directed arc
from Ci to Cj, which denotes the “strength of impact” of the concept Ci on the
concept Cj (the value wij is inserted in the cell that corresponds to line i and
column j). If wij = 0, then it denotes that Ci and Cj are not knowledge related
concepts.
• KL is a function that at each concept Ci associates the sequence of its activa-
tion degree. In other worlds, KLi(t) indicates the value of a concept’s knowledge
level at the moment t.
• f is a transformation function. For the definition of the transformation
function the following limitation has to be taken into account. Only the
­
knowledge level of the most recently read concept affects the knowl-
edge level of a domain concept, each time. The reason for this is the fact
that the ­learner’s knowledge level is affected either by the new knowledge
that s/he has obtained, or by the knowledge that s/he has forgot, each time.
Consequently, the KL value of a concept is affected only by the KL value of
the most recently read concept, regarding the weight of the directed arc that
connects them. Therefore, the transformation function for a FR-CN, which is
used to represent the knowledge domain of the learning material, is defined
as: KLi(t + 1) = f(KLi(t) ± wji*pj*KLi(t)/100), where pj is the percentage of
the difference on the value of the knowledge level of the most recently read
concept Cj, with pj = (KLj(t + 1) − KLj(t))*100/KLj(t). Also, the + is used
in case of increase and the − is used in case of decrease.
For example, the matrixes ORD (Table 2.1), PART (Table 2.2) and IMPACT
(Table 2.3) for the FR-CN that depicts in Fig. 2.10 are the following:
At the ORD matrix the value of the cell ORD [i, j], which corresponds to the
line i and column j, can be 1, although there is no a direct arc in the c­ orresponding
FR-CN that connects the node-concept Ci with the node-concept Cj and
declares “precedes” relation between the particular concepts. The reason for that
is the fact that an indirect relation of type “precedes” can be exist between the par-
ticular concepts. For example, in the FR-CN of Fig. 2.10, the concept C3 precedes
the concept C2 due to the fact that the concept C7 precedes the concept C2 and the
2.3 Fuzzy Logic for Knowledge Representation 39

Table 2.1  ORD
C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13 C14 C15
C1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1
C2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C3 0 1 0 0 1 1 1 1 1 1 1 1 1 1 1
C4 0 1 0 0 1 1 1 1 1 1 1 1 1 1 1
C5 0 1 0 0 0 0 1 1 1 1 1 1 1 1 1
C6 0 1 0 0 0 0 1 1 1 1 1 1 1 1 1
C7 0 1 0 0 0 0 0 0 0 1 1 1 1 1 1
C8 0 1 0 0 0 0 0 0 0 1 1 1 1 1 1
C9 0 1 0 0 0 0 0 0 0 1 1 1 1 1 1
C10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C11 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C12 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C13 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C14 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C15 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Table 2.2  PART
C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13 C14 C15
C1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C4 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
C5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C6 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0
C7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C8 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
C9 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
C10 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
C11 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
C12 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
C13 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0
C14 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0
C15 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0

concept C3 precedes the concept C7. Therefore, ORD [3, 2] = 1. Similarly, C4

precedes C8 because C4 is part-of the concept C3, which precedes the concept C7
whose part is the concept C8. C3 precedes C7 due to the fact that C3 precedes C5,
which precedes C7. As a result, ORD [4, 8] = 1.
40 2 Fuzzy Logic in Student Modeling

Table 2.3  IMPACT
C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13 C14 C15
C1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C6 0 0 0 0 0 0 0 0 0 0 0 +0.35 0 0 0
C7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C8 0 0 0 0 0 0 0 0 0 0 0 0 0 0 +1
C9 0 0 0 0 0 0 0 0 0 0 0 0 0 0 +0.63
C10 0 0 0 0 0 0 0 0 0 0 0 0 +0.8 0 0
C11 0 0 0 0 0 0 0 0 0 0 0 +1 0 0 0
C12 0 0 0 0 0 −0.5 0 0 0 0 −1 0 0 0 0
C13 0 0 0 0 0 0 0 0 0 −0.2 0 0 0 +0.45 0
C14 0 0 0 0 0 0 0 0 0 0 0 0 −0.3 0 +1
C15 0 0 0 0 0 0 0 −1 −0.17 0 0 0 0 −0.72 0

2.3.1.1 Application of FR-CN for the Representation of the Knowledge

Domain of the Programming Language ‘C’

An application of Fuzzy Related-concepts Networks to a real situation is needed to

understand the above, described approach for knowledge domain representation.
That is the aim of the particular section, in which the description of the knowledge
domain of a programming tutoring system is presented. In particular, the knowl-
edge domain of the programming tutoring system is the programming language
‘C’. The aim of the particular tutoring system is to teach learners so the principles
and structures of the programming language ‘C’, as the logic of programming. So,
the learning material includes not only expressions, operations and statements of the
programming language ‘C’, but also it includes algorithms, like calculating sums,
averages and maximums or minimums. Thereby, the learning material is decom-
posed in domain concepts which concern declarations of variables and constants,
expressions and operators, input and output expressions, the sequential execution
of a program, the if, if-else and if-else if statements, the iteration statements (for
loop, while loop, do…while loop), sorting and searching algorithms, arrays, func-
tions (Table 2.4).
Learners of programming languages have different backgrounds and their
knowledge of a concept of the programming language, which they are taught, is
subject to change. A new concept may be completely unknown to the learner but in
other circumstances it may be partly or completely known due to previous related
knowledge of the learner. For example, if a learner already knows an algorithm
(e.g., calculating the sum of integers in a ‘for’ loop), there is no need to learn another
similar algorithm (e.g., counting in a ‘for’ loop). Similarly, if a learner knows a pro-
gramming structure (e.g., one-dimensional arrays), it is easier to understand another
2.3 Fuzzy Logic for Knowledge Representation 41

Table 2.4  Learning material of the programming language ‘C’

C1. Basics C1.1. Constants and C5. Iteration C5.1. While statement
variables structure
C1.2. Assignment statement Unknown no of C5.2. Calculating sum
loops in a while loop
C1.3. Arithmetical operators C5.3. Counting in a
while loop
C1.4. Comparative operators C5.4. Calculating avgr
in a while loop
C1.5. Logical operators C5.5. Calculating max/
min in a while loop
C1.6. Mathematical C5.6. Do…while
functions statement
C1.7. Input-output statements
C2. Sequence C2.1. A simple program C6. Arrays C6.1 One-dimensional
structure structure arrays
C6.2. Searching
C3. Conditional C3.1. If statement C6.3. Sorting
structures C3.2. If…else if C6.4. Two-dimensional
C3.2.1 Methodology of find- arrays
ing max/min
C3.3. Nested if C6.5. Processing per
row
C6.6. Processing per
column
C4. Iteration C4.1. For statement C6.7. Processing of
structure diagonals
Concrete no of C4.2. Calculating sum in a C7. C7.1. Functions
loops for loop Sub-programming
C4.3. Counting in a for loop
C4.4. Calculating avgr in a
for loop
C4.5. Calculating max/min
in a for loop

programming structure (e.g., multidimensional arrays), so this new structure should

not be considered as being completely unknown to the learner. On the other hand,
domain concepts, which were previously known by the learner, may be completely
or partly forgotten. For example, if a learner has difficulties in calculating a sum in
a ‘while’ loop, her/his knowledge of the previous domain concept of “calculating a
sum in a ‘for’ loop” has eroded. Therefore, there is the need to represent the knowl-
edge dependencies that exist between the domain concepts of the learning mate-
rial of the programming language. This is achieved using Fuzzy Related-Concepts
Network. The FR-CN for the knowledge domain of the programming language ‘C’
that is described in Table 1.7 is depicted in Fig. 2.11. Tables 2.5, 2.6 and 2.7 are a part
of the matrixes ORD, PART, IMPACT of the FR-CN of Fig. 2.11 correspondingly.
The whole matrixes are presented in the Appendix A.
42 2 Fuzzy Logic in Student Modeling

Fig. 2.11  The FR-CN of the knowledge domain of the programming language ‘C’ (it is
decomposed in four graphs). a The “precedence” and “part-of” relations of the FR-CN; b The
knowledge dependence relations for the domain concepts of the section 3; c The knowledge
dependence relations for the domain concepts of the section 6; d The knowledge dependence
relations for the domain concepts of the sections 4 and 5
2.3 Fuzzy Logic for Knowledge Representation 43

Fig. 2.11  (continued)

The value 1 on the directed arc that connects two dependent domain concepts
of the FR-CN implies that if a learner knows a domain concept, then s/he may
know a related domain concept at the same degree. For example, if a learner has
been tested and found to have known the “for” loop and the “while” loop and this
44 2 Fuzzy Logic in Student Modeling

Table 2.5  A sample of the ORDER matrix of the FR-CN of Fig. 2.11
C1 C1.1 C1.2 C1.3 C1.4 C1.5 C1.6 C1.7 C2 C2.1 C3 C3.1 C3.2 C3.2.1 C3.3
C1.7 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1
C2 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1
C2.1 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1
C3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C3.1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0

Table 2.6  A sample of the PART matrix of the FR-CN of Fig. 2.11
C1 C1.1 C1.2 C1.3 C1.4 C1.5 C1.6 C1.7 C2 C2.1 C3 C3.1 C3.2 C3.2.1 C3.3
C2.1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
C3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
C3.1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
C3.2 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
C3.2.1 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0

Table 2.7  A sample of the IMPACT matrix of the FR-CN of Fig. 2.11
C4 C4.1 C4.2 C4.3 C4.4 C4.5 C5 C5.1 C5.2 C5.3 C5.4 C5.5 C5.6
C4.1 0 0 0 0 0 0 0 0 0 0 0 0 0
C4.2 0 0 0 +0.45 +0.81 0 0 0 +1 +0.45 +0.39 0 0
C4.3 0 0 −0.42 0 +0.34 0 0 0 +0.42 +1 +0.41 0 0
C4.4 0 0 −1 −0.45 0 0 0 0 +1 +0.45 +0.52 0 0
C4.5 0 0 0 0 0 0 0 0 0 0 0 +1 0

learner knows how to calculate sum in a “for” loop, s/he will also know how to
calculate sum in a “while” loop, since the methodology is the same.
Experts on programming have defined so the domain concepts of the learning
material, as their relations (“precedence”, “part-of”, “knowledge dependence”).
In particular, ten professors of computer programming, whose experience counts
12 years at least, are responsible for the definition and structure of the knowl-
edge domain. They were, also, asked to determine, empirically, the knowledge
dependencies that exist between the defined domain concepts of the learning
material, as well as their “strength of impact” on each other. The FR-CN that
is depicted in Fig. 2.11 has been mapped according to the mean of the experts’
answers (due to its complexity, it has been decomposed in four graphs).
The information that is derived from the above matrixes concerns:
• The order in which the domain concepts of the leaning material have to be
delivered.
• Which domain concepts belong to another general domain concept of the learning
material.
• The knowledge dependencies that exist between the domain concepts of the
learning material and their “strength of impact”.
2.3 Fuzzy Logic for Knowledge Representation 45

For example, the domain concept C1 is delivered before concept C2 and ­concept
C4.2 is delivered before the domain concept C4.4. That is derived from the values of
the cells ORDER [1, 9] (Table 1.8a) and ORDER [18, 20] (Table 1.8b), which are 1
both. On the other hand, the ORDER [18, 21] = 0 (Table 1.8b) denotes that the con-
cept C4.2 is not necessary to be taught before the concept C4.5. Furthermore, C3.2.1
belongs to the concepts C3 and C3.2 as PART [14, 11] = 1 and PART [14, 13] = 1
(Table 2.1a). In addition, the learner’s knowledge level on the concept C4.4 affects
the particular learner’s knowledge level on the previously delivered concepts C4.2,
C4.3, C5.2, C5.3 and C5.5. This information is derived from the matrix IMPACT.
In particular, the values IMPACT [20, 18] = −1 and IMPACT [20, 19] = −0.45
(Table 2.2b) denote that the knowledge level of concept C4.4 affects the knowledge
level of C4.2 and C4.3, and its “strength of impact” on C4.2 and C4.3 are 1 and 0.45
correspondingly. Similarly, the values IMPACT [20, 24] = + 1, IMPACT [20,
25] = + 0.45 and IMPACT [20, 26] = +0.52 (Table 2.2b) denote that the knowledge
level of concept C4.4 affects the knowledge level of the following concepts C5.2, C5.3
and C5.5, and its “strength of impact” on the particular concepts are 1, 0.45 and 0.52
correspondingly. However, the value IMPACT [20, 21] = 0 (Table 2.2b) denote that the
knowledge level of concept C4.4 does not affect the knowledge level of the concept C4.5.

2.4 A Novel Rule-Based Fuzzy Logic System

for Modeling Automatically the Learning
or Forgetting Process of a Student

Learning is not a “black or white” process. The definition of the learner’s knowledge
level is a moving target. In other words, it is not a straightforward task to define for
each learner which concepts are unknown, known or assimilated and at what degree.
The particular process is confronted with uncertainty and human subjectivity. One
possible approach to deal with this is fuzzy set techniques, with their ability to nat-
urally represent human conceptualization. That is the reason for the integration of
fuzzy logic techniques into the student model.
Fuzzy logic is the solution for recognizing and modeling the increase and/or
decrease of the learner’s knowledge level on a domain concept in relation with
her/his performance on other related domain concepts of the learning material. In
particular, the presented rule-based fuzzy logic module is responsible for identifying and
updating the student’s knowledge level of all the concepts of the knowledge domain.
Its operation is based on the Fuzzy Related-Concepts Network that is used to
represent the structure of the learning material and the dependencies that exist
between the domain concepts. It uses fuzzy sets to represent the student’s knowl-
edge level and a mechanism of rules over the fuzzy sets, which is triggered after a
change has occurred on the student’s knowledge level of a domain concept. This
mechanism updates the student’s knowledge level of all related with this concept,
concepts. With this approach the alterations on the state of student’s knowledge
level, such as forgetting or learning are represented.
46 2 Fuzzy Logic in Student Modeling

The presented rule-based fuzzy logic module includes the following three steps:
Step 1 Definition of the fuzzy sets:
In the particular step, the definition of the fuzzy sets, which represent the
learner’s knowledge level on a domain concept (i.e. {“Unknown”, “Known”,
“Learned”} or {“Unknown”, “Insufficiently Known”, “Known”, “Learned”,
“Assimilated”}), is carried out. Fuzzy sets are used to characterize the change-
able learner’s knowledge level. Therefore, FS1, FS2, …, FSn are the defined
fuzzy sets, for the educational adaptive system.
Step 2 Definition of the membership functions:
In the particular step, the membership functions of the determined fuzzy sets
FS1, FS2, …, FSn is defined. The membership functions (Fig. 2.12) are defined
as follows (x indicates the learner’s degree of success on a particular domain
concept; xi-1, xi, xi+1, xi+2 are thresholds that indicate particular degrees of
success like 0, 50, 100):

 1, x ≤ x1
µFS1 = 1 − xx−x 1
2 −x1
, x1 < x < x2
0, x ≥ x2


x−x2i−3


 x2i−2 −x2i−3 , x2i−3 < x < x2i−2

1, x2i−2 ≤ x ≤ x2i−1

∀i �= 1andi �= nµFSi = x−x2i−1
1− x2i −x2i−1 , x2i−1 < x < x2i



 0, x ≤ x2i−3 or x ≥ x2i

 x−x2n−3
 x2n−2 −x2n−3 x2n−3 < x < x2n−2
µFSn = 1 x2n−2 ≤ x ≤ x2n−1
0 x ≤ x2n−3


The knowledge level of a domain concept changes in a continuous way. Meaning

that the knowledge level of a domain concept usually passes gradually from the

Fig. 2.12  The membership functions μFSi

2.4 A Novel Rule-Based Fuzzy Logic System … 47

unknown state to the learned and assimilated state. Membership values corre-
spond to percentages of the offered knowledge in a way that they cover 100 % of
it, at any time. This gives a more natural and understandable way of representa-
tion. For example, it would be non-intuitive to say that domain concept “A” is 0.5
(50 %) Insufficiently Known and 0.6 (60 %) Known for a student, given that 0.5
plus 0.6 gives 1.1 (110 %). So, the sum of the concept’s percentage of different
knowledge levels has to be 100 %, or 1 if the membership value of a concept to a
knowledge level category is from 0 to 1. So, the following expression stands:
µFS1 + µFS2 + µFS3 + . . . + µFSn = 1
Therefore, a set (μFS1, μFS2, μFS3, … ,μFSn) is used express the student
knowledge of a domain concept.
Step 3 Definition of the fuzzy rules:
When there is a dependency between two domain concepts, then the knowl-
edge level of the one domain concept can affect the knowledge level of the other
domain concept. More specifically, the following are taken into account:
• Considering the knowledge level of Ci, the knowledge level of its following
domain concept Cj is increased or decreased.
• Considering the knowledge level of Cj, the knowledge level of its prerequisite
domain concept Ci is increased or decreased.
Consequently, the student model expands when a change on the knowledge level of
a domain concept causes increase on the knowledge level of the related concepts,
or it is minimized when a change on the knowledge level of a domain concept
causes decrease on the knowledge level of the related concepts with this concept.
In this document, D is defined to represent the knowledge dependency between two
domain concepts. The symbolism μD(Ci, Cj) is used to represent the “strength of
impact” of Cj on Ci and the symbolism μD(Cj, Ci) is used to represent the “strength
of impact” of Ci on Cj. The values of μD(Ci, Cj) and μD(Ci, Cj) are the values of the
arcs that depict the “knowledge dependencies” relations between the concepts of
the learning material in the FR-CN of the knowledge domain (Sect. 3.1).
Concerning two domain concepts Ci and Cj where Ci is taught before Cj, the
knowledge level of the concepts can change according to the following rules.
These rules depict how the changes on the knowledge level of the domain con-
cepts of the learning material for a student occur, revealing her/his learning state.
In particular, they reveal if s/he learns or not or if s/he forgets. If the knowledge
level of a concept is decreased, then the system infers that the student does not
learn. If the knowledge level of a previously taught concept is decreased, then
the system infers that the student forgets. If the knowledge level of a concept is
increased, then the system infers that the student learns, and if the knowledge
level of all the related concepts is improved continuously, then the system infers
that the student assimilates the learning material.
The rules are based on Kavčič’s (2004b) work. That work models mainly how
the student’s knowledge level of the prerequisites concepts that the student
48 2 Fuzzy Logic in Student Modeling

had read previously, is improved when s/he performs better in following

­concepts. In this way Kavčič’s work deals only with how learning progresses.
In her work there are no rules that imply the possible decrease of knowledge
via the student’s forgetting of some previously learned concepts. Moreover,
another important problem that is not dealt with in Kavčič’s work is the
fact that in static educational systems, students are often required to repeat
previously known concepts thought the following chapters. However, this
practice is quite generic and does not take into account individual features
of a student such as how fast they learn or how well they remember previ-
ously taught concepts. As such, educational systems do not adapt their pace
on individual students. In view of the above, in the presented rule-based
fuzzy module, Kavčič’s rules have been expanded to deal with the above
problems. The rules with these novelties that lead to the dynamic personali-
zation of teaching are presented below. In the following rules, FSx, FSy are
fuzzy sets that represent knowledge levels with FSx < FSy, and KL() denotes
the “Knowledge Level of”.

• Based on updates of the KL(Ci), the KL(Cj) is improved according to:

R1: If the same fuzzy sets are active for both Ci and Cj, then KL(Cj) = FSx with
µFSy (Cj ) = max[µFSx (Ci ), µFSx (Ci ) ∗ µD (Ci , Cj )
where FSx is the last active fuzzy set. Subtract the value (new μFSx
(Cj)—­previous
μFSx (Cj)) from the others μFSy(Cj) (FSy < FSx) sequentially until µFSi = 1.
R2: If KL(Cj) = FSx and KL(Ci) = FSy, then KL(Cj) = FSy with
µFSy (Cj ) = µFSy (Ci ) ∗ µD (Ci , Cj )
• Based on updates of the KL(Ci), the KL(Cj) is deteriorated according to:
R3: If KL(Cj)
  = FSn, then
if µFS1 Cj + µFS2 Cj + · · · + µFSn−1 Cj < µFSi (Ci ) ∗ µD Ci , Cj , where
    

i < n, then the corresponding value is subtracted by μFSn(Cj)

else it does not change.
R4: If KL(Cj) = FSy and KL(Ci) = FSx, then KL(Cj) = FSx with
   
µFSx Cj = µFSx (Ci ) ∗ µD Ci , Cj
• Based on updates of the KL(Cj), the KL(Ci) is improved according to:
R5: If the same fuzzy sets are active for both Ci and Cj, then KL(Cj) = FSx with
   
µFSx (Ci ) = max[µFSx (Ci ), µFSx Cj ∗ µD Ci , Cj ]
where FSx is the last active fuzzy set. Subtract the value (new μFSx
(Ci)—­previous
μFSx (Ci)) from the others μFSy(Ci) (FSy < FSx) sequentially until µFSi = 1
R6: If KL(Ci) = FSx and KL(Cj) = FSy, then KL(Ci) = FSy with

µFSy (Ci ) = µFSy (Cj ) ∗ µD (Cj , Ci )

2.4 A Novel Rule-Based Fuzzy Logic System … 49

• Based on updates of the KL(Cj), the KL(Ci) is deteriorated according to:

R7: If KL(Ci) = FSn with μFSn(Ci) = 1, then it does not change

R8: The formula xi = 1 − µ D Ci ,C j ∗ xi + min[µ D Ci ,C j ∗ xi ,µ D

    

Ci ,C j ∗ x j ], where xi and xj are the values of the criterion, which determines



the fuzzy sets that are active each time for Ci and Cj respectively, is used (for
the calculation of previous xi, the membership value of the upper active fuzzy
set is used). Then, using the new xi, the KL(Ci) is determined, calculating the
membership  functions.
• Limitation: µFSi = 1

2.4.1 Integration of the Fuzzy Rules

The application of the fuzzy rules of the step 3 that was described above deals
with the problem of estimating wrongly the knowledge level of a domain con-
cept. In particular, consider the fuzzy sets {“Uknown”, “Known”, “Well-Known”,
“Learned”} and the set of their membership functions (μUn, μK, μWK, μL) that
represent the student’s knowledge level of a domain concept. Let’s the domain
concept Ci to be 100 % ‘Learned’ and the “strength of impact” of Ci on the fol-
lowing concept Cj to be 0.3. The knowledge level of Cj is 100 % ‘Unknown’.
According to the rule R2, the knowledge level of Cj will become 30 % ‘Learned’.
However, that it means that the rest 70 % of the concept Cj is ‘Known’? The
answer is no. The rest 70 % of the Cj can be ‘Unknown’, ‘Known’, ‘Well-Known’
or ‘Learned’, or different parts of it can belong to a different fuzzy set (i.e. 10 %
‘Unknown’, 20 % ‘Known’ and 40 % ‘Well-Known’). In addition, let’s the set that
describes the knowledge level of the domain concept Ci to be (0.8, 0.2, 0, 0) (e.g.
80 % ‘Unknown’ and 20 % ‘Known’ → KL(Cj) = 0.2 ‘Known’) and the “strength
of impact” of Ci on its following concept Cj to be 0.6. The knowledge level of Cj is
20 % ‘Learned’. According to the rule R4, the knowledge level of Cj will become
60 % ‘Known’. However, that it means that the rest 40 % of the concept Cj is
‘Uknown’? The answer is no. It can be any of the above fuzzy sets.
A solution to this problem is to keep data for each domain concept of the
learning material concerning the different part of the particular concept that can
be affected be other related concepts. In such a way, the system can be informed
each time about the knowledge level of each separate part of the particular domain
concept and it is able to draw conclusions about the learner’s knowledge level on
the overall domain concept. For example, according to the Fig. 2.10 (Sect. 3.1)
the domain concept C12 is affected by both concepts C11 and C6. Initially
is KL(C6) = KL(C11) = KL(C12) = 100 % ‘Uknown’. During the learning
process, the concept C6 is delivered to the learner firstly. The learner’s knowledge
level on the particular concept becomes 20 % ‘Well-Known’ and 80 % ‘Known’
(KL(C6) = 20 % Well-Known). According to the rule R2, the learner’s knowl-
edge level on the domain concept C12 will become 7 % ‘Well-Known’ and 28 %
‘Known’. The other part, however, of C12 is not affected by C6. So, its knowledge
50 2 Fuzzy Logic in Student Modeling

level remains ‘Unknown’. Therefore, C12 is 7 % ‘Well-Known’, 28 % ‘Known’

and 65 % ‘Unknown’. As a result, the system will advise the learner to read
C12. Also, according to R6, the learner’s knowledge level of C11 will become
7 % ‘Well-Known’ 28 % ‘Known’ and 65 % ‘Unknown’ because C12 affects C11
with “strength of impact” 1 (Fig. 2.10). Then, the concept C11 is delivered to the
learner. The learner’s knowledge level on the particular concept becomes 40 %
‘Learned’ and 60 % ‘Well-Known’ (KL(C11) = 40 % Learned). According to R2
is KL(C12) = 40 % Learned (40 % ‘Learned’ and 60 % ‘Well-Known’), due to the
fact that the “strength of impact” of C11 on C12 is 1. Therefore, the system will
consider that the concept the learner knows C12, and it will not advise her/him to
read the particular concept. In addition, C12 affects C6. The “strength of impact”
of the particular knowledge dependency is 0.5. Therefore, according to the rule
R6, the learner’s knowledge level of C6 will become 20 % ‘Learned’ and 30 %
‘Well-known’. However, because the previous knowledge level of C6 was 20 %
‘Well-Known’ and 80 % ‘Known’, the system will consider that the rest 50 % of
C6 remains ‘Known’. Thereby, although the learner’s knowledge level on C6 has
been improved, the system will advise the learner to revise the domain concept C6.

2.4.2 Application of the Presented Rule-Based Fuzzy Logic

System in a Programming Tutoring System

In this chapter an application of the presented rule-based fuzzy logic system is

described. In particular, the presented rule-based fuzzy logic system is used to
model the cognitive states of learners of the programming language ‘C’.
Step 1 Definition of the fuzzy sets:
The defined fuzzy sets are the following:
• Unknown (Un): the degree of success in the domain concept is from 0 to 50 %.
• Moderate Known (MKn): the degree of success in the domain concept is from
40 to 70 %.
• Known (Kn): the degree of success in the domain concept is from 60 to 80 %.
• Learned (L): the degree of success in the domain concept is from 75 to 90 %.
• Assimilated (A): the degree of success in the domain concept is from 85 to
100 %.
Step 2 Definition of the membership functions:
The membership functions of the fuzzy sets Un, MKn, Kn, L and A are depicted
in Fig. 2.13 and are the following (x indicates the learner’s degree of success on
a particular domain concept):

 1, x ≤ 40
µUn = 1 − x−40 10 , 40 < x < 50
0, x ≥ 50

2.4 A Novel Rule-Based Fuzzy Logic System … 51

Fig. 2.13  The membership functions of the fuzzy sets of the programming tutoring system for ‘C’

x−40


 10 , 40 < x < 50
1, 50 ≤ x ≤ 60

µMKn = x−60

 1− 10 , 60 < x < 70
0, x ≤ 40 or x ≥ 70


x−60


 10 , 60 < x < 70
1, 70 ≤ x ≤ 75

µKn = x−75

 1 − 5 , 75 < x < 80
0, x ≤ 60 or x ≥ 80


x−75


 5 , 75 < x < 80
1, 80 ≤ x ≤ 85

µL = x−85

 1 − 5 , 85 < x < 90
0, x ≤ 75 or x ≥ 90


x−85

 5 , 85 < x < 90
µA = 1, 90 ≤ x ≤ 100
0, x ≤ 85


Therefore, a set (μUn, μMKn, μKn, μL, μA) is used to express the student knowl-
edge of a domain concept.
Experts on programming and teachers of the programming language ‘C’
have defined the limits of each fuzzy set. In particular, they were asked to deter-
mine the lower and higher values of the degree of success that characterize a
domain concept as ‘Unknown’, ‘Moderate Known’, ‘Known’, ‘learned’ and
‘Assimilated’. The mean values of their answers consist the base for the definition
of the limits of the presented fuzzy sets.
52 2 Fuzzy Logic in Student Modeling

Step 3 Definition of the fuzzy rules:

Concerning two domain concepts Ci and Cj where Ci is taught before Cj, the
knowledge level of the concepts can change according to the following rules
(μD(Ci, Cj) and μD(Cj, Ci) indicate the “strength of impact” of Ci on Cj and of
Cj on Ci correspondingly. Their values are the values of the arcs that depict the
“knowledge dependencies” relations between the concepts of the learning material
in the FR-CN (Sect. 3.1.1 Fig. 2.11)):
• Based on updates of the KL(Ci), the KL(Cj) is improved according to:
Subtract the value (new μx(Cj)—previous μx(Cj)) from the others μy(Cj)
sequentially until µUn + µMKn + µKn + µL + µA = 1, where x = {MKn, Kn,
L, A} and y = {Un, MKn, Kn, L} with y < x.
R1: If the same fuzzy setsare active
 for both
 Ci and Cj, then:
– If KLA(Cj) > 0: µA Cj  = max µA Cj , µA (Ci ) ∗ µD Ci , Cj 
 

– Else If KLL(Cj) > 0: µL Cj =  max µL Cj , µL (Ci ) ∗ µD Ci , C j

– Else If KLKn(Cj) > 0: µKn Cj = max µKn Cj , µKn

(C i ) ∗ µD C i , Cj
– Else If KLMKn(Cj) > 0: µMKn Cj = max µMKn Cj , µMKn (Ci ) ∗ µD Ci , Cj
  

R2:
(a) If KL(Cj) = Un and KL(Ci) = MKn, then KL(Cj) = MKn with
   
µMKn Cj = µMKn (Ci ) ∗ µD Ci , Cj
(b) If KL(Cj) = Un and KL(Ci) = Kn, then KL(Cj) = Kn with
   
µKn Cj = µKn (Ci ) ∗ µD Ci , Cj
(c) If KL(Cj) = Un and KL(Ci) = L, then KL(Cj) = L with
   
µL Cj = µL (Ci ) ∗ µD Ci , Cj
(d) If KL(Cj) = Un and KL(Ci) = A, then KL(Cj) = A with
   
µA Cj = µA (Ci ) ∗ µD Ci , Cj
(e) If KL(Cj) = MKn and KL(Ci) = Kn, then KL(Cj) = Kn with
   
µKn Cj = µKn (Ci ) ∗ µD Ci , Cj
(f) If KL(Cj) = MKn and KL(Ci) = L, then KL(Cj) = L with
   
µL Cj = µL (Ci ) ∗ µD Ci , Cj
(g) If KL(Cj) = MKn and KL(Ci) = A, then KL(Cj) = A with
   
µA Cj = µA (Ci ) ∗ µD Ci , Cj
(h) If KL(Cj) = Kn and KL(Ci) = L, then KL(Cj) = L with
   
µL Cj = µL (Ci ) ∗ µD Ci , Cj
2.4 A Novel Rule-Based Fuzzy Logic System … 53

(i) If KL(Cj) = Kn and KL(Ci) = A, then KL(Cj) = A with

   
µA Cj = µA (Ci ) ∗ µD Ci , Cj
(j) If KL(Cj) = L and KL(Ci) = A, then KL(Cj) = A with
   
µA Cj = µA (Ci ) ∗ µD Ci , Cj

• Based on updates of the KL(Ci), the KL(Cj) is deteriorated according to:

R3: If KL(Cj)
  = A, then
– if µUn Cj + µMKn Cj + µKn Cj + µL Cj < µx (Ci ) ∗ µD Ci , Cj , where
       

x = {Un, MKn, Kn, L}, then the corresponding value is subtracted by μA(Cj)
– else it does not change.
R4:
(a) If KL(Cj) = L and KL(Ci) = Kn, then KL(Cj) = Kn with
   
µKn Cj = µKn (Ci ) ∗ µD Ci , Cj
(b) If KL(Cj) = L and KL(Ci) = MKn, then KL(Cj) = MKn with
   
µMKn Cj = µMKn (Ci ) ∗ µD Ci , Cj
(c) If KL(Cj) = L and KL(Ci) = Un, then KL(Cj) = Un with
   
µUn Cj = µUn (Ci ) ∗ µD Ci , Cj
(d) If KL(Cj) = Kn and KL(Ci) = MKn, then KL(Cj) = MKn with
   
µMKn Cj = µMKn (Ci ) ∗ µD Ci , Cj
(e) If KL(Cj) = Kn and KL(Ci) = Un, then KL(Cj) = Un with
   
µUn Cj = µUn (Ci ) ∗ µD Ci , Cj
(f) If KL(Cj) = MKn and KL(Ci) = Un, then KL(Cj) = Un with
   
µUn Cj = µUn (Ci ) ∗ µD Ci , Cj

• Based on updates of the KL(Cj), the KL(Ci) is improved according to:

R5: If the same fuzzy sets are active
 for both Ciand Cj, then:
– If KLA(Ci) > 0: µA (Ci ) = max µA(Ci ), µA Cj ∗µD Cj , Ci 
 

– Else If KLL(Ci) > 0: µL (Ci ) = max µL (Ci ), µL Cj ∗ µD Cj , C

i
– Else If KLKn(Ci) > 0: µKn (Ci ) = max µKn (Ci ), µKn Cj ∗µD Cj , Ci


– Else If KLMKn(Ci) > 0: µMKn (Ci ) = max[µMKn (Ci ), µMKn Cj ∗ µD Cj , Ci ]



Subtract the value (new μx(Ci)—previous μx(Ci)) from the others μy(Ci)
sequentially until µUn + µMKn + µKn + µL + µA = 1, where x = {MKn, Kn,
L, A} and y = {Un, MKn, Kn, L} with y < x.
54 2 Fuzzy Logic in Student Modeling

R6:
(a) If KL(Ci) = Un and KL(Cj) = MKn, then KL(Ci) = MKn with
   
µMKn (Ci ) = µMKn Cj ∗ µD Cj , Ci
(b) If KL(Ci) = Un and KL(Cj) = Kn, then KL(Ci) = Kn with
   
µKn (Ci ) = µKn Cj ∗ µD Cj , Ci
(c) If KL(Ci) = Un and KL(Cj) = L, then KL(Ci) = L with
   
µL (Ci ) = µL Cj ∗ µD Cj , Ci
(d) If KL(Ci) = Un and KL(Cj) = A, then KL(Ci) = A with
   
µA (Ci ) = µA Cj ∗ µD Cj , Ci
(e) If KL(Ci) = MKn and KL(Cj) = Kn, then KL(Ci) = Kn with
   
µKn (Ci ) = µKn Cj ∗ µD Cj , Ci
(f) If KL(Ci) = MKn and KL(Cj) = L, then KL(Ci) = L with
   
µL (Ci ) = µL Cj ∗ µD Cj , Ci
(g) If KL(Ci) = MKn and KL(Cj) = A, then KL(Ci) = A with
   
µA (Ci ) = µA Cj ∗ µD Cj , Ci
(h) If KL(Ci) = Kn and KL(Cj) = L, then KL(Ci) = L with
   
µL (Ci ) = µL Cj ∗ µD Cj , Ci
(i) If KL(Ci) = Kn and KL(Cj) = A, then KL(Ci) = A with
   
µA (Ci ) = µA Cj ∗ µD Cj , Ci
(j) If KL(Ci) = L and KL(Cj) = A, then KL(Ci) = A with
   
µA (Ci ) = µA Cj ∗ µD Cj , Ci

• Based on updates of the KL(Cj), the KL(Ci) is deteriorated according to:

R7: If KL(Ci) = A with μA(Ci) = 1, then it does not change.

R8: The formula xi = 1 − µ D Ci ,C j ∗ xi + min[µ D Ci ,C j ∗ xi ,µ D

    

Ci ,C j ∗ x j ], where xi and xj are the degree of success, which determine the



fuzzy sets that are active each time for Ci and Cj respectively, is used (for the
calculation of previous xi, the membership value of the upper active fuzzy
set is used). Then, using the new xi, the KL(Ci) is determined, calculating the
membership functions.
• Limitation L1: µUn + µMKn + µKn + µL + µA = 1.
2.4 A Novel Rule-Based Fuzzy Logic System … 55

2.4.2.1 Examples of Operation

The above described rule-based fuzzy logic system was used in a postgradu-
ate program in the field of informatics at the University of Piraeus in Greece. It
was used in order to offer dynamically personalized e-training in computer pro-
gramming and the language C. At the beginning, all the domain concepts of the
learning material were considered to be ‘Unknown’ for the learners. At the next
interactions, the system delivered to them the appropriate learning material for
each individual student’s needs by adapting instantly to the learner’s individ-
ual learning pace. The KL value of each domain concept was determined by the
results of the tests. There were two kinds of tests: (i) the tests that corresponded
to each individual domain concept of the learning material (practice tests), (ii) the
final tests that corresponded to the sections of the learning material (they included
exercises of a variety of domain concepts). In particular, each time the learner read
a domain concept, s/he had to complete a corresponding practice test. When, the
learner had completed successfully all the practice tests of the domain concepts of
a section (e.g. iterations with concrete number of loops, arrays, sub-programming),
then s/he had to complete the final test of the section. If s/he succeeded to the final
test, then s/he transited to a next section. Otherwise, s/he had advised to revise
some domain concepts. Representative examples of the system’s implementation
follow.
• Example 1
George had learned the sections 1 (domain concepts 1.1 to 1.7) and 2 (domain
­concept 2.1) and she was taught the domain concepts of the section 3 (domain
concepts 3.1 to 3.3) (Interaction I of Table 2.8). He read the concept C3.1. Then,
he was examined in the particular domain concept and succeeded 78 %. According
to the above, the value of the defined membership functions for concept C3.1
become μUn = 0, μMKn = 0, μKn = 0.4, μL = 0.6 and μA = 0. According to the
FR-CN (Fig. 2.11) the concept C3.1 affects the following concepts C3.2 and C3.3
with “strength of impact” 0.5 and 0.2 correspondingly. Consequently, applying the
fuzzy rule R2 (b) and (c), KL(C3.2) becomes 20 % ‘Known’ and 30 % ‘Learned’.
The rest 50 % of the particular concept remains ‘Unknown’ (Interaction II of
Table 2.8). Similarly, applying the same rules, KL(C3.3) becomes 8 % ‘Known’
and 12 % ‘Learned’. The rest 80 % of the particular concept remains ‘Unknown’
(Interaction II of Table 2.8). Therefore, although concepts C3.2 and C3.3 are not
completely unknown to George, the system advises him to read them.
• Example 2
Kate had learned the sections 1 (domain concepts 1.1 to 1.7), 2 (domain con-
cept 2.1), 3 (domain concepts (3.1 to 3.3) and the concepts 4.1, 4.5 and 5.5
(Interaction I of Table 2.9). She read the concept C4.2 to improve her knowledge
level. Then, she was examined in the particular domain concept and succeeded
86 %. According to the above, the value of the defined membership functions
for concept C4.2 become μUn = 0, μMKn = 0, μKn = 0, μL = 0.8 and μA = 0.2.
56 2 Fuzzy Logic in Student Modeling

Table 2.8  George’s progress

Domain concepts Learner’s knowledge
Interaction I (μUn, μMKn, Interaction II (μUn,
μKn, μL, μA) μMKn, μKn, μL, μA)
1.1 Constants and variables (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
1.2 Assignment statement (0,0, 0, 0.08, 0.92) (0,0, 0, 0.08, 0.92)
1.3 Arithmetic operators (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
1.4 Comparative operators (0,0, 0,0.08, 0.92) (0,0, 0,0.08, 0.92)
1.5 Logical operators (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
1.6. Mathematic functions (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
1.7 Input-output statements (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
2.1 A simple program’s structure (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
3.1 If statement (1, 0, 0, 0, 0) (0, 0, 0.4, 0.6, 0)
3.2 If…else if (1, 0, 0, 0, 0) (0.5, 0, 0.2, 0.3, 0)
3.2.1 Finding max, min (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
3.3 Nested if statement (1, 0, 0, 0, 0) (0.8, 0, 0.08, 0.12, 0)
4.1 For statement (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
4.2 Calc. sum in a for loop (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
4.3 Counting in a for loop (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
4.4 Calc. avrg in a for loop (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
4.5 Calc. max/min in a for loop (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
5.1 While statement (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
5.2 Calc. sum in a while loop (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
5.3 Counting in a while loop (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
5.4 Calc. avrg in a while loop (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
5.5 Calc. max/min in a while loop (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
5.6 Do…until (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.1 One-dimension arrays (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.2 Searching (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.3 Sorting (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.4 Two-dimensions arrays (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.5 Processing per rows (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.6 Processing per column (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.7 Processing of diagonals (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
7.1 Functions (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)

According to the FR-CN (Fig. 2.11) the concept C4.2 affects the following concepts
C4.3, C4.4, C5.2, C5.3 and C5.4 with “strength of impact” 0.45, 0.81, 1, 0.45 and 0.39
correspondingly. Consequently, applying the fuzzy rule R2 (c) and (d), KL(C4.3)
becomes 36 % ‘Learned’ and 9 % ‘Assimilated’. The rest 55 % of the particular
concept remains ‘Unknown’ (Interaction II of Table 3.2). Similarly, applying the
same rules, KL(C4.4) becomes 64.8 % ‘Learned’ and 16.2 % ‘Assimilated’ (the rest
19 % of the particular remains ‘Unknown’), KL(C5.2) becomes 80 % ‘Learned’
2.4 A Novel Rule-Based Fuzzy Logic System … 57

Table 2.9  Kate’s progress

Domain concepts Learner’s knowledge
Interaction I (μUn, μMKn, Interaction II (μUn, μMKn,
μKn, μL, μA) μKn, μL, μA)
1.1 Constants and variables (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
1.2 Assignment statement (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
1.3 Arithmetic operators (0, 0, 0, 0.02, 0.098) (0, 0, 0, 0.02, 0.098)
1.4 Comparative operators (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
1.5 Logical operators (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
1.6. Mathematic functions (0, 0, 0, 0.12, 0.88) (0, 0, 0, 0.12, 0.88)
1.7 Input-output statements (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
2.1 A simple program’s (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
structure
3.1 If statement (0, 0, 0, 0.3, 0.7) (0, 0, 0, 0.3, 0.7)
3.2 If…else if (0, 0, 0, 0.4, 0.6) (0, 0, 0, 0.4, 0.6)
3.2.1 Finding max, min (0, 0, 0, 0.1, 0.9) (0, 0, 0, 0.1, 0.9)
3.3 Nested if statement (0, 0, 0, 0.4, 0.6) (0, 0, 0, 0.4, 0.6)
4.1 For statement (0, 0, 0, 0.73, 0.27) (0, 0, 0, 0.73, 0.27)
4.2 Calc. sum in a for loop (1, 0, 0, 0, 0) (0, 0, 0, 0.8, 0.2)
4.3 Counting in a for loop (1, 0, 0, 0, 0) (0.55, 0, 0, 0.36, 0.09)
4.4 Calc. avrg in a for loop (1, 0, 0, 0, 0) (0.19, 0, 0, 0.648, 0.162)
4.5 Calc. max/min in a for (0, 0, 0, 0.67, 0.33) (0, 0, 0, 0.67, 0.33)
loop
5.1 While statement (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
5.2 Calc. sum in a while loop (1, 0, 0, 0, 0) (0, 0, 0, 0.8, 0.2)
5.3 Counting in a while loop (1, 0, 0, 0, 0) (0.55, 0, 0, 0.36, 0.09)
5.4 Calc. avrg in a while loop (1, 0, 0, 0, 0) (0.61, 0, 0, 0.312, 0.078)
5.5 Calc. max/min in a while (0, 0, 0, 0.67, 0.33) (0, 0, 0, 0.67, 0.33)
loop
5.6 Do…until (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.1 One-dimension arrays (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.2 Searching (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.3 Sorting (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.4 Two-dimensions arrays (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.5 Processing per rows (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.6 Processing per column (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.7 Processing of diagonals (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
7.1 Functions (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)

and 20 % ‘Assimilated’, KL(C5.3) becomes 36 % ‘Learned’ and 9 % ‘Assimilated’

(the rest 55 % of the particular concept remains ‘Unknown’) and KL(C5.4)
becomes 31.2 % ‘Learned’ and 7.8 % ‘Assimilated’ (the rest 61 % of the particular
concept remains ‘Unknown’) (Interaction II of Table 3.3). Therefore, the increase
of Kate’s knowledge level on C4.2 improves automatically her knowledge level
58 2 Fuzzy Logic in Student Modeling

Table 2.10  Nick’s progress

Domain concepts Learner’s knowledge
Interaction I (μUn, μMKn, Interaction II (μUn, μMKn,
μKn, μL, μA) μKn, μL, μA)
1.1 Constants and variables (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
1.2 Assignment statement (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
1.3 Arithmetic operators (0, 0, 0, 0.02, 0.098) (0, 0, 0, 0.02, 0.098)
1.4 Comparative operators (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
1.5 Logical operators (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
1.6. Mathematic functions (0, 0, 0, 0.12, 0.88) (0, 0, 0, 0.12, 0.88)
1.7 Input-output statements (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
2.1 A simple program’s (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
structure
3.1 If statement (0, 0, 0, 0.3, 0.7) (0, 0, 0, 0.3, 0.7)
3.2 If…else if (0, 0, 0, 0.4, 0.6) (0, 0, 0, 0.4, 0.6)
3.2.1 Finding max, min (0, 0, 0, 0.1, 0.9) (0, 0, 0, 0.1, 0.9)
3.3 Nested if statement (0, 0, 0, 0.4, 0.6) (0, 0, 0, 0.4, 0.6)
4.1 For statement (0, 0, 0, 0.73, 0.27) (0, 0, 0, 0.73, 0.27)
4.2 Calc. sum in a for loop (0, 0, 0, 0.8, 0.2) (0, 0, 1, 0, 0)
4.3 Counting in a for loop (0, 0, 0, 0.6, 0.4) (0, 0, 0, 1, 0)
4.4 Calc. avrg in a for loop (0, 0, 0, 0.7, 0.3) (0, 0, 1, 0, 0)
4.5 Calc. max/min in a for (0, 0, 0, 0.67, 0.33) (0, 0, 0, 0.67, 0.33)
loop
5.1 While statement (0, 0, 0, 1, 0) (0, 0, 0, 1, 0)
5.2 Calc. sum in a while loop (0, 0, 0, 0.8, 0.2) (0, 0, 1, 0, 0)
5.3 counting in a while loop (0, 0, 0, 0.6, 0.4) (0, 0, 0.45, 0.15, 0.4)
5.4 Calc. avrg in a while loop (0, 0, 0, 0.7, 0.3) (0, 0, 0.81, 0, 0.19)
5.5 Calc. max/min in a while (0, 0, 0, 0.67, 0.33) (0, 0, 0, 0.67, 0.33)
loop
5.6 Do…until (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.1 One-dimension arrays (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.2 Searching (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.3 Sorting (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.4 Two-dimensions arrays (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.5 Processing per rows (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.6 Processing per column (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
6.7 Processing of diagonals (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
7.1 Functions (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)

on other related domain concepts, also. Indeed, the fact that the knowledge level
of concept C5.2 became automatically from 100 % ‘Unknown’, 80 % ‘Learned’
and 20 % ‘Assimilated’, without Kate read it, is particular important. This change
triggers the system to infer that C5.2 is already known for Kate.
2.4 A Novel Rule-Based Fuzzy Logic System … 59

• Example 3
Nick had learned the sections 1 (the domain concepts 1.1 to 1.7), 2 (the domain
concept 2.1), 3 (the domain concepts 3.1 to 3.3), 4 (the domain ­concepts 4.1 to
4.5) and some domain concepts 5.1 to 5.5 of the section 5 (Interaction I of
Table 2.10). He revised the concept C5.2. During the revision, he was exam-
ined in the particular domain concept and succeeded 73 %. According to the
above, the value of the defined membership functions for concept C5.2 become
μUn = 0, μMKn = 0, μKn = 1, μL = 0 and μA = 0. According to the FR-CN
(Fig. 2.11) the concept C5.2 affects the preceding concepts C4.2, C4.3, C4.4
and the following concepts C5.3 and C5.4 with “strength of impact” 1, 0.45,
0.81, 0.45 and 0.81 correspondingly. Consequently, applying the fuzzy rule
R8 is: x4.2 = (1 − 1) ∗ 86 + min[1 ∗ 86, 1 ∗ 73] = 73. That degree of suc-
cess corresponds to the fuzzy set ‘Known’ with μKn = 1. (Interaction II of­
Table 3.4). Similarly, applying the same rule, KL(C4.3) becomes 100 % ‘Learned’,
and KL(C4.4) becomes 100 % ‘Known’ (Interaction II of Table 3.4). Furthermore,
according to the rules R3 and R4 (a), KL(C5.3) becomes 45 % ‘Known’, 15 %
‘Learned’ and 40 % ‘Assimilated’ and KL(C5.4) becomes 70 % ‘Known’ and 30 %
‘Assimilated’ (Interaction II of Table 2.10).

2.5 Conclusions and Discussion

Learning is a complicated process. It cannot be accurately said that a learner

knows or does not know a domain concept. For example, a new domain concept
may be completely unknown to the learner but in other circumstances it may be
partly known due to previous related knowledge of the learner. On the other hand,
domain concepts, which were previously known by the learner, may be completely
or partly forgotten. Hence, currently they may be partly known or completely
unknown. In this sense, the level of knowing cannot be accurately represented.
Finally, the teaching process itself changes the status of knowledge of a user.
This is happened due to the fact that a learner accepts new concepts while being
taught. Furthermore, the learner’s knowledge is a moving target. The knowledge
level of a domain concept is increased when the student’s performance is improved.
Alternatively, it is decreased when the student forgets. Improvement of the knowl-
edge level of a domain concept should lead to the increase of the k­ nowledge level
of all the related concepts (prerequisite and following), with his concept. Similarly,
poor performance on a domain concept should lead to decrease of the knowledge
level of all the related concepts with this concept.
In view of the above, an effective adaptive tutoring system has to be responsible
for tracking cognitive state transitions of learners with respect to their progress or
non-progress. The alterations on the state of student’s knowledge level are not lin-
ear. They deal with uncertainty. Thus, a solution to represent these is fuzzy logic.
Therefore, the target of this section was to develop a rule-based fuzzy logic system,
which models the cognitive state transitions of learners, such as forgetting, learning
60 2 Fuzzy Logic in Student Modeling

Table 2.11  Correlation of an e-shop and an adaptive e-learning system concerning the presented
rule-based fuzzy logic system
E-shop E-learning
Nodes Products Domain concepts
Arcs Preferences’ dependencies Knowledge dependencies
Fuzzy sets Descriptions of a preference (e.g. Descriptions of knowledge level (e.g.
‘uninterested’, ‘interested’, ‘liked’, ‘unknown’, ‘insufficiently known’,
‘preferred’) ‘known’, ‘learned’)
Changeable Preferences Knowledge level
states

or assimilating. The presented rule-based fuzzy logic system identifies and updates
each time the student’s knowledge level not only for the current concept, which is
delivered to the learner, but also for all the related concepts with this concept. To
achieve that, the system considers either the learner’s performance or the knowledge
dependencies that exist between the domain concepts of the learning material. In
the particular rule-based fuzzy logic system, fuzzy sets are used in order to describe
how well each individual domain concept is known and learned. Furthermore, it
uses a mechanism of rules over the fuzzy sets, which is triggered after any change
of the value of the knowledge level of a domain concept and updates the values of
the knowledge level of all the related domain concepts with that. Therefore, the edu-
cational system, which has integrated the particular rule-based fuzzy logic system,
is able to makes dynamic decisions on how the teaching syllabus is presented to the
learner to fit his/her personal needs and learning pace.
The operation of the system is based on the knowledge domain representation that
is implemented through a Fuzzy Related-Cognitive Network. This kind of knowl-
edge domain representation helps to manage to represent either the order in which
the domain concepts of the learning material have to be taught and organized, or the
knowledge dependencies that exist between the domain concepts. This is signifi-
cant because the knowledge level of a domain concept increases or decreases due to
changes on the knowledge level of a related domain concept. The design of the learn-
ing material and the definition of the individual domain concepts that it includes, are
based on the knowledge and experience of domain experts. Furthermore, the contri-
bution of domain experts is significant for the definition of the knowledge dependen-
cies that exist among the domain concepts of the learning material and their “strength
o impact” on each other.
The presented rule-based fuzzy logic system is applicable to systems, in which the
user’s changeable state and/or preferences are affected by the existing dependencies
among the system’s elements (like concepts, preferences, events, choices). Thereafter,
the particular system could be implemented in adaptive systems other than adaptive
tutoring system. For example, it could be used in an e-shop, where the preference of
an online shopper for particular products can be used in order to guess and propose
her/him other products that the user is likely to be interested in. In the Table 2.11 the
correlation of an e-shop and an adaptive e-learning system is presented concerning
the particular rule-based fuzzy logic system (Table 2.11).
Chapter 3
A Novel Hybrid Student Model
for Personalized Education

Abstract The goal of each web-based educational system is to offer effective learning
such as real-classroom education and further more. To achieve this goal, the web-based
educational system has to adapt dynamically to each individual student’s needs and
preferences. A solution to this is the student model, which allows the understanding
and identification of each individual student’s needs. In this chapter a novel student
model, which is called F.O.S., is presented. F.O.S. combines three different student
modeling approaches. It combines an overlay model with stereotypes and a rule-based
mechanism. Furthermore, F.O.S. has been fully implemented in a web-based educa-
tional application, which teaches the programming language ‘C’. The particular hybrid
student model allows each individual learner to complete the training program in her/
his own learning pace and abilities.

3.1 Introduction

The rapid development of computer and Internet technologies, and the accessibility of
e-learning applications by a large and heterogeneous group of learners at any time and
place, has led to a rapid and significant growth of Web-based learning environments.
Web-based educational systems offer easy access to knowledge domains and learning
processes from everywhere for everybody at any time. Therefore, they facilitate the
access in knowledge. The goal of each web-based educational system is to offer effec-
tive learning such as real-classroom education and further more. However, traditional
web-based and standalone educational systems still have several shortcomings when
compared to real-life classroom teaching, such as lack of contextual and adaptive sup-
port, lack of flexible support of the presentation and feedback, lack of the collabora-
tive support between students and systems (Xu et al. 2002). This is happened because
at the classroom level, teachers can readjust each time the instructional process and
the teaching strategy considering the student’s needs and abilities. So, the challenge is
to develop Web-based educational systems that adapt dynamically to each individual
student for effective delivery of domain knowledge.

© Springer International Publishing Switzerland 2015 61

K. Chrysafiadi and M. Virvou, Advances in Personalized Web-Based Education,
Intelligent Systems Reference Library 78, DOI 10.1007/978-3-319-12895-5_3
62 3 A Novel Hybrid Student Model …

Web-based educational systems offer easy access to knowledge domains

and learning processes from everywhere for everybody at any time. As a result,
users of web-based educational systems are of varying backgrounds. They have
heterogeneous needs, different levels of knowledge and abilities. That is the
reason why researches in the field of e-learning have expanded their interests on
adaptive e-learning, which is suitable for teaching heterogeneous student popu-
lations (Schiaffino et al. 2008). An adaptive system must be capable of managing
learning paths adapted to each user, monitoring user activities, interpreting those
using specific models, inferring user needs and preferences and exploiting user
and knowledge domain to dynamically facilitate the learning process (Boticario
et al. 2005). In other words, an adaptive educational system has to provide
personalization to the specific needs, knowledge and background of each indi-
vidual student.
A solution is the student model. Student modeling has been introduced in
Intelligent Tutoring Systems, but its use has been extended to most current
educational software applications that aim to be adaptive and personalized.
A student model allows understanding and identification of student needs.
By keeping a model for every user, a system can successfully personalize its
content and utilize available resources accordingly (Kyriacou 2008). For
example, in an adaptive educational application, a student model can be used
to achieve accurate student diagnosis and predict a student’s needs. In return,
it offers individualized courses (Gaudioso et al. 2010), adaptive navigation
support (Castillo et al. 2009), help and feedback to students (Tsiriga and Virvou
2003a; Chrysafiadi and Virvou 2008), allowing them to learn in their own pace
(Chrysafiadi and Virvou 2013c).
The student’s model dimensions and properties correspond to the physical
student’s features and characteristics (Yang et al. 2010). Therefore, in order to
construct a student model, it has to be considered what information and data about
a student should be gathered. The student’s characteristics are: the knowledge
level, the errors and misconceptions, the learning preferences and style, other cog-
nitive features, the emotions, the motivation and meta-cognitive skills. To model
them there is a variety of student modeling techniques to choose: overlay model,
stereotypes, perturbation, constraint-based model, learning machine algorithms,
fuzzy logic, Bayesian networks etc. (Chrysafiadi and Virvou 2013b). However in
most cases there is the need to model more than one student’s characteristics. That
is achieved by using a hybrid student model bringing together features of different
techniques of student modeling.

3.2 Related Work

Each student modeling technique considers, usually, only one or a limited num-
ber of students’ characteristics. However, a student model should consider a sig-
nificant number of student’s characteristics to be effective. Therefore, the need
3.2 Related Work 63

to model a variety of student’s characteristics creates the need for hybrid student
models. A hybrid student model allows the tutoring system to carry out the per-
sonalization efficiently. That is the reason why many adaptive and/or personalized
tutoring systems perform student modeling combing different modeling techniques,
like overlay model with stereotypes, stereotypes with cognitive theories, Bayesian
networks with machine learning techniques etc.
Many researchers have used a hybrid student model, which brings together
various features of different techniques of student modeling, in order to com-
bine various aspects of student’s characteristics. For example, Web-EasyMath
(Tsiriga and Virvou 2002, 2003c) uses a combination of stereotypes with the
machine learning technique of the distance weighted k-nearest neighbor algo-
rithm, in order to initialize the model of a new student. The student is first
assigned to a stereotype category concerning her/his knowledge level and
then the system initializes all aspects of the student model using the distance
weighted k-nearest neighbor algorithm among the students that belong to the
same stereotype category with the new student. A combination of stereotypes
with machine learning techniques has been, also, used in Web-PTV (Tsiriga
and Virvou 2003a, b) and GIAS (Castillo et al. 2009). Furthermore, Inventado
et al. (2010) and Baker et al. (2010) have used a combination of Bayesian
networks and machine learning technique in order to observe students’ reac-
tions and adjust the instruction automatically to each individual learner. Also,
Balakrishnan (2011) has build a student model upon ontology of machine learning
strategies in order to model the effect of affect on learning and recognize for any
learning task, what learning strategy, or combination thereof, is likely to be the
most effective. Millán and Pérez-de-la-cruz (2002) have improved the accu-
racy and efficiency of the diagnosis process through a student model, which
applied Bayesian networks and Adaptive Testing Theory (cognitive theory).
Other adaptive and/or personalized tutoring systems that have used a com-
pound student model, which combines Bayesian networks with cognitive the-
ories, are: ABM (Hernández et al. 2010); AMPLIA (Viccari et al. 2008) and
PlayPhysics (Muñoz et al. 2011). Virvou and Kabassi (2002) have added more
“human” reasoning to F-SMILE using a novel combination of HPR (cognitive
theory) with a stereotype-based mechanism. In addition, the student model of
TADV (Kosba et al. 2003, 2005) combines an overlay model with fuzzy tech-
niques, to represent the knowledge of individual students and their communi-
cation styles. Kavčič (2004a) has been used a similar combination of student
modeling techniques. Furthermore, InfoMap (Lu et al. 2005, 2007) uses an
overlay student model in combination with a buggy model for identification of
the deficient knowledge. Also, KERMIT maintains two kinds of student mod-
els: a constraint-based model and an overlay model (Suraweera and Mitrovic
2004). Glushkova (2008) has applied a qualitative overlay student model to rep-
resent learners’ knowledge level to DeLC system. However, because she wanted
to model, also, learners’ manner of access to training resources, their prefer-
ences, habits and behaviors during the learning process, she have combined the
overlay model with stereotype modeling. A combination of stereotypes with
64 3 A Novel Hybrid Student Model …

overlay model has been performed in ELaC (Chrysafiadi and Virvou 2013c).
Moreover, AUTO-COLLEAGUE (Tourtoglou and Virvou 2008, 2012) performs
student modeling through a hybrid student model based on perturbation and the
stereotype-based modeling technique. A combination of fuzzy logic and
machine learning techniques has been used in ADAPTAPlan (Jurado et al.
2008), while overlay model has been combined with ontologies in Personal
Reader (Dolog et al. 2004), OPAL (Cheung et al. 2010) and IWT (Albano
2011). It is remarkable to refer that a compound student model can include
more than two student modeling techniques. For example, Surjono and Maltby
(2003) have combined an overlay model with perturbation technique and stereo-
types; Chrysafiadi and Virvou (2014) have combined fuzzy techniques with ste-
reotypes and overlay model; the student model of INSPIRE (Grigoriadou et al.
2002; Papanikolaou et al. 2003) combines stereotypes and an overlay model
with fuzzy logic techniques; and the student model of DEPTHS (Jeremić et al.
2012) is a combination of stereotype and overlay modeling with fuzzy rules.
Conclusions about the most common combination of student modeling tech-
niques are drawn considering the hybrid student models of the literature review.
An overlay student model usually is combined with stereotypes or fuzzy logic
techniques. Stereotypes are blended, mainly, with overlay, but they are also
combined with machine learning or fuzzy logic techniques. Perturbation stu-
dent model is combined only with overlay and stereotypes. Machine learning
techniques are used mostly to support stereotype modeling, but there is, also,
an interest to combine them with Bayesian networks. Cognitive theories are
applied with stereotypes and Bayesian Networks. Fuzzy logic is usually used
with overlay or stereotype student models. Bayesian networks are blended,
mainly, with machine learning techniques and cognitive theories, but they are,
also, combined with stereotypes. Ontologies are primarily combined with overlay
student modeling.

3.3 The F.O.S. Hybrid Student Model

A hybrid student model, which brings together various features of different

techniques of user modeling, is the solution for offering a more adaptive
learning system. The reason for this is the fact that the student model needs
to combines various aspects of student’s characteristics that is both domain
dependent and domain independent in order to carry out the personalization
efficiently (Yang et al. 2010). This way, the model not only can exhibit unique
individual characteristics and preferences of each learner by monitoring
and tracing the changes of their knowledge, skills, interests, but also classify
the learners according to their performance, individual learning behaviors
and activities (Yang et al. 2010). That is the reason for the development of a
novel hybrid student, which combined overlay technique and stereotypes with
fuzzy logic.
3.3 The F.O.S. Hybrid Student Model 65

Fig. 3.1  The hybrid student model F.O.S.

The particular hybrid student model is called Fuzzy logic system, Overlay
and Stereotypes (F.O.S.). It includes a rule-based fuzzy logic system, which
is responsible for tracking and updating the knowledge level of each domain
concept of the learning material for each individual system; an overlay model,
which represents the knowledge level of the student; and a three-dimensional
stereotype model (Fig. 3.1). The fuzzy rules are used to define the learner’s
knowledge level of each domain concept of the learning material. Therefore, the
results of the application of that rules determine the overlay model. Information
of the overlay model is used to define the first dimension of the stereotype
model, which concerns the learner’s knowledge level and vary from novices to
experts. The second dimension, which concerns the type of errors that a learner
can make, and the third dimension, which concerns the prior knowledge of the
student on related knowledge domain fields, are not determined by the rest parts
of the student model.

3.3.1 Fuzzy Rules

It is not a straightforward task to define for each learner which concepts are
unknown, known or assimilated and at what degree. The student’s progress
assessment includes statements like “The concept A is 72 % ‘Known’ for the
student”, “The concept X is 60 % ‘insufficiently known’ for the learner”, “The
student has learnt 100 % the concept Y”, “The concept Z is 100 % ‘unknown’
for student Z”. This information is imprecise. One possible approach to deal
with this is fuzzy set techniques, with their ability to naturally represent human
conceptualization. That is the reason for the integration of fuzzy logic techniques
into the hybrid student model.
The rule-based fuzzy logic system is responsible for identifying and
­updating the student’s knowledge level of all the concepts of the knowledge
66 3 A Novel Hybrid Student Model …

domain. It uses fuzzy sets to represent the student’s knowledge level and a
mechanism of rules over the fuzzy sets, which is triggered after a change has
occurred on the student’s knowledge level of a domain concept. This mechanism
updates the student’s knowledge level of all related with this concept, concepts.
Its operation is based on the knowledge dependencies that exist between the
domain concepts of the learning material and their “strength of impact” on
each other.
Fuzzy sets are used to characterize the changeable user’s state. For example,
(“Unknown”, “Known”, “Learned”} or (“Unknown”, “Insufficiently Known”,
“Known”, “Learned”, “Assimilated”} are the fuzzy sets of educational adap-
tive systems. Therefore, FS1, FS2, … , FSn are the defined fuzzy sets and µFSi ,
i = 1, 2, 3, … , n are the corresponding membership functions. Therefore, a set
(µFS1 , µFS2 , µFS3 , . . . , µFSn ) is used to express the student knowledge of a domain
concept with µFS1 + µFS2 + µFS3 + · · · + µFSn = 1. The fuzzy rules are depicted
in Fig. 3.2.

Fig. 3.2  The fuzzy rules

3.3 The F.O.S. Hybrid Student Model 67

The results of the application of the particular fuzzy rules update the overlay
model of the hybrid student model.

3.3.2 Overlay Model

One of the most popular and common used student models is the overlay model.
It was invented by Stansfield et al. (1976) and has been used in many systems ever
since. The main assumption underlying the overlay model is that a student may
have incomplete but correct knowledge of the domain. Therefore, according to the
overlay modeling, the student model is a subset of the domain model (Martins et al.
2008; Vélez et al. 2008), which reflects the expert-level knowledge of the subject
(Brusilovsky and Millán 2007; Liu and Wang 2007). The differences between the
student’s and the expert’s set of knowledge are believed to be the student’s lack
of skills and knowledge, and the instructional objective is to eliminate these dif-
ferences as much as possible (Bontcheva and Wilks 2005; Michaud and McCoy
2004; Staff 2001). Consequently, the domain is decomposed into a set of ele-
ments and the overlay model is simply a set of masteries over those elements
(Nguyen and Do 2008). The pure overlay model assigns a Boolean value, yes
or no, to each element, indicated whether the student knows or does not know
this element, while in its modern form, an overlay model represents the degree
to which the user knows such a domain element by using a qualitative measure
(good-average-poor) or a quantitative measure such as the probability that the
student knows the concept (Brusilovsky and Millán 2007).
A fuzzy-weighted qualitative overlay model is used in the presented hybrid
student model. A qualitative weighted overlay model is an extension of the pure
overlay model that can distinguish several levels of student’s knowledge about
each concept representing user knowledge of a concept as a qualitative value
(Brusilovsky and Anderson 1998; Papanikolaou et al. 2003). In the presented
novel hybrid student model, the overlay model uses qualitative values, like
(‘unknown’, ‘insufficiently known’, ‘known’, ‘learned’), which corresponds to the
fuzzy sets. Furthermore, it uses a set of fuzzy values (µFS1 , µFS2 , µFS3 , . . . , µFSn ),
which expresses the degree in which each of the above fuzzy sets (qualitative
values) are active. For example, if (‘unknown’, ‘insufficiently known’, ‘known’,
‘learned’) are the qualitative values and (0, 0.3, 0.6, 0.1) is the fuzzy values that
characterize the concept C1 in the overlay model, then it means that the particular
concept is 30 % ‘insufficiently known’, 60 % ‘known’ and 10 % ‘learned’. That
is the reason for the name ‘fuzzy-weighted qualitative overlay model’.
Figure 3.3 depicts an example of the presented fuzzy-weighted qualitative
overlay model. The concepts, which are colored green, belong to the subset of
the domain model that the learner knows or has assimilated. The presented fuzzy-
weighted qualitative overlay model is used to model the variations of the l­earner’s
knowledge level. Particularly, it is used to inform the system which domain
68 3 A Novel Hybrid Student Model …

Fig. 3.3  A fuzzy-weighted qualitative overlay model [the fuzzy set in each node corresponds to
the qualitative values (‘unknown’, ‘insufficiently known’, ‘known’, learned’)]

concepts are learned, which domain concepts are partly known and which domain
concepts are completely unknown.

3.3.3 Stereotypes

Another common used approach of student modeling is stereotyping. Stereotypes

were introduced to user modeling by Rich (1979) in the system called GRUNDY.
The main idea of stereotyping is to create groups of students with common charac-
teristics. Such groups are called stereotypes. In other words, a stereotype normally
contains the common knowledge about a group of users. A new user will be assigned
into a related stereotype if some of his/her characteristics match the ones contained
in the stereotype. Each stereotypes has a set of trigger conditions, which activate the
stereotype if they are true, and a set of retraction conditions, which deactivate the
stereotype if they are true to Kay (2000).
The stereotype student model of the presented hybrid student model is three-
dimensional (Fig. 3.4). The first dimension (KL) consists of stereotypes that rep-
resent the learner’s knowledge level. They vary from novices to experts. The value
of KL is defined considering the information of the fuzzy-weighted qualitative
overlay model. A learner is classified to a knowledge level (KL) stereotype cat-
egory according to which domain concepts the learner knows and how well. The
particular stereotype category gives information about the learning material that
should be delivered to the learner. The second dimension (ErrTyp) consists of two
stereotypes and concerns the type of errors that a learner can make. It helps the
system to reason the learner’s performance. For example, the system can infer if
3.3 The F.O.S. Hybrid Student Model 69

Fig. 3.4  The three-

dimensional stereotype model

the learner reads, if s/he has difficulties in understanding, if s/he is careless, if s/he
has confused with a prior knowledge on a related concept etc. Finally, the third
dimension (PrK) concerns prior knowledge of the student on related knowledge
domain fields. In this way, the tutoring system is able to distinguish if an error
occurs due to non-learning or due to affecting by prior knowledge.
The stereotypes are updated each time new information about the learner
is required. New information about the learner is obtained each time s/he inter-
acts with the system. More concretely, each time the learner interacts with the
system, s/he takes a test, the results of which determine the learner’s knowledge
and update her/his overlay model. The first dimension of the stereotype student
model receives information from the overlay model and determines the value of
KL. The stereotype categories of the second and third dimension, to which the
learner should be classified, are not affected by the information that is received by
the overlay model. The stereotype category of the second dimension to which the
learner belongs each time, is determined by the type of errors that s/he does dur-
ing the test. Also, the third dimension is determined by the learner during her/his
registration to the system.

3.4 Operation of F.O.S.

When the learner interacts with the system for the first time, s/he asked to enter
static information like her/his age, name and the prior knowledge (PrK) that
s/he has on related fields with the knowledge domain of the system. Initially,
s/he is considered to be novice. After that, the student model is updated each
time new information about the learner is required. New information about the
learner is obtained each time s/he interacts with the system. More concretely,
each time the learner interacts with the system, s/he takes a test, the results of
which determine the learner’s knowledge and errors and update her/his over-
lay model. In particular, each time the learner interacts with the system, s/he
reads a domain concept Ci and takes a test to assess her/his knowledge on the
particular domain concept. The results of the test determine either the learner’s
70 3 A Novel Hybrid Student Model …

knowledge on the domain concept Ci, or the type of errors (ErrTyp) that s/he
made. Then, the system applies the fuzzy rules in order to identify and update
the student’s knowledge level of all the related concepts with Ci. The values
of the knowledge level of all the domain concepts of the learning material
that came off the application of the fuzzy rules are used for the definition of
the fuzzy-weighted qualitative overlay model. So, the overlay student model
is updated.
The new information of the overlay model triggers the KL stereotype
category. Therefore, the system decides to which stereotype category of knowl-
edge level, it should classify the learner concerning the learner’s knowledge
level of the domain concepts of the learning material. If s/he succeeds in the
test, then s/he is transited to a next knowledge level and the value of KL is
increased. Otherwise, if s/he fails, then s/he remains to the same knowledge
level or s/he is returned to a previous knowledge level, according to her/his
errors. In particular, if the learner’s poor performance does not affect the knowl-
edge level of other related domain concepts, which belong to a previous section,
then the value of KL remains the same; otherwise the value of KL is decreased.
If the learner makes errors that correspond to concepts of previous section,
then the system infers that s/he has forgot something from previous sections of
the learning material. In particular, if the learner makes errors, which consider
concepts of previous knowledge level, then the system checks the value of the
stereotype that corresponds to her/his prior knowledge on related fields. If the
system decides that the errors were made due to confusion with prior knowl-
edge, then it does not classify the learner to a previous knowledge level, but it
points out the error. Otherwise, it classifies the learner to a previous knowledge
level reducing the value of KL.
Therefore, the operation of F.O.S. is described in the following steps:
1. The learner registers into the system and enters her/his age, name and her/his
prior knowledge (PrK) on related fields.
2. The learning material is delivered to her/him.
3. S/he is examined in domain concept Ci taking a test.
4. The system identifies the type of errors that the learner made (ErrTyp).
5. If the learner makes errors due to confusion with prior knowledge, then it does
not consider them into calculating her/his performance, but it points out the error.
6. The system identifies the learner’s knowledge level on the concept Ci.
7. The system applies the fuzzy rules and defined the alterations on the learner’s
knowledge level of all the related with Ci concepts, considering either the
knowledge dependencies that exist between the domain concepts of the learn-
ing material or the current learner’s knowledge level on the domain concepts of
the learning material.
8. The system updated the fuzzy-weighted qualitative overlay model.
9. The system advises the overlay model and classifies the learner to the appropriate
stereotype category of knowledge level (KL).
3.4 Operation of F.O.S. 71

The operation of F.O.S. leads the system to make inferences about the
changes of the user’s state and make useful adaptation decisions, offering
dynamic adaptation to users’ needs. With this approach the system identifies
the alterations on the state of student’s knowledge level, such as forgetting or
learning are represented. These states determine the progress of the learner each
time. They are revealed by the transition from one stereotype of the student
model to another. Thus, the system has to decide which stereotypes have to be
activated and which stereotypes have to be deactivated, at each interaction of a
learner with educational application.
As Tretiakov et al. (2005) have been stated, a state-chart diagram can been
used to show the sequence of a student’s mind. Consequently, the educational
system has to construct statechart diagrams to track the cognitive state transitions

Table 3.1  The relationship between cognitive state and KL stereotype transitions

Cognitive state Stereotype transitions
S/he does not S/he doesn’t read No transition to other KL stereotype
learn S/he reads but not learn (KL remains the same)
S/he reads but s/he has difficulty
in understanding
S/he learns Transition to the next KL stereotype
(KL increases)
S/he forgets Transition to a previous KL stereo-
type (KL decreases)
S/he reaches target knowledge Continuous transition to the
advanced KL stereotypes
(KL increases continuously)

Fig. 3.5  Sequence of the changes of the learner’s cognitive state in relation with her/his progress
or no-progress
72 3 A Novel Hybrid Student Model …

of learners. Table 3.1 depicts the relation between a learner’s cognitive state and
the transitions among the KL stereotypes of the presented hybrid student model.
Furthermore, Fig. 3.5 depicts the sequence of the changes of a learner’s cognitive
state in relation with her/his progress or no-progress.

3.5 Application of F.O.S. in a Programming

Tutoring System

The F.O.S. has been incorporated into an innovative integrated e-learning environ-
ment for computer programming and the language ‘C’. The F.O.S. is responsible
for identifying and updating the student’s knowledge level, taking the different
pace of learning of each individual learner into account. The system can adapt
dynamically to each individual learner’s needs by scheduling the sequence of les-
sons on the fly. This personalization allows each learner to complete the e-training
course at their own pace and according to their ability.
In particular, the system retains static information about each student, such
as her/his previous experience on computer programming and the programming
languages that s/he already knows. It also retains dynamic information, such as
errors, misconceptions and progress. F.O.S. allows the system to recognize when
a new domain concept is completely unknown to the learner, or when it is partly
known due to the learner having previous related knowledge. Furthermore, the
system recognizes when a previously-known domain concept has been completely
or partly forgotten by the learner. Thus it models either the possible increase or
decrease of the learner’s knowledge. Also, each time the system checks if the
learner’s errors were due to possible confusion with features of another previ-
ously-known programming language. In this case, it responds accordingly. Below,
the application of F.O.S. in the programming tutoring system of the language ‘C’
is presented.

3.5.1 Fuzzy Rules

The defined fuzzy sets and their membership functions are the following:
• Unknown (Un): the degree of success in the domain concept is from 0 to 50 %.
• Moderate Known (MKn): the degree of success in the domain concept is from
40 to 70 %.
• Known (Kn): the degree of success in the domain concept is from 60 to 80 %.
• Learned (L): the degree of success in the domain concept is from 75 to 90 %.
• Assimilated (A): the degree of success in the domain concept is from 85 to
100 %.
3.5 Application of F.O.S. in a Programming Tutoring System 73


 1, x ≤ 40
µUn = x−40
1− 10 , 40 < x < 50
0, x ≥ 50


x−40


 10 ,
40 < x < 50
1, 50 ≤ x ≤ 60

µMKn = x−60

 1 − 10 , 60 < x < 70
0, x ≤ 40 or x ≥ 70


x−60


 10 ,60 < x < 70
1, 70 ≤ x ≤ 75

µKn = x−75

 1 − 5 , 75 < x < 80
0, x ≤ 60 or x ≥ 80


x−75


 5 ,75 < x < 80
1,
80 ≤ x ≤ 85

µL = x−85

 1 − 5 , 85 < x < 90
0, x ≤ 75 or x ≥ 90


x−85

 5 , 85 < x < 90
µA = 1, 90 ≤ x ≤ 100
0, x ≤ 85


Concerning two domain concepts Ci and Cj where Ci is taught before Cj, the
knowledge level of the concepts can change according to the fuzzy rules that are
depicted in Fig. 3.6 (how the knowledge level of Cj changes according to updates
of the knowledge level of Ci) and Fig. 3.7 (how the knowledge level of Ci changes
according to updates of the knowledge level of Cj).

3.5.2 Overlay Model

The qualitative values of the fuzzy-weighted qualitative overlay model are the
defined fuzzy sets. In other words, they are the values: ‘unknown’, ‘moderate
known’, ‘known’, ‘learned’ and ‘assimilated’. Therefore, the overlay model uses a
quintet (μUn, μMKn, μKn, μL, μA), which expresses the degree in which each of
the above qualitative values are active (Fig. 3.8). For example, (0, 0, 0.6, 0.3, 0.1)
declares that the domain concept is 60 % ‘known’, 30 % ‘learned’ and 10 % ‘assim-
ilated’. Similarly, (0.7, 0.3, 0, 0, 0) declares that the concept is 70 % ‘Unknown’
and 30 % ‘moderate known’.
74 3 A Novel Hybrid Student Model …

Fig. 3.6  The fuzzy rules (how the knowledge level of Cj changes according to updates of the
knowledge level of Ci)

3.5.3 Stereotypes

3.5.3.1 KL Stereotype Category

The particular stereotype category is the first dimension of the third-dimension

stereotype model. It concerns the knowledge level of the learner. Its value
­represents the expertise of the learner in the algorithms and the programming
language ‘C’. The value of the KL stereotype varies from “novice” users, who
do not have a structural knowledge of programming and are unable to give an
3.5 Application of F.O.S. in a Programming Tutoring System 75

Fig. 3.7  The fuzzy rules (how the knowledge level of Ci changes according to updates of the
knowledge level of Cj)

acceptable answer in most cases, to “expert” users, who are able to select, use
and combine programming structures creating complex programs.
For the definition of these stereotypes the conceptual framework for analyzing
students’ knowledge of programming that was developed by McGill and Volet (1997)
and the evaluation method of knowledge of programming that was developed by
deRaadt (2007) have been advised. McGill and Violet discern three knowledge types
in the view of cognitive psychology: declarative (the basic knowledge of an object),
procedural (how to use declarative knowledge for problem solving and decision
making), strategic (upper ­knowledge level), and three knowledge types in the view
of educational research: syntactic (basic knowledge), conceptual (be able to combine
76 3 A Novel Hybrid Student Model …

Fig. 3.8  An example of the fuzzy weighted qualitative overlay model of the presented program-
ming tutoring system

knowledge, ­analytical thought) and strategic (integrated knowledge). De Raadt sug-

gests five knowledge levels:
• No answer: no knowledge
• Pre-structural: substantial lack of knowledge
• One-structural: ability to describe a part of code
• Multi-structural: ability to describe a program line-line
• Relational: ability to describe the whole of a program.
Therefore, the eight stereotypes of the KL stereotype category (Table 3.2) were
defined considering the above frameworks. A learner is classified to a KL stereotype
category according to which chapters the learner knows and how well. That kind of
information is derived from the fuzzy-weighted qualitative overlay model. The KL
stereotype category to which the learner has been classified, gives information about
the learning material that should be delivered to the learner (Fig. 3.9).

3.5.3.2 ErrTyp Stereotypes Category

ErrTyp stereotype takes two values: prone to syntax errors and prone to ­logical
errors. Syntax errors are recognized if they belong in one of the following
­categories: anagrammatism of commands’ names, omission of the definition of
data, using invalid command names etc. They, usually, indicate that the learner
has not read carefully and has not known adequately the chapters that correspond
to her /his knowledge level. Logical errors are usually errors of design and occur
3.5 Application of F.O.S. in a Programming Tutoring System 77

Table 3.2  Values of the KL stereotype category

KL stereotype Knowledge type Knowledge level
McGill and Volet (1997) de Raadt (2007)
Stereotype 1: novice No knowledge Level 1
Stereotype 2: the learner knows the basics Declarative—syntactic Pre-structural
of the programming language C and the
sequence structure of programming
Stereotype 3: the learner knows basics of Declarative—conceptual One-structural
the programming language C, the sequence
structure and the structures of choice
Stereotype 4: the learner knows basics of Procedural—syntactic One-structural
the programming language C, the sequence
structure, the structures of choice and the
iteration structure with concrete number
of loops
Stereotype 5: the learner knows basics of Procedural—syntactic Multi-structural
the programming language C, the sequence
structure, the structures of choice and the
iteration structures with concrete or unknown
number of loops
Stereotype 6: the learner knows basics of Procedural—conceptual Multi-structural
the programming language C, the sequence
­structure, the structures of choice, the i­teration
structures and one-dimensional arrays
Stereotype 7: the learner/he knows basics of Procedural—conceptual Relational
the programming language C, the sequence
structure, the structures of choice, the itera-
tion structures and all the type of arrays
Stereotype 8: experts Strategic Relational

in case of misconceptions of the program and of the semantics and operation of

the commands. They, usually, indicate that the learner has a difficulty in under-
standing the instructions and their logic.

3.5.3.3 PrK Stereotype Category

Furthermore, the system should known if a learner has prior knowledge of another
programming language in order to be able to distinguish if an error occurs due to
non-learning or due to affecting by another language. This kind of information is
derived by the 3rd-dimension of the stereotype model (PrK). Therefore, the PrK
stereotype category is associated with other programming languages that the
learner may already know. In particular, PrK takes the following values: ‘none’,
‘Basic’, ‘Pascal’, ‘Java’. If a learner does not know a programming language, PrK
takes the value “none”. If a learner knows more than one programming language of
the above, PrK takes two or all these programming languages. So, it takes either the
value ‘none’, or one or more values of the set {Basic, Pascal, Java}.
78 3 A Novel Hybrid Student Model …

C1.1. constants & variables

C1.2. assignment statement

C1.3. arithmetic opertors
C1. Basics C1.4. comparative operators

C1.5. logical opertors

C1.6. mathematical functions
C1.7. input-output statements

C2. Sequence C2.1. a simple program struture

structure

C3.1. if statemet
C3. Conditional
C3.2. if...else if
Structures
C3.2.1 methodology of finding max/min
C3.3. nested if

C4.1. for statemet

C4. lteration C4.2. calculating sum in a for loop

Structure
C counting in a for loop
Concrete no 4.3.
of loops C4.4. calculating avgr in a for loop
C4.5. calculating max/min in a for loop

C5.1. while statement

C5.2. calculating sum in a while loop

C5. lteration C5.3. counting in a while loop
Structure
Unknown no of C5.4. calculating avgr in a while loop
loops C5.5. calculating max/min in a while loop

C5.6. do...while statement

C6.1. one-dimenstional arrays

C6. Arrays C6.2. searching

C6.3. sorting

C6.4. two-dimenstional arrays

C6.5. processing per now
C6.6. processing per column
C6.7. processing of diagonals

C7. sup-
C7.1. functions
programming

Fig. 3.9  The delivered learning material concerning the learner’s KL stereotype

3.5 Application of F.O.S. in a Programming Tutoring System 79

Fig. 3.10  The transitions among the learner’s cognitive states

3.5.4 Cognitive State Transitions of Learners

of the Programming Tutoring System

Taking into account all the above, it can be concluded that a domain concept
passes through five states (‘unknown’, ‘moderate known’, ‘known’, ‘learned’,
‘assimilated’) during the interaction of the user with the system. Furthermore, a
learner passes through several states during the learning process. S/he can learn or
not, forget, assimilate or not etc. These states determine the progress of the learner
each time; they determine the transition from one KL stereotype to another. Also,
the transition from one KL stereotype to another can reveal the state of the learner.
Figure 3.10 depicts the transitions among the learner’s cognitive states for the
­presented programming tutoring system.

3.6 Examples of Operations

The presented programming tutoring system was used in a postgraduate program

in the field of informatics at the University of Piraeus, in order to offer dynami-
cally personalized e-training in computer programming and the language C.
The learners used the system without attending any complementary course on
programming, over a period of six months. They had different ages and diverse
80 3 A Novel Hybrid Student Model …

backgrounds. Examples of such backgrounds are physics, mathematics, computer

science, education, human and social science.
At the beginning, all learners were considered to be novices. At the next inter-
actions, the system delivered to them the appropriate learning material for each
individual student’s needs by adapting instantly to the learner’s individual learning
pace. The system’s adaptation decisions were based on the values of the student
model. The student model is updated each time the learner interacts with the ­system
and takes a test. There were two kinds of tests: (1) the tests that corresponded to
each individual domain concept of the learning material (practice tests), (2) the
final tests that corresponded to the sections of the learning material (they included
exercises of a variety of domain concepts). In particular, each time the learner read
a domain concept (Ci), s/he had to complete a corresponding practice test. When,
the learner had completed successfully (μA(Ci) = 1 or μA(Ci) + μL(Ci) = 1)
all the practice tests of the domain concepts of a section (e.g. iterations with
concrete number of loops, arrays, sub-programming), then s/he had to com-
plete the final test of the section. If s/he succeeded to the final test (μA(Ci) = 1
or μA(Ci) + μL(Ci) = 1 for all the concept Ci of the particular section), then
s/he transited to a next section. Otherwise, s/he had advised to revise some domain
­concepts. Representative examples of the system’s implementation follow.
Example 1 Elena’s current student model has the following values: KL = 3,
ErrTyp = “prone to syntax errors”, PrK= “none”. The value KL = 3 comes off
her current overlay model (Table 3.3, column ‘before’). ErrTyp is “prone to syn-
tax errors” due to the fact that she had made usually errors that concern anagram-
matism of commands’ names or invalid symbolisms of operands or commands’
names. Also, PrK= “none” indicates that Elena does not have previous knowledge
on computer programming.
She is examining in C4.2: “calculating sum in a ‘for’ loop” and is succeeding 92 %.
So, the quintet, which describes Elena’s knowledge level on C4.2, is (0, 0, 0, 0, 1).
However, according to the “strength of impact” of the knowledge dependencies that
exist between the domain concepts of the learning material (Table 2.2), C4.2 affects
45 % the concept C4.3, 81 % the concept C4.4, 100 % the concept C5.2, 45 % the con-
cept C5.3, and 39 % the concept C5.4.
According to the rule R2 (d) over the fuzzy sets (Fig. 3.6) the following occur
(Table 3.3, column ‘after—interaction I’):
• μA(C4.3) = 0.45 and it remains 55 % ‘Unknown’ (μUn(C4.3) = 0.55) So, the
quintet for C4.3 is (0.55, 0, 0, 0, 0.45).
• μA(C4.4) = 0.81 and it remains 19 % ‘Unknown’ (μUn(C4.4) = 0.19) So, the
quintet for C4.4 is (0.19, 0, 0, 0, 0.81).
• μA(C5.2) = 1. So, the quintet for C5.4 is (0, 0, 0, 0, 1).
• μA(C5.3) = 0.45 and it remains 55 % ‘Unknown’ (μUn(C5.3) = 0.55) So, the quintet
for C5.3 is (0.55, 0, 0, 0, 0.45).
• μA(C5.4) = 0.52 and it remains 55 % ‘Unknown’ (μUn(C5.4) = 0.48). So, the
quintet for C5.4 is (0.48, 0, 0, 0, 0.52).
3.6 Examples of Operations 81

Table 3.3  Elena’s progress

Domain Learner’s knowledge
concepts Before After
(KL = 3) Interaction I Interaction II Interaction Interaction IV
(KL = 3) (KL = 3) III (KL = 3) (KL = 4)
C1.1 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C1.2 (0,0, 0, 0.08, (0,0, 0, 0.08, (0,0, 0, 0.08, (0,0, 0, 0.08, (0,0, 0, 0.08,
0.92) 0.92) 0.92) 0.92) 0.92)
C1.3 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C1.4 (0,0, 0,0.08, (0,0, 0,0.08, (0,0, 0,0.08, (0,0, 0,0.08, (0,0, 0,0.08,
0.92) 0.92) 0.92) 0.92) 0.92)
C1.5 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C1.6 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C1.7 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C2.1 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C3.1 (0, 0, 0, 0.2, (0, 0, 0, 0.2, (0, 0, 0, 0.2, (0, 0, 0, 0.2, (0, 0, 0, 0.2,
0.8) 0.8) 0.8) 0.8) 0.8)
C3.2 (0, 0, 0, 0.6, (0, 0, 0, 0.6, (0, 0, 0, 0.6, (0, 0, 0, 0.6, (0, 0, 0, 0.6,
0.4) 0.4) 0.4) 0.4) 0.4)
C3.2.1 (0, 0, 0, 0.3, (0, 0, 0, 0.3, (0, 0, 0, 0.3, (0, 0, 0, 0.3, (0, 0, 0, 0.21,
0.7) 0.7) 0.7) 0.7) 0.79)
C3.3 (0, 0, 0, 0.74, (0, 0, 0, 0.74, (0, 0, 0, 0.74, (0, 0, 0, 0.74, (0, 0, 0, 0.74,
0.26) 0.26) 0.26) 0.26) 0.26)
C4.1 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C4.2 (1, 0, 0, 0, 0) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C4.3 (1, 0, 0, 0, 0) (0.55, 0, 0, 0, (0.28, 0, 0, (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
0.45) 0.27, 0.45)
C4.4 (1, 0, 0, 0, 0) (0.19, 0, 0, 0, (0, 0, 0, 0.6, (0, 0, 0, 0.55, (0, 0, 0, 0.55,
0.81) 0.4) 0.45) 0.45)
C4.5 (0.63, 0, 0, (0.63, 0, 0, (0.63, 0, 0, (0.63, 0, 0, (0, 0, 0, 0, 1)
0.11, 0.26) 0.11, 0.26) 0.11, 0.26) 0.11, 0.26)
C5.1 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C5.2 (1, 0, 0, 0, 0) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C5.3 (1, 0, 0, 0, 0) (0.55, 0, 0, 0, (0.28, 0, 0, (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
0.45) 0.27, 0.45)
C5.4 (1, 0, 0, 0, 0) (0.48, 0, 0, 0, (0.17, 0, 0, (0.1, 0, 0, (0.1, 0, 0,
0.52) 0.31, 0.52) 0.18, 0.72) 0.18, 0.72)
C5.5 (0.63, 0, 0, (0.63, 0, 0, (0.63, 0, 0, (0.63, 0, 0, (0, 0, 0, 0, 1)
0.11, 0.26) 0.11, 0.26) 0.11, 0.26) 0.11, 0.26)
C5.6 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.1 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.2 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.3 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.4 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.5 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
(continued)
82 3 A Novel Hybrid Student Model …

Table 3.3  (continued)
Domain Learner’s knowledge
concepts Before After
(KL = 3) Interaction I Interaction II Interaction Interaction IV
(KL = 3) (KL = 3) III (KL = 3) (KL = 4)
C6.6 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.7 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C7.1 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)

Therefore, after the interaction I Elena’s student model becomes: KL = 3,

ErrTyp = “prone to logical errors”, PrK= “none”. The value of the ErrTyp
­stereotype changes and becomes “prone to logical errors” due to the fact that
the last errors of Elena are associated with the semantics and operation of the
commands. This value indicates that Elena has a difficulty in understanding the
instructions and their logic.
At the next interaction, Elena is examined in the domain concept C4.4 and is
succeeding 87 %. C4.4 affects 100 % the concept C4.2, 45 % the concept C4.3,
100 % the concept C5.2, 45 % the concept C5.3, and 52 % the concept C5.4.
According to the fuzzy rules (Figs. 3.6 and 3.7) the following occur (Table 3.3,
column ‘after—interaction II’):
• According to R7, μA(C4.2) remains 1.
• According to R5 and R6 (c), μA(C4.3) = 0.45, μL(C4.3) = 0.27 and it remains
28 % ‘Unknown’ (μUn(C4.3) = 0.28). So, the quintet for C4.3 is (0.28, 0, 0, 0.27,
0.45).
• According to R3 μA(C5.2) remains 1.
• According to R1 and R2 (c) μA(C5.3) = 0.45, μL(C5.3) = 0.27 and it remains
28 % ‘Unknown’ (μUn(C5.3) = 0.28). So, the quintet for C5.3 is (0.28, 0, 0, 0.27,
0.45).
• According to R1 and R2 (c) μA(C5.4) = 0.52, μL(C5.4) = 0.31 and it remains 17 %
‘Unknown’ (μUn(C5.4) = 0.17). So, the quintet for C5.4 is (0.17, 0, 0, 0.31, 0.52).
After the interaction II Elena’s student model becomes: KL = 3, ErrTyp = “prone
to logical errors”, PrK= “none”.
At the next interactions, Elena is examining in the domain concepts C4.3 and
C4.5 and is succeeding 95 and 90 % respectively. Applying the fuzzy rules, the
knowledge level of Elena changes as it is presented in Table 3.3, column ‘after—
interaction III and interaction IV’.
When all the domain concepts of the section 4 (i.e. the domain concepts 4.1 to
4.5) become learned, the value of the stereotype category KL of Elena’s student
model increases and becomes 4. Now, the domain concepts of section 5 (i.e. the
domain concepts 5.1 to 5.6) should be delivered to Mary. However, the domain
concepts C5.2, C5.3, C5.4 and C5.5 are considered already learned. Therefore Elena
has to read only the domain concepts C5.1 and C5.6.
3.6 Examples of Operations 83

Table 3.4  Kostas’ progress Domain Learner’s knowledge

concepts Before After
(KL = 6) (KL = 6)
C1.1 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C1.2 (0,0, 0, 0.08, 0.92) (0,0, 0, 0.08, 0.92)
C1.3 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C1.4 (0,0, 0,0.08, 0.92) (0,0, 0,0.08, 0.92)
C1.5 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C1.6 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C1.7 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C2.1 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C3.1 (0, 0, 0, 0.2, 0.8) (0, 0, 0, 0.2, 0.8)
C3.2 (0, 0, 0, 0.6, 0.4) (0, 0, 0, 0.6, 0.4)
C3.2.1 (0, 0, 0, 0.3, 0.7) (0, 0, 0, 0.3, 0.7)
C3.3 (0, 0, 0, 0.74, 0.26) (0, 0, 0, 0.74, 0.26)
C4.1 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C4.2 (0, 0, 0, 0.12, 0.88) (0, 0, 0, 0.12, 0.88)
C4.3 (0, 0, 0, 0.25, 0.75) (0, 0, 0, 0.25, 0.75)
C4.4 (0, 0, 0, 0.24, 0.76) (0, 0, 0, 0.24, 0.76)
C4.5 (0, 0, 0, 0.31, 0.69) (0, 0, 0, 0.31, 0.69)
C5.1 (0, 0, 0, 0.3, 0.7) (0, 0, 0, 0.3, 0.7)
C5.2 (0, 0, 0, 0.12, 0.88) (0, 0, 0, 0.12, 0.88)
C5.3 (0, 0, 0, 0.25, 0.75) (0, 0, 0, 0.25, 0.75)
C5.4 (0, 0, 0, 0.23, 0.77) (0, 0, 0, 0.23, 0.77)
C5.5 (0, 0, 0, 0.31, 0.69) (0, 0, 0, 0.31, 0.69)
C5.6 (0, 0, 0, 0.16, 0.84) (0, 0, 0, 0.16, 0.84)
C6.1 (0, 0, 0.15, 0.6, (0, 0, 0.15, 0.6, 0.25)
0.25)
C6.2 (0, 0, 0.5, 0.5, 0) (0, 0, 0.5, 0.5, 0)
C6.3 (0, 0, 0.3, 0.43, (0, 0, 0.3, 0.43, 0.27)
0.27)
C6.4 (0, 0, 0.2, 0.3, 0.5) (0, 0, 0, 0.5, 0.5)
C6.5 (0, 0, 0.73, 0.1, (0, 0, 0, 0.8, 0.2)
0.17)
C6.6 (0, 0, 0.73, 0.1, (0, 0, 0.11, 0.63,
0.17) 0.17)
C6.7 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C7.1 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)

Example 2 Kostas’ current student model has the following values: KL = 6,

ErrTyp = “prone to logical errors”, PrK= “{Basic, Pascal}”. The value KL = 6
comes off his current overlay model (Table 3.4, column ‘before’). ErrTyp
is “prone to logical errors” due to the fact that he had made usually errors that
84 3 A Novel Hybrid Student Model …

concern the semantics and operation of the commands. Also, PrK= = “{Basic,
Pascal}” indicates that Kostas knows the programming languages ‘Basic’ and
‘Pascal’, already.
He is examining in C6.5: “processing arrays per row” and is succeeding 86 %.
So, the quintet, which describes Kostas’ knowledge level on C6.5, is (0, 0, 0, 0.8,
0.2). However, according to the “strength of impact” of the knowledge dependen-
cies that exist between the domain concepts of the learning material (Table 2.2),
C6.5 affects 99 % the concept C6.4 and 77 % the concept C6.6. According to the
fuzzy rules (Figs. 3.6 and 3.7) the following occur (Table 3.4, column ‘after’):
• According to R5 and R6 (h) are μL(C6.4) = 0.79 and μA(C6.4) = 0.5. However,
due to limitation L1 is μL(C6.4) = 0.5. So, the quintet for C6.4 is (0, 0, 0, 0.5,
0.5).
• According to R1 and R2 (h) are μL(C6.6) = 0.62, μA(C6.4) = 0.17 and it
remains 11 % ‘Known’ (μKn(C6.6) = 0.11). So, the quintet for C6.6 is (0, 0,
0.11, 0.62, 0.17).
Consequently, Kostas remains to the same knowledge level (KL = 6), but the system
infers that he does not need to read the domain concepts C6.4.
Example 3 Stella’s current student model has the following values: KL = 3,
ErrTyp = “prone to logical errors”, PrK= “none”. The value KL = 3 comes off
her current overlay model (Table 3.5, column ‘before’). ErrTyp is “prone to logical
errors” due to the fact that she had made usually errors that concern the semantics
and operation of the commands. PrK= “none” indicates that Stella does not have
previous knowledge on computer programming.
She is examining in C4.2: “calculating sum in a ‘for’ loop” and is succeeding 95 %.
So, the quintet, which describes Stella’s knowledge level on C4.2, is (0, 0, 0, 0, 1).
However, according to the “strength of impact” of the knowledge dependencies that
exist between the domain concepts of the learning material (Table 2.2), C4.2 affects
45 % the concept C4.3, 81 % the concept C4.4, 100 % the concept C5.2, 45 % the con-
cept C5.3, and 39 % the concept C5.4.
According to the fuzzy rules (Figs. 3.6 and 3.7) the following occur (Table 3.5,
column ‘after’):
• According to R2 (j) is are μA(C4.3) = 0.45. However, the current value
of μA(C4.3) is 0.75. Therefore, according to R1 μA(C4.3) = 0.75 and
μL(C4.3) = 0.25 So, the quintet for C4.3 is (0, 0, 0, 0.25, 0.75).
• According to R2 (j) IS μA(C4.4) = 0.81 and it remains 19 % ‘Learned’
(μL(C4.4) = 0.19). So, the quintet for C4.4 is (0, 0, 0, 0.19, 0.81).
• According to R2 (j) is μA(C5.2) = 1. So, the quintet for C5.2 is (0, 0, 0, 0, 1).
• According to R2 (j) μA(C5.3) = 0.45 and it remains 35 % ‘Learned’
(μL(C5.3) = 0.35) So, the quintet for C5.3 is (0, 0, 0, 0.35, 0.45).
• According to R2 (j) μA(C5.4) = 0.39 and it remains 61 % ‘Learned’
(μUn(C5.4) = 0.61). So, the quintet for C5.4 is (0, 0, 0, 0.61, 0.39).
3.6 Examples of Operations 85

Table 3.5  Stella’s progress Domain concepts Learner’s knowledge

Before After
(KL = 3) (KL = 5)
C1.1 (0, 0, 0, 0.1, 0.9) (0, 0, 0, 0.1, 0.9)
C1.2 (0,0, 0, 0.32, 0.68) (0,0, 0, 0.32, 0.68)
C1.3 (0, 0, 0, 0.12, 0.88) (0, 0, 0, 0.12,
0.88)
C1.4 (0,0, 0,0.02, 0.98) (0,0, 0,0.02, 0.98)
C1.5 (0, 0, 0, 0.23, 0.77) (0, 0, 0, 0.23,
0.77)
C1.6 (0, 0, 0, 0.3, 0.7) (0, 0, 0, 0.3, 0.7)
C1.7 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C2.1 (0, 0, 0, 0.16, 0.84) (0, 0, 0, 0.16,
0.84)
C3.1 (0, 0, 0, 0.2, 0.8) (0, 0, 0, 0.2, 0.8)
C3.2 (0, 0, 0, 0.6, 0.4) (0, 0, 0, 0.6, 0.4)
C3.2.1 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C3.3 (0, 0, 0, 0.42, 0.58) (0, 0, 0, 0.42,
0.58)
C4.1 (0, 0, 0, 0.18, 82) (0, 0, 0, 0.18, 82)
C4.2 (0, 0, 0, 1, 0) (0, 0, 0, 0, 1)
C4.3 (0, 0, 0, 0.25, 0.75) (0, 0, 0, 0.25,
0.75)
C4.4 (0, 0, 0, 1, 0) (0, 0, 0, 0.19,
0.81)
C4.5 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C5.1 (0, 0, 0, 0.3, 0.7) (0, 0, 0, 0.3, 0.7)
C5.2 (0, 0, 0, 1, 0) (0, 0, 0, 0, 1)
C5.3 (0, 0, 0, 0.45, 0.35) (0, 0, 0, 0.35,
0.45)
C5.4 (0, 0, 0, 0.81, 0.19) (0, 0, 0, 0.61,
0.39)
C5.5 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C5.6 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C6.1 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.2 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.3 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.4 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.5 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.6 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.7 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C7.1 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
86 3 A Novel Hybrid Student Model …

Consequently, the system infers that Stella’s knowledge level has increased in more
concepts than C4.2. Thereby, it advises her overlay model and increases the value of
KL stereotype, classifying her to an upper knowledge level to read the following learn-
ing material. In particular, the value of KL becomes 5 declaring that the system consid-
ers that Stella does not need to read concepts that correspond to knowledge level 4.
Example 4 The parameters of Dimitris’ student model had the following values:
KL = 6, ErrTyp = “prone to syntax errors”, PrK= “Pascal”. After having com-
pleted a test which involved exercises on two-dimensional arrays, the system dis-
covered that he made more than 40 % errors on the assignment statement, and more
specifically he used the symbol := rather than the symbol =. The system checked
the values of the parameter PrK and was informed that he already knows the pro-
gramming language Pascal. Thus, it was assumed that Dimitris used the symbol
:= for assignment due to his previous knowledge on Pascal. So, the system’ reac-
tion was to stress the error, but it did not classify him to the knowledge level 1.
Example 5 Alexis’s current student model has the following values: KL = 4,
ErrTyp = “prone to logical errors”, PrK= “Pascal”. The value KL = 4 comes off
his current overlay model (Table 3.6, column ‘before’). ErrTyp is “prone to logical
errors” due to the fact that he had made usually errors that concern the seman-
tics and operation of the commands. PrK= “none” indicates that Alexis knows the
­programming language ‘Pascal’.
He is examining in C5.4: “calculating average in while ‘for’ loop” and is
­succeeding 76 %. So, the quintet, which describes his knowledge level on C5.4,
is (0, 0, 0.8, 0.2, 0). However, according to the “strength of impact” of the
­knowledge dependencies that exist between the domain concepts of the learning
material (Table 2.2), C5.4 affects the concept C4.2, C4.3, C4.4, C5.2 and C5.3.
According to the fuzzy rules (Figs. 3.6 and 3.7) the knowledge level of the
related concepts is deteriorated as follows (Table 3.6, column ‘after’):
• According to R7 C4.2 remains 100 % ‘Assimilated’.
• According to R8 is x4.3 = (1 − 0.41) × 88.6 + min[0.41 × 88.6, 0.41 × 76] =
83.46. Therefore, KL(C4.3) = L with μL(C4.3) = 1. So, the quintet for C4.3 is
(0, 0, 0, 1, 0).
• According to R8 is x4.4 = (1 − 0.52) × 89 + min[0.52 × 89, 0.52 × 76] = 84.8.
Therefore, KL(C4.4) = L with μL(C4.4) = 1. So, the quintet for C4.3 is (0, 0, 0, 1, 0).
• According to R7 C5.2 remains 100 % ‘Assimilated’.
• According to R8 is x5.3 = (1 − 0.41) × 88.6 + min[0.41 × 88.6, 0.41 × 76] =
83.44. Therefore, KL(C5.3) = L with μL(C5.3) = 1. So, the quintet for C4.3 is
(0, 0, 0, 1, 0).
Thereby, the system infers that Alexis has forgotten the concepts C4.3, C4.4, C5.2
and C5.3 and it classifies him to the previous corresponding knowledge level to
revise them, reducing the value of KL to 3.
Example 6 George, a novice student, interacted with the system for the first time.
He completed the test that involved exercises, which concerned the basic concepts
of “C”. The results are depicted in Fig. 3.11. Particularly, he does 29 % errors
3.6 Examples of Operations 87

Table 3.6  Alexis’ progress Domain Concepts Learner’s Knowledge

Before After
(KL = 4) (KL = 3)
C1.1 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C1.2 (0,0, 0, 0.4, 0.6) (0,0, 0, 0.4, 0.6)
C1.3 (0, 0, 0, 0. 2, 0.8) (0, 0, 0, 0. 2, 0.8)
C1.4 (0,0, 0,0.15, 0.85) (0,0, 0,0.15, 0.85)
C1.5 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C1.6 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C1.7 (0, 0, 0, 0.43, 0.57) (0, 0, 0, 0.43,
0.57)
C2.1 (0, 0, 0, 0.12, 0.88) (0, 0, 0, 0.12,
0.88)
C3.1 (0, 0, 0, 0.2, 0.8) (0, 0, 0, 0.2, 0.8)
C3.2 (0, 0, 0, 0.4, 0.6) (0, 0, 0, 0.4, 0.6)
C3.2.1 (0, 0, 0, 0.3, 0.7) (0, 0, 0, 0.3, 0.7)
C3.3 (0, 0, 0, 0.52, 0.48) (0, 0, 0, 0.52,
0.48)
C4.1 (0, 0, 0, 0.3, 0.7) (0, 0, 0, 0.3, 0.7)
C4.2 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C4.3 (0, 0, 0, 0.28, 0.72) (0, 0, 0, 1, 0)
C4.4 (0, 0, 0, 0.2, 0.8) (0, 0, 0, 1, 0)
C4.5 (0, 0, 0, 0.1, 0.9) (0, 0, 0, 0.1, 0.9)
C5.1 (0, 0, 0, 0.3, 0.7) (0, 0, 0, 0.3, 0.7)
C5.2 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C5.3 (0, 0, 0, 0.28, 0.72) (0, 0, 0, 1, 0)
C5.4 (0, 0, 0.4, 0.2, 0.4) (0, 0, 0.8, 0.2, 0)
C5.5 (0, 0, 0, 0.1, 0.9) (0, 0, 0, 0.1, 0.9)
C5.6 (0, 0, 0, 0.47, 0.53) (0, 0, 0, 0.47,
0.53)
C6.1 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.2 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.3 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.4 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.5 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.6 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.7 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C7.1 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)

Fig. 3.11  Test’s results of George

88 3 A Novel Hybrid Student Model …

on variables and constants, 33 % errors on the assignment statement and 50 %

errors on the arithmetic operators. Therefore, his current student model has the
­following values: KL = 1, ErrTyp = “prone to syntax errors”, PrK= “none”. The
system infers that George did not read the corresponding domains. So, he is not
­transmitted to the next KL stereotype, but he is advised to revise the concepts of
his ­current KL stereotype.

Table 3.7  Anna’s progress Domain concepts Learner’s knowledge

Before After
(KL = 2) (KL = 2)
C1.1 (0, 0, 0, 0, 1) (0, 0, 0, 0, 1)
C1.2 (0,0, 0, 0.2, 0.8) (0,0, 0, 0.2, 0.8)
C1.3 (0, 0, 0, 0.1, 0.9) (0, 0, 0, 0.1, 0.9)
C1.4 (0,0, 0, 0, 1) (0,0, 0, 0, 1)
C1.5 (0, 0, 0, 0.08, 0.82) (0, 0, 0, 0.08, 0.82)
C1.6 (0, 0, 0, 0.3, 0.7) (0, 0, 0, 0.3, 0.7)
C1.7 (0, 0, 0, 0.23, 77) (0, 0, 0, 0.23, 77)
C2.1 (0, 0, 0, 0.16, 0.84) (0, 0, 0, 0.16, 0.84)
C3.1 (0, 0, 0, 0.62, 0.38) (0, 0, 0.11, 0.89, 0)
C3.2 (0, 0, 0, 1, 0) (0, 0, 1, 0, 0)
C3.2.1 (0, 0, 0, 0.2, 0.8) (0, 0, 0, 0.2, 0.8)
C3.3 (0, 0, 0.3, 0.7, 0) (0, 0, 0.64, 0.36, 0)
C4.1 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C4.2 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C4.3 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C4.4 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C4.5 (0.63, 0, 0, 0.07, 0.3) (0.63, 0, 0, 0.07,
0.3)
C5.1 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C5.2 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C5.3 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C5.4 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C5.5 (0.63, 0, 0, 0.07, 0.3) (0.63, 0, 0, 0.07,
0.3)
C5.6 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.1 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.2 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.3 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.4 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.5 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.6 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C6.7 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
C7.1 (1, 0, 0, 0, 0) (1, 0, 0, 0, 0)
3.6 Examples of Operations 89

Example 7 Anna’s current student model has the following values: KL = 2,

ErrTyp = “prone to logical errors”, PrK= “Pascal”. The value KL = 2 comes off
her current overlay model (Table 3.7, column ‘before’). ErrTyp is “prone to logical
errors” due to the fact that she had made usually errors that concern the semantics
and operation of the commands. PrK= “none” indicates that Anna knows the pro-
gramming language ‘Pascal’.
She is examining in C3.2: “if…else if” and is succeeding 72 %. So, the quin-
tet, which describes Anna’s knowledge level on C3.2, is (0, 0, 1, 0, 0). However,
according to the “strength of impact” of the knowledge dependencies that exist
between the domain concepts of the learning material (Table 2.2), C3.2 affects
100 % the concept C3.1 (C3.1 affects 50 % C3.2) and 64 % the concept C3.3.
According to the fuzzy rules (Figs. 3.6 and 3.7) the following occur (Table 3.7,
column ‘after’):
• According to R8 is x3.1 = (1 − 0.5) × 86.9 + min[0.5 × 86.9, 0.5 × 72] =
79.45. Therefore, μKn(C3.1) = 0.11 and μL(C3.1) = 0.89. So, the quintet for C4.3
is (0, 0, 0.11, 0.89, 0).
• According to R4 (a) μKn(C5.4) = 0.64 and it remains 36 % ‘Learned’
(μL(C5.4) = 0.36). So, the quintet for C5.4 is (0, 0, 0.64, 0.36, 0).
Consequently, Anna’s knowledge level has been deteriorated. Therefore, the sys-
tem does not transit Anna to another knowledge level stereotype category (KL
remains 2). It consults her to revise the above domain concepts. Furthermore,
Anna made errors concerning the equality operator. In particular, she used the
symbol “=” rather that “==”. However, the system does not consult her to revise
the corresponding domain concept. It informs her only about the particular error.
This is happened, due to the fact the value of PrK of Anna’s student model is
‘Pascal’. Therefore, the system infers that Anna used the symbol “=” for equality
operator due to confusion with her previous knowledge on Pascal.

3.7 Conclusions

In this section, a novel hybrid student model was presented. The presented student
model combines an overlay model and stereotypes with fuzzy logic techniques. In
particular, the student model is based on an overlay model, which represents the
knowledge level of a learner. The determination of the student’s knowledge level
of each domain concept, as well as the updating of the student model are based
on the fuzzy logic technique that have been incorporated into the student model.
Fuzzy sets are used in order to describe how well each individual domain con-
cept is known, learned and assimilated. In addition, the student model includes a
mechanism of rules over the fuzzy sets, which is triggered after any change of the
value of the knowledge level of a domain concept and updates the values of the
knowledge level of all the related domain concepts with this.
90 3 A Novel Hybrid Student Model …

According to the learner’s knowledge level and errors, the system attached her/him
to the appropriate stereotype category, which concerns the student’s knowledge level.
The transition of a learner from one stereotype category of knowledge level to another
depicts the state of the learner. In other words, the transition of a learner from one
stereotype category of knowledge level to another reveals if s/he has learned or not
a domain concept, if s/he has forgot a concept or if s/he has assimilated it. In addi-
tion, the presented model includes two more stereotype categories. The one category
concerns the type of errors and helps the system to reason the poor performance of
the learner. The other stereotype category concerns the student’s prior knowledge on
other related knowledge domains and helps the system either to identify the learner’s
knowledge level or to infer if the student’s errors are caused by confusion with her/his
prior knowledge.
The presented novel student model approach has been fully implemented in a
web-based educational application, which teaches the programming language ‘C’.
The presented system provides adaptation of the instructional material, taking into
account the individuality of learners in terms of background, skills and pace of
learning. The particular student model includes a fuzzy-weighted qualitative over-
lay model; a rule-based fuzzy system; and a three-dimensional stereotype model.
The first dimension of the three-dimensional stereotype model consists of eight
stereotypes that represent the learner’s knowledge level; the second dimension
consists of two stereotypes that concern the type of programming errors (logical
or syntactic) and the third dimension concerns prior knowledge of the student on
other programming languages.
The particular student model allows each learner to complete the e-learning
course at their own pace, taking decisions about which concepts should be
delivered, which concepts need revision and which concepts are known and
­
do not need rereading. In this way, the system helps learners to save time and
effort d­ uring the learning process. The system identifies the alterations on the
state of ­students’ knowledge level, recognizes the misconceptions and needs of a
learner, and r­easons them. It tracks the cognitive state transitions of learners by
­constructing automatically state-chart diagrams. Thereby, the system recognizes if
a s­ tudent learns or not, if s/he reads or not, if s/he has difficulty in understanding,
if s/he forgets, if s/he has confused with other programming languages that s/he
has p­ reviously learned.
Chapter 4
Evaluation

Abstract In this chapter the evaluation of the hybrid student model F.O.S., which
incorporates fuzzy logic techniques, is presented. The presented evaluation pro-
cess is performed applying the evaluation framework PERSIVA (Chrysafiadi and
Virvou 2013a), which includes both questionnaires and observations through
experiments. In particular, either the educational impact (i.e. performance, satis-
faction, change of learners’ attitudes) or the effectiveness and validity adaptation
of the personalized and/or adaptive tutoring system are assessed. The evaluated
hybrid student model was integrated in a programming tutoring system for the
programming language ‘C’. Students of a postgraduate program in the field of
Informatics on the University of Piraeus, Greece, used the particular system to
learn how to program with the programming language ‘C’. The evaluation results
were very encouraging. They demonstrated that the presented student modeling
approach had a positive impact on the learners’ performance and on the learning
process. Furthermore, they showed that the system made valid and meaningful
adaptation decisions.

4.1 Introduction

The past decades the interest on web-based learning environments and tools has
been witnessed a rapid growth. The lack of time and place limitations and the
ability of web-based educational systems to offer instructions to large and hetero­
geneous groups of learners play a major role in the rapid growth of this interest.
However, web-based educational systems and computer tutors have to react and make
decisions like human tutors, in order to offer as effective learning as a real-life
classroom educational process. A solution to this is the technology of Intelligent
Tutoring Systems (ITSs), which are computer-based tutors that aim to provide the
same level of student specific help as a human tutor (Mitrovic et al. 2007), making
learning process more adaptive and student oriented (Jarusek and Pelánek 2012).
A core component in any intelligent or adaptive tutoring system that represents
many of the student’s features, such as knowledge and individual traits, is the

© Springer International Publishing Switzerland 2015 91

K. Chrysafiadi and M. Virvou, Advances in Personalized Web-Based Education,
Intelligent Systems Reference Library 78, DOI 10.1007/978-3-319-12895-5_4
92 4 Evaluation

student model (Brusilovsky and Millán 2007). It affects automated tutoring sys-
tems in making instructional decisions (Li et al. 2011), since a student model ena-
bles understanding and identification of students’ needs (Sucar and Noguez 2008).
Although, the adaptation generated by student modeling techniques often tend
to improve the interaction of the learner with the educational system, most of the
time the exploitation of such techniques makes the system more complex, less
predictable and buggier. As a consequence, it should be evaluated whether or not
the student model really improves the system (Gena 2005; Chin 2001). Therefore,
the evaluation of a student model is a crucial factor. Even though the evaluation
of adaptive systems is a difficult task due to the complexity of such systems, as
shown by many studies (Lavie et al. 2005; Markham et al. 2003; Del Missier and
Ricci 2003), several researchers have attempted to assess the student model of
their adaptive system. An assessment of the student model that SQL-Tutor uses is
presented in Mitrovic et al. (2002). Also, Weibelzahl and Weber (2003) performed
the evaluation of the accuracy of the student model of an adaptive learning system,
called the HTML-Tutor. A more recent attempt to assess the effectiveness and the
accuracy of the student model, which was applied in an ITS for learning software
design patterns, was done by Jeremić et al. (2009).
Although, there are many evaluation methods available in literature review,
as Mulwa et al. (2011) have mentioned, there is no a standard agreed measure-
ment framework for assessing the value and effectiveness of the adaptation yielded
by adaptive systems. The most common practice of evaluation is experiments.
However, there is not an accurate, clear and agreed framework in which an experi-
ment for the assessing of a student model should be performed. Furthermore, it
is important to not only evaluate but also to ensure that the evaluation uses the
correct methods, since an incorrect method can lead to wrong conclusions (Gena
and Weibelzahl 2007). Besides a well-designed evaluation framework underlines
the success of an approach and its potential value to others (Dempster 2004).
For this reason, the presented evaluation process is performed applying the
evaluation framework PERSIVA (Chrysafiadi and Virvou 2013a), which includes
both questionnaires and observations through experiments. Applying the particular
evaluation framework, either the educational impact (i.e. performance, satisfaction,
change of learners’ attitudes) or the adaptation of the personalized and/or adap-
tive tutoring system is assessed. The evaluation of the educational impact is based
on the Kirkpatrick’s model (1979). Furthermore, experiments play a major role in
the particular evaluation method, as they are appropriate for the evaluation of the
effectiveness and successfulness of user models (Chin 2001; Virvou and kabassi
2004).
The remainder of this section is organized as follows. Initially, the evaluation
framework PERSIVA is presented. Then, the evaluation criteria, process and popu-
lation are described. The presentation of the evaluation’s results follows. Finally,
the conclusions drawn from the evaluation process are presented.
4.2 The Evaluation Method 93

4.2 The Evaluation Method

4.2.1 The Evaluation Framework PERSIVA

The used evaluation framework PERSIVA (Chrysafiadi and Virvou 2013a) assesses:
• Students’ satisfaction about the e-learning program. More concretely, infor-
mation about the feelings, thoughts and satisfaction of the learners’ about
the adaptivity and effectiveness of the e-learning education environment was
gathered.
• Students’ performance on the knowledge domain.
• The changes that were caused on the individual state of the students.
• The results of the e-learning program to students’ progress.
• The validity of the conclusions drawn by the student model concerning the
aspects of the students’ characteristics.
• The validity of the adaptation decision making of the student model.

4.2.2 The Evaluation Criteria

The evaluation of an adaptive educational system is a complex process. It includes

evaluation of two different educational aspects. The first aspect concerns the effec-
tiveness of the educational program, and the other aspect concerns the effectiveness
of the student model. Therefore, the presented evaluation includes the following
two levels:
• Level 1: Evaluation of the educational impact. This level includes assess-
ment of the learners’ performance and satisfaction; evaluation of the effect of
the e-learning program on the behavior and thoughts of students about computer
programming and distance learning; evaluation of the effects of the particular
e-learning program to students’ progress on their further studies and assessment
of how it helped the students to learn other programming languages.
• Level 2: Evaluation of adaptation. This level is responsible to give answers in
the following questions:
– How satisfied are the learners about the system’s adaptive responses to their
needs?
– How important and essential is to model the particular student’s characteris-
tics (prior knowledge, knowledge level, type of errors etc.)?
– How valid are the conclusions drawn by the student model concerning the
aspects of the students’ characteristics (i.e. their background)?
– The decisions for adaptivity are valid and effective?
94 4 Evaluation

4.2.3 The Evaluation Process

The presented evaluation process includes experiment research and questionnaires.

Below, the particular evaluation processes that were used to assess each individual
evaluation criteria were described.
• Learners’ general satisfaction: For gathering this kind of information a ques-
tionnaire (Questionnaire A, Appendix B) was used. The questions were close-
ended based on Likert scale with five responses ranging from “Very much”
(5) to “Not at all” (1). The questions were divided into six categories based on
the type of information that were evaluated. The questions of the first category
are related to the quality of the content. The questions of the second category
concern the quality of instruction. The third and fourth categories concern the
friendliness and usefulness of the programming tutoring system correspond-
ingly. The questions of the fifth category are aimed to evaluate the adaptivity of
the system. The final question concerns the overall rating of the system.
• Learners’ performance: The learner’s performance is defined by either her/his
degree of success in tests and exercises or the times that s/he needed to read
a particular domain concept. Therefore, factor p was calculated to assess the
learners’ performance. In particular, the value of factor p is derived if the learn-
er’s degree of success in a particular domain concept is divided with the times
that s/he needs to read the particular concept (Eq. 4.1). The better the degree
of success is, the higher the value of p is. The lower the value of times that
the learner read the concept is, the higher the value of p is. Also, the lower the
degree of success is, the lower the value of p is. The higher the value of times
that the learner read the concept is, the lower the value of p is. Learners’ perfor-
mance was measured by conducting an experiment with an experimental group
(the group of students which used the presented programming tutoring system)
and a control group (the group of students which used a similar programming
tutoring system from which the presented student model was absent). Factor p
was calculated and compared for both groups.
degree of success in Ci
p= (4.1)
reading times of Ci
• Changes on learners’ behavior and thoughts about computer program-
ming: For gathering this kind of information a questionnaire (Questionnaire B,
Appendix B) was used. The questions were close-ended based on Likert scale
with five responses ranging from “Very much” (5) to “Not at all” (1).
• Changes on learners’ behavior and thoughts about e-learning: For gather-
ing this kind of information a questionnaire (Questionnaire C, Appendix B) was
used. The questions were close-ended based on Likert scale with five responses
ranging from “Very much” (5) to “Not at all” (1).
• Results on learners’ further studies: The assessment of that criterion is per-
formed with both questionnaire and experiment. In particular, a questionnaire
(Questionnaire D, Appendix B), which included five close-ended questions
4.2 The Evaluation Method 95

based on Likert scale with five responses ranging from “Very much” (5) to
“Not at all” (1), was used. The experiment included an experimental group (the
group of students which used the presented programming tutoring system) and
two control groups (students who had not used some programming tutoring sys-
tem). The average degrees that the students succeeded in two related with pro-
gramming lessons, which they taught after the use of the programming tutoring
system, were calculated and compared for each of the three groups.
• Learners’ satisfaction about the system’s adaptive responses to their
needs: For gathering this kind of information a questionnaire (Questionnaire E,
Appendix B) was used. The questions were close-ended based on Likert scale
with five responses ranging from “Very much” (5) to “Not at all” (1).
• The validity of the conclusions drawn by the student model concerning the
aspects of the students’ characteristics: It is performed assessing the results of
the system’s use in relation with the different backgrounds of the students. More
concretely, the students, who were used the presented programming tutoring
system, were divided into three categories according to their background knowl-
edge. These three categories are arts, science fields (other than computer
science) and computer science related fields. Learners, which have studies in
the field of human, social, political and educational sciences, belong to arts.
Learners, who have studies in the field of mathematic, physic, business and
economic sciences, belong to science fields. Learners, who have studies in
the field of computer, engineering and system sciences, belong to computer
science-related fields. It is obvious that students, who belong to the last cat-
egory, have a previous knowledge on computer programming. Thereby, their
progress should be better than the students of the other two categories. Also,
the particular learners should be advised to read a chapter fewer times than the
others. Furthermore, the learners, who belong to science fields, should have
a good learning pace, since their studies offer them a way of thinking that is
close to the programming logic. Whilst, the students of the arts’ category should
be advised to read more times the learning material, until to learn the knowl-
edge domain, since they often have no idea about computer programming.
Consequently, for each background category, the average reading times of each
domain concept and the percentage of learners that are advised to return to a
previous concept in order to revise it were measured. Furthermore, those values
were compared in order to assess if the conclusions drawn by the system con-
cerning the characteristics of student model are valid.
• The validity of the adaptation decision making of the student model: The
evaluation of the particular criterion includes assessment of the system’s adapta-
tion decisions about the return of a learner to a domain concept, which was con-
sidered as learner and/or assimilated, in order to revise it and about the inference
of the system that a learner should not read a particular domain concept at all.
The particular assessment is performed using a questionnaire and conducting an
experiment. In particular, a questionnaire (Questionnaire F, Appendix B) with
close-ended questions based on Likert scale with five responses ranging from
“Very much” (5) to “Not at all” (1), was used to ask the learners’ opinion about
96 4 Evaluation

the appropriateness or necessity of the returns to a domain concept, which was

considered as learner and/or assimilated, in order to revise it. In addition, learners’
performance in a final test was measured and compared for two groups (group A:
the group of students which used the presented programming tutoring system, and
group B: the group of students which used a similar programming tutoring system
with the same organized content, in which the learner chooses if s/he will return
to revise the concept that the system indicates her/him and/or if s/he will read or
not the suggested concept each time). Furthermore, the percentage of times that a
learner needed finally to read a domain concept that the system had advised her/
him not to read was measured.
The questionnaires A, E and F (Appendix B) were given to the students after their
participation in the training program. However, the learners were asked to com-
plete the questionnaires B, C and D (Appendix B), after, almost, 2 years of their
participation in the training program. The reason for that is the fact that the evalu-
ation of the changes on learners’ behavior and thoughts and of training program’s
results on learners’ further studies (which correspond to the evaluation levels of
behavior and results of the Kirkpatrick’s model) need at least a two-year evalua-
tion period (Jeremić et al. 2009). Furthermore, the Questionnaires B and C were,
also, given to participants before the use of the systems, in order to compare the
answers before and after their participation in the training program.

4.2.4 The Evaluation Population

For the experiment two groups of students were used. Learners of both groups
were students of a postgraduate conversion course program in the field of infor-
matics at the University of Piraeus. They had different ages, varying from 22 to
50, and backgrounds. Examples of such backgrounds are physics, mathematics,
computer science, education, human and social science. The number of students,
which belong to either each age category or background category, is the same for
both groups (Table 4.1). The reason for this is the fact that the homogeneity of the
experiment’s samples simplifies the experiment’s performing. Furthermore, 40 %
of the learners of the group A had a prior knowledge on computer programming.
The learners of group B that knew already a programming language were 45.74 %.
The distribution of students’ knowledge on other languages for both groups is
depicted in Table 4.2.

Table 4.1  Distribution of students’ ages and backgrounds

Ages 22–30 31–40 41–50
68.57 % 22.86 % 8.57 %
Background Arts Science (other than computers) Computer science related
34.29 % 28.57 % 37.14 %
4.2 The Evaluation Method 97

Table 4.2  Distribution of Language Java (%) Pascal (%) Basic (%)

students’ knowledge of other
Group A 21.43 28.57 50
programming languages
Group B 22.22 27.78 45

Both groups consisted of 35 students. The students of group A used the pre-
sented programming tutoring system for learning the programming language ‘C’.
The particular, programming tutoring system uses a fuzzy student model, which
helps the system to infer about the learner’s knowledge level on each domain con-
cept of the learning material and advise the learner not to read a concept at all or
return her/him to a previous learnt concept to revise it. The students of group B
used a similar educational system from which the fuzzy student model was absent.
Both systems had the same knowledge domain, which is divided into 31 concepts,
but the system, which was used by the students of group B, delivers the concepts
of the learning material in sequence without taking into account how the learner’s
performance on a domain concept may affect her/his knowledge of another con-
cept. The students, who used that system, had to read all the concepts one time at
least. Furthermore, according to the learner’s degree of success on tests, the sys-
tem decided if s/he had to return to revise a concept or if s/he had to be transited to
the next section of the learning material. The learners of both groups used the cor-
responding systems without attending any complementary course on programming,
over a period of 6 months.

4.3 Results

4.3.1 Learners’ General Satisfaction

Learners’ satisfaction about the educational system and program is positive. The
average learners’ overall rating of satisfaction is 4.38. The level of their satisfac-
tion about the quality of the content, the quality of instruction, the system’s use-
fulness, friendliness and adaptivity is high. The results of the corresponding
questionnaire (Questionnaire A, Appendix B) are depicted in Fig. 4.1. This infor-
mation is easy to collect, but does not tell enough about the learning success.

4.3.2 Learners’ Performance

One of the main goals of the system is to adapt dynamically the teaching sequence
to the users’ individual level of knowledge. In this sense the evaluation’s aim is
to evaluate the individualization of the teaching rather than evaluating the success
of the teaching method alone. In other words, it is evaluated how the student was
taught and whether s/he learned successfully rather than just whether s/he learned
98 4 Evaluation

4.5

4.4

4.3

4.2

4.1

3.9

3.8

3.7
quality of quality of friendliness usefulness adaptivity overall
conctent instruction rating

Fig. 4.1  Learners’ general satisfaction about the programming tutoring system

successfully. As such, a classical pre-test and post-test methodology could not be

sufficient for the aimed evaluation. The value of factor p, which is derived dividing
the learner’s degree of success in a particular domain concept with the times that
s/he needs to read the particular concept, was measured for group A and group B.
Then, the two average values were compared to assess the impact of the presented
fuzzy student model on the student’s performance and progress.
According to Grubišić et al. (2009) experiment used in the e-learning systems’
effectiveness evaluation change the independent variable (tutoring strategy) while
measuring the depended variable (effects on learning). In the presented work, the
experiment has one dependent variable and one independent variable with two val-
ues (two independent groups). The dependent variable is the factor p. The inde-
pendent variable is the groups (group A and group B). Due to the fact that the
experiment involves one dependent variable and one independent variable with
two levels, the two-independent sample t-test should be used to analyze the exper-
iment’s data. The particular statistical test is used to determine whether the dif-
ferent average scores of two groups, represents a real difference between the two
populations, or just a chance difference in our samples (Carver and Nash 2009;
Norusis 2009). The results of the experiment are depicted in Table 4.3.
However, the different averages scores can be occurred by chance or due to
differences on the education, knowledge level and abilities of the learners of the
two groups. The Levene’s test for equality of variances is used to ensure that the
above results were not occurred randomly or due to differences of students’ char-
acteristics of the two groups. According to the Levene’s test, if the value “Sig.”
is less than 0.05, then the two variances are significantly different, otherwise the
variability in two groups is about the same. The “Sig.” value of the experiment is
0.282 (Table 4.4). Therefore, the two variances are approximately equal. Then, the
value of “Sig. (2-tailed)” (Table 4.5) is checked in order to infer if the two means
4.3 Results 99

Table 4.3  Group statistics

Group N Mean Std. deviation Std. error
mean
p Group A 35 84.8734 16.35326 2.76421
Group B 35 59.3209 14.17618 2.39621

Table 4.4  Levene’s test for equality of variances

Levene’s test for equality t-test for equality of
of variances means
F Sig. t df
p Equal variances assumed 1.174 0.282 6.985 68
Equal variances not assumed 6.985 66.658

Table 4.5  Independent samples test

t-test for equality of means
Sig. Mean Std. error 95 % confidence
(2-tailed) difference difference interval of the
difference
Lower Upper
p Equal variances assumed 0.000 25.55257 3.65823 18.25268 32.85246

are statistically different. If this value is greater than 0.05, then there is no statis-
tically difference between the means and the difference is likely due to chance.
The experiment’s “Sig. (2-tailed)” value is 0. Thereby, the differences between
the means are statistically significant and are not a result of chance. Therefore, the
presented fuzzy programming tutoring system has a positive effect on the learner’s
performance and progress.

4.3.3 Changes on Learners’ Behavior and Thoughts

About Computer Programming

Learners’ behavior and thoughts towards the computer programming have

positively changed. The questionnaire B that was answered by the learners and the
mean of students’ answers are displayed in Appendix B. The results of the ques-
tionnaire are depicted in Figs. 4.2, 4.3, 4.4 and 4.5. The results show that the state
towards computer programming has been improved. The higher improvement is
observed for learners with background on arts. It is remarkable the increase of the
willingness of the particular learners to be engaged in computer programming pro-
jects. Furthermore, the number of learners, who believe that computer program-
ming can facilitate some everyday processes and are motivated to use it in their
jobs, has been increased significantly.
100 4 Evaluation

5
4.5
4
3.5
3
2.5 before
2 after
1.5
1
0.5
0
arts science (other than computer science
computers) related

Fig. 4.2  Change’s on learners’ behavior and thoughts about computer programming

4.3.4 Changes on Learners’ Behavior and Thoughts

About E-Learning

Learners’ behavior and thoughts about distance learning have positively changed. The
results of the questionnaire reveal that the students’ perception about e-learning has
improved. They seem more willing to be involved in e-learning programs. Furthermore,
their experience with the web-based programming tutoring system seems to have con-
vinced the learners about the effectiveness and usefulness of the e-learning. The results
of the learners’ answers to the questionnaire C (Appendix B) are depicted in Figs. 4.6,
4.7, 4.8 and 4.9.

4.3.5 Results on Learners’ Further Studies

The results of the e-learning program to the learners’ progress on their further
studies are satisfactory. The results of the questionnaire reveal that the e-learning
program helped the users. The results of the learners’ answers to the questionnaire
D (Appendix B) are depicted in Fig. 4.10.
However, a supplementary empirical research is essential for gathering this
kind of information. For this reason, an experiment was conducted. In the particular
experiment, the average degrees that students of three groups succeeded in two com-
puter programming related lessons were calculated. Both lessons were delivered
to students after the use of the programming tutoring system. The lesson 1 con-
cerned the computer programming language ‘Java’. The lesson 2 concerned the
digital signal processing of speech and audio. The latter lesson is not a lesson of
computer programming language, but it includes methodologies and techniques of
 4.3
Results

Fig. 4.3  Learners’ answers with arts background about computer programing

101
 102

Fig. 4.4  Learners’ answers with science (other than computers) background about computer programing
4
Evaluation
 4.3
Results

Fig. 4.5  Learners’ answers with computer-related science background about computer programing
103
104 4 Evaluation

4.5
4
3.5
3
2.5 before
2
1.5 after
1
0.5
0
arts science (other than computer-related
computers) science

Fig. 4.6  Changes on learners’ behavior and thoughts about e-learning

Do you think that e-learning systems

can facilitate the educational process?

Do you think that e-learning systems

can provide an effective education?

Do you interested in participating in an after

e-learning training program?
before
How useful do you think that is the
computer-based education?

What is you opinion about distance

learning?

0 1 2 3 4 5

Fig. 4.7  Learners’ answers with arts background about e-learning

algorithms and computer programming. Each of the three groups includes 35 stu-
dents of the department of informatics of the University of Piraeus. The students
of group A used the presented programming tutoring system before the teaching
of the above lessons, while the students of group B and group C did not used some
programming tutoring system. The students’ distribution according to their back-
grounds is the same for three groups. Particularly, they include 13 students with
background on arts, 10 students with background on sciences other than computers,
and 12 students with background on computer-related sciences. The results of the
experiment were compared in order to assess the effect of the presented web-based
programming tutoring system on learners’ further studies.
4.3 Results 105

Do you think that e-learning systems

can facilitate the educational process?

Do you think that e-learning systems

can provide an effective education?

Do you interested in participating in an after

e-learning training program?
before
How useful do you think that is the
computer-based education?

What is you opinion about distance

learning?

0 1 2 3 4 5

Fig. 4.8  Learners’ answers with science (other than computers) background about e-learning

Do you think that e-learning systems

can facilitate the educational process?

Do you think that e-learning systems

can provide an effective education?

Do you interested in participating in an after

e-learning training program?
before

How useful do you think that is the

computer-based education?

What is you opinion about distance

learning?

0 1 2 3 4 5

Fig. 4.9  Learners’ answers with computer-related science background about e-learning

The particular experiment has two dependent variables and one independent
variable with three levels. The dependent variables are the learners’ degrees in the
computer science lessons. The independent variable is the group of learners and
takes three values: group A, group B and group C. Due to the fact that the experi-
ment involves more than one dependent variables at a time, the multivariate analysis
of variance (MANOVA) should be used to analyze the experiment’s data. The results
of the experiment are depicted in Table 4.6. According to the experiment’s results,
106 4 Evaluation

Did the educational software help you in the

elaboration of tasks and activities
considering your studies?
Did the educational software help you to
understand better other lessons of computer
science?

Did the educational software help you in

your studies?

Did the educational software help you to

learn other programming languages?

Did the educational software help you to

understand better the logic of programming?

0 1 2 3 4 5

Fig. 4.10  Learners’ answers about the results of the e-learning program to their further studies

Table 4.6  Descriptive Group Mean Std. deviation N

statistics
Grade_on_ Group A 8.4286 1.00837 35
lesson1 Group B 7.3714 1.16533 35
Group C 7.2857 1.17752 35
Total 7.6952 1.22572 105
Grade_on_ Group A 7.5714 0.85011 35
lesson2 Group B 6.6571 0.96841 35
Group C 6.6857 0.96319 35
Total 6.9714 1.01391 105

the performance of the students, who used the presented web-based tutoring system
(group A), is better than the other students. Actually, their performance is better
even in lesson of ‘digital signal processing of speech and audio’, which does not
concern computer programming directly.
The homogeneity of covariances and the homogeneity of variances have
to be checked to ensure that the different averages scores were not occurred by
chance or due to differences on the group’s populations. MANOVA checks
the homogeneity of covariances conducting the “Box’s test of equality of
covariance”. If the “Sig.” value is less than 0.001, then the assumption of
homogeneity of covariances is violated. In the particular experiment, that value is
0.344 (Table 4.7). Therefore, there is homogeneity of covariances. Furthermore,
MANOVA checks the homogeneity of variances conducting the “Levene’s test
4.3 Results 107

Table 4.7  Box’s test of Box’s M 6.953

equality of covariance
F 1.126
matrices
df1 6
df2 259,299.692
Sig. 0.344

of equality of error variances”. If the “Sig.” value is higher than 0.05, then the
variability in experiment’s groups is about the same. In the particular experiment,
that value is 0.714 for ‘grade_on_lesson1’ and 0.854 for ‘grade_on_lesson2’
(Table 4.8). Consequently, both ‘grade_on_lesson1’ and ‘grade_on_lesson2’ have
homogeneity of variances.
Then, MANOVA conducts the ‘Wilks’ Lambda test’ to determine whether the
differences of the means are statistically significant. Lambda varies between 0 and
1. If the “Sig.” value of the test is less than 0.0005, then it means that the inde-
pendent variable has a significant effect on the dependent variable. The lambda
value of the independent variable ‘group’ is 0.760 and the corresponding “Sig.”
value is 0.000 (Table 4.9). Therefore, learners’ performance on lessons 1 and 2
were significantly dependent on the use of the programming tutoring system.
Then, tests of “Between-Subjects Effects” determine how the dependent variables
differ for the values of the independent variable and if the independent variable has
a significant effect on both or only on one dependent variable. From the results
that are depicted in Table 4.10, is concluded that the group (so, the use of the
programming tutoring system) has a significant effect on both ‘grade_on_les-
son1’ (Sig. = 0.000 < 0.0005) and ‘grade_on_lesson2’ (Sig. = 0.000 < 0.0005).

Table 4.8  Levene’s test of equality of error variances

F df1 df2 Sig.
Grade_on_lesson1 0.338 2 102 0.714
Grade_on_lesson2 0.158 2 102 0.854

Table 4.9  Multivariate tests

Effect Value F Hypothesis df Error df Sig.
Intercept Pillai’s trace 0.987 3,891.756a 2.000 101.000 0.000
Wilks’ lambda 0.013 3,891.756a 2.000 101.000 0.000
Hotelling’s trace 77.064 3,891.756a 2.000 101.000 0.000
Roy’s largest root 77.064 3,891.756a 2.000 101.000 0.000
Group Pillai’s trace 0.241 6.978 4.000 204.000 0.000
Wilks’ lambda 0.760 7.440a 4.000 202.000 0.000
Hotelling’s trace 0.316 7.896 4.000 200.000 0.000
Roy’s largest root 0.314 16.026b 2.000 102.000 0.000
aExact Statistic
bThe statistics is an upper bound on F that yields a lower bound on the significance level
108 4 Evaluation

Table 4.10  Tests of between-subjects effects

Source Dependent Sig. Partial eta Noncent. Observed
variable squared parameter power*
Group Grade_on_lesson1 0.000 0.182 22.621 0.991
Grade_on_lesson2 0.000 0.177 21.923 0.990
*Computed using alpha = .05

Table 4.11  Multiple comparisons

Dependent (I) (J) group Mean dif- Std. Sig. 95 % confidence interval
variable group ference error Lower Upper
(I–J) bound bound
Grade_on_ Group A Group B 1.0571* 0.26767 0.000 0.4205 1.6938
lesson1 Group C 10.1429* 0.26767 0.000 0.5062 1.7795
Group B Group A −1.0571* 0.26767 0.000 −1.6938 −0.4205
Group C 0.0857 0.26767 0.945 −0.5509 0.7223
Group C Group A −1.1429* 0.26767 0.000 −1.7795 −0.5062
Group B −0.0857 0.26767 0.945 −0.7223 0.5509
Grade_on_ Group A Group B 0.9143* 0.22204 0.000 0.3862 1.4424
lesson2 Group C 0.8857* 0.22204 0.000 0.3576 10.4138
Group B Group A −0.9143* 0.22204 0.000 −1.4424 −0.3862
Group C −0.0286 0.22204 0.991 −0.5567 0.4995
Group C Group A −0.8857* 0.22204 0.000 −1.4138 −0.3576
Group B 0.0286 0.22204 0.991 −0.4995 0.5567
*The mean difference is significant at the .05 level

The results of Table 4.11 show that both ‘grade_on_lesson1’ and ‘grade_on_les-
son2’ are statistical significantly different between either group A and group B
(Sig. < 0.0005) or group A and group C (Sig. < 0.0005). However, the dependent
variables are not statistical significantly different between group B and group C.
Therefore, the use of the presented web-based programming tutoring system has a
positive effect on the results of the further studies of the learners.

4.3.6 Learners’ Satisfaction About the System’s Adaptive

Responses to Their Needs

The learners’ satisfaction about the system’s adaptive responses to their need is
very satisfactory. The results of the questionnaire reveal that the programming
tutoring system considers the learners needs and reacts dynamically each time to
meet them. The results of the learners’ answers to the Questionnaire E (Appendix
B) are depicted in Fig. 4.11.
4.3 Results 109

Did the system’s inferences about your

knowledge level on each domain concept …

Was the system’s decision not to read some

concepts meaningful?

Were the returns to already learned concepts

meaningful?

Did the system return you to read again a

concept that that you knew it?

Did you need to read some concepts that the

system considered to be learned?

Did the prompt for revision was useful and

appropriate?

How time did you spend on issues that you

already known?

Did the educational system correspond to

your needs and knowledge level each time?

0 1 2 3 4 5

Fig. 4.11  Learners’ answers about the system’s adaptive responses to their needs

4.3.7 The Validity of the Conclusions Drawn by the Student

Model Concerning the Aspects of the Students’
Characteristics

The conclusions that are drawn by the system concerning the aspects of students’
characteristics seem to be satisfactory valid. According to the results the system
advises the learners with studies on arts fields to read a domain concept more
times than the learners who had been involved with the logic of programming
before (Fig. 4.12). The lower average times of reading corresponds to the learners
with background on computer-related sciences. Furthermore, the system advises
the learners with studies on arts to return to a domain concept in order to revise
it more times than the other learners (Fig. 4.13). This is evident, since learners
with no previous knowledge and experience on computer programming, have
difficulty in assimilating the learning material of the particular knowledge domain.
The average time of returns to a domain concept for revision is very low for the
learners with background on computer-related sciences. It is logical, since the
most learners with background on computer-related sciences have already been
involved in computer programming and thus it is easy for them to deal with the
learning material of the system’s knowledge domain.
110 4 Evaluation

1.6 1.45

1.4 1.17
1.2
0.91
1
0.8
0.6
0.4
0.2
0
arts sciences (other than computer-related
computers) sciences

Fig. 4.12  The average times of reading per background category

6.54
7
6
5
4
2.7
3
2
0.75
1
0
arts sciences (other than computer-related
computers) sciences

Fig. 4.13  The average times of returns to concepts for revision per background category

In addition, an experiment was conducted to ensure the validity of the conclu-

sions drawn by the student model concerning the aspects of the students’ charac-
teristics. In particular, the mean values of times of reading of a domain concept
of the learning material were measured for group A and group B. Furthermore,
the statistical significance of the difference of the mean values of reading times
between the learners of different backgrounds was checked. ANOVA is the statisti-
cal method that was chosen for the analysis of the experiment’s data. The different
mean values of reading times for both groups are presented in Table 4.12.
In the particular experiment, there is homogeneity of variances (the “Sig.”
value of “Levene’s test” is higher than 0.05—Table 4.13). Furthermore, the “Sig.”
value (Sig. = 0.000 < 0.0005) of the Table 4.14 declares that there is a signifi-
cant difference on the times of reading between the learners of different back-
grounds. Moreover, the comparisons of Table 4.15 show that for group A the
mean value of reading times is significantly different between the learners with
4.3 Results 111

Table 4.12  Descriptive statistics of reading times

Background Mean Std. deviation N
Times_of_ Group A Arts 1.4533 0.15744 12
reading Computer science related 0.9092 0.07805 13
Science (other than computers) 1.1760 0.12955 10
Total 1.1720 0.26259 35
Group B Arts 1.7325 0.20064 12
Computer science related 1.2400 0.15138 13
Science (other than computers) 1.4340 0.19202 10
Total 1.4643 0.27526 35

Table 4.13  Levene’s test of equality of error variances for reading times

F df1 df2 Sig.
2.223 5 64 0.063

Table 4.14  Tests of between-subjects effects for reading times

Source Type III sum of squares df Mean square F Sig.
Corrected model 4.869a 5 0.974 40.300 0.000
Intercept 121.273 1 121.273 5,018.842 0.000
Group 1.447 1 1.447 59.894 0.000
Background 3.357 2 1.679 69.473 0.000
Group * background 0.017 2 0.008 0.342 0.712
Error 1.546 64 0.024
Total 128.040 70
Corrected total 6.415 69

Table 4.15  Multiple Comparisons about reading times for group A

(I) background (J) Background Mean difference Std. Sig.
(I–J) error
Arts Computer science related 0.5441* 0.04988 0.000
Science (other than 0.2773* 0.05335 0.000
computers)
Computer science related Arts −0.5441* 0.04988 0.000
Science (other than −0.2668* 0.05241 0.000
computers)
Science (other than Arts −0.2773* 0.05335 0.000
computers) Computer science related 0.2668* 0.05241 0.000
*The mean difference is significant at the .05 level

background on arts and the learners with background on sciences other than com-
puters (Sig. < 0.0005); between the learners with background on sciences other
than computers and the learners with background on computer-related sciences
(Sig. < 0.0005); and the learners with background on arts and the learners with
112 4 Evaluation

Table 4.16  Multiple comparisons about reading times for group B

(I) Background (J) Background Mean difference Std. Sig.
(I–J) error
Arts Computer science related 0.4925* 0.07250 0.000
Science (other than 0.2985* 0.07755 0.002
computers)
Computer science related Arts −0.4925* 0.07250 0.000
Science (other than −0.1940* 0.07618 0.041
computers)
Science (other than Arts −0.2985* 0.07755 0.002
computers) Computer science related 0.1940* 0.07618 0.041
*The mean difference is significant at the .05 level

background on computer-related sciences (Sig. < 0.0005). On the other hand, the
results of the Table 4.16 show that for group B the mean value of reading times
is significantly different only between the learners with background on arts and
learners with background on computer-related sciences (Sig. < 0.0005). For group
B, the mean value of reading times does not differ significantly between neither
the learner with background on arts and learners with background on sciences
other than computers (Sig. = 0.002 > 0.0005) nor the learners with background on
sciences other than computers and learners with background on computer-related
sciences (Sig. = 0.041 > 0.0005). Therefore, the student model of the presented
programming tutoring system makes valid conclusions concerning the aspects of
the students’ characteristics.

4.3.8 The Validity of the Adaptation Decision Making

of the Student Model

The results of the validity of the system’s decision making were very encouraging.
It has been answered that the percentage of time that the learners spent to revise
a previous domain concept is at all waste of time. The results of the learners’
answers to the questionnaire F (Appendix B) are depicted in Fig. 4.14.
In addition, learners’ performance in a final test was measured and compared
for two groups (group A: the group of students which used the presented
programming tutoring system, and group B: the group of students which used a
similar programming tutoring system with the same organized content, in which
the learner chooses if s/he will return to revise the concept that the system
indicates her/him and/or if s/he will read or not the suggested concept each time).
The two-independent sample t-test was used to analyze the experiment’s data. The
results of the experiment are depicted in Table 4.17. The variability in two groups
is about the same (value “Sig.” = 0.106 > 0.05 Table 4.18). Therefore, the different
average scores did not occur by chance or due to differences on the education,
knowledge level and abilities of the learners of the two groups. Furthermore, the
differences between the mean values are statistically significant; since the value of
4.3 Results 113

Did the returns to a previous domain concept

for revision help you to learn computer
programming better?

Were the returns to already learned concepts

meaningful?

Did the returns to a previous domain concept

correspond to your need for revision?

Did the prompt for revision was useful and

appropriate?

How many times the returns to a previous

read concept for revision concerned
concepts that you actually knew?

Were the returns to a previous learned

concept for revision a waste of time?

0 1 2 3 4 5

Fig. 4.14  Learners’ answers about about the appropriateness or necessity of the returns to a
domain concept for revision

Table 4.17  Mean values of performance

Group N Mean Std. deviation Std. error
mean
Performance Group A 35 90.1463 3.78963 0.64056
Group B 35 82.7143 5.30023 0.89590

Table 4.18  Independent samples test for performance

Levene’s test t-test for equality of means
for equality of
variances
F Sig. t df Sig. Mean Std. error
(2-tailed) difference difference
Performance Equal 2.676 0.106 6.748 68 0.000 7.43200 1.10135
variances
assumed
Equal vari- 6.748 61.56 0.000 7.43200 1.10135
ances not
assumed
114 4 Evaluation

“Sig. (2-tailed)” (Table 4.18) is lower than 0.05. Consequently, the system’s
decisions about adaptive responses to the learner’s needs are valid, since learners
seem to assimilate the learning material and succeed a better performance finally.
In addition, the percentage of times that a learner needed, finally, to read a
domain concept that the system had advised her/him not to read it is 7.14 %. This
percentage is sufficiently satisfactory to be able to lead to the conclusion that the
decisions, which are made by the system based on the student model, are valid.

4.4 Conclusions

The system’s evaluation revealed that the combination of fuzzy sets with overlay
model and stereotypes contributes, significantly, to the adaptation of the learning
process to the learning pace of each individual learner. The results of the evalu-
ation demonstrated learning improvements and successful adaptation to students’
needs. In particular, the learners’ overall rating of the presented web-based pro-
gramming tutoring system is very high. The participant learners were very
satisfied with the quality of content and quality, with the tutoring system’s friend-
liness and usefulness, and with the system’s adaptive responses to their needs.
Furthermore, the integration of the presented novel fuzzy student model into the
programming tutoring system improved significantly the student’s performance.
Also, the learners, who used the presented e-learning tutoring system, obtained
a more positive state and behavior towards computer programming and distance
learning. The assessment results showed, also, that the e-learning program helped
learners to their further studies satisfactory. In addition, the evaluation results
revealed that the presented novel fuzzy approach of student modeling improves
the efficiency of the adaptation of the instructional process. In particular, both,
the conclusions that are drawn by the system concerning the aspects of students’
characteristics and the adaptation decision-making were valid. The system advises
the student model and adapts instantly the sequence of learning lessons concerning
the students’ characteristics with success. Consequently, the presented novel hybrid
student model contributes significantly to the adaptation process and helps the
system to provide a personalized and effective educational process for learning of
computer programming.
Conclusions and Discussion

Applicability of the Novel Fuzzy Student Modeling Approach

The target of this book was to present a survey of student modeling approaches and
fuzzy logic application and a novel approach that combines fuzzy logic techniques
for offering individualized instruction and personalized support in adaptive educa-
tional systems. The presented student modeling approach performs individualized
instruction, adapting the delivery of the knowledge domain to the individual learner’s
learning needs and pace. The presented approach models either how learning pro-
gresses or how the student’s knowledge can decrease. It automatically models the
learning or forgetting process of a student. In particular, the system keeps track of
cognitive state transitions of learners with respect to their progress or no-progress.
Thereby, it reveals if a student learns or not, if s/he forgets and reasons these states.
Therefore, the system allows each individual learner to complete the e-learning
course at her/his own pace, taking decisions about the concepts of the learning mate-
rial that have to be delivered to them, the concepts that need revision and the con-
cepts that are known and do not need further reading. In this way, the system helps
learners to save time and effort during the learning process.
The presented novel approach was fully implemented and evaluated. Particularly,
an original integrated environment for personalized e-training in programming and
the language ‘C’ was developed. This system was used by the students of a post-
graduate program in the field of Informatics in the University of Piraeus, Greece. The
evaluation was based on close-ended questionnaires and experimental research. The
results of the evaluation were very encouraging. They demonstrated that the system
models the student’s cognitive state and adapts dynamically to her/his individual needs
by scheduling the sequence of lessons on the fly, allowing her/him to complete the
e-training course at her/his own pace and according to her/his ability. The encourag-
ing results motivate the implementation of the presented novel approach of tutoring
system that combines fuzzy techniques to educational environments of knowledge
domains other than computer programming.

© Springer International Publishing Switzerland 2015 115

K. Chrysafiadi and M. Virvou, Advances in Personalized Web-Based Education,
Intelligent Systems Reference Library 78, DOI 10.1007/978-3-319-12895-5
116 Conclusions and Discussion

Contribution to Knowledge Domain Representation

The operation of the system is based on the knowledge domain representation

that is implemented through a Fuzzy Related-Concepts Network (FR-CN). This
kind of knowledge domain representation helps to manage to represent either the
order in which the domain concepts of the learning material have to be taught and
organized, or the knowledge dependencies that exist among the domain concepts.
This is significant because the knowledge level of a domain concept increases or
decreases due to changes on the knowledge level of a related domain concept.
The design of the learning material and the definition of the individual domain
concepts that it includes, are based on the knowledge and experience of domain
experts. Furthermore, the contribution of domain experts is significant for the defi-
nition of the knowledge dependencies that exist among the domain concepts of the
learning material and their “strength o impact” on each other.
The particular knowledge domain representation approach helps the system to
recognize either the domain concepts that are already partly or completely known for
a learner or the domain concepts that s/he has forgot, taking into account the learner’s
knowledge level of the related concepts of the learning material. As a consequence,
the presented knowledge domain representation approach contributes to the improve-
ment of the navigation support that an adaptive and/or personalized learning system
provides. Furthermore, the presented approach represents the knowledge domain in
a more realistic way. It constitutes a prototype for an adaptive and/or personalized
tutoring system for delivering the learning material to each individual learner dynamically,
taking into account her/his learning needs and different learning pace.

Contribution to Student Modeling

The target of this book was to show how fuzzy sets can be combined with other
student modeling techniques to promote adaptivity and personalization in educa-
tional applications. The evaluation of the novel approach, which was presented
in this book, revealed that the incorporation of fuzzy techniques into the student
model contributes significantly to the adaptation of the learning process to the learning
pace of each individual learner. The presented novel fuzzy student modeling approach
allows the system to identify the appropriate domain concepts that correspond to each
individual learner’s knowledge level and educational needs. Therefore, it improves
the efficiency of the adaptivity of the instructional process.
The presented fuzzy student modeling approach models automatically the
learning or forgetting process of a student. In particular, it helps the learners that
already know some concepts of the teaching material to save time and effort dur-
ing the learning process. Furthermore, the presented approach helps the system
to recognize which domain concepts of the learning material that the student has
already learned in previous interactions, s/he has forgotten and adapts the presen-
tation of material accordingly.
Conclusions and Discussion 117

The ability of the presented novel fuzzy student model to recognize the altera-
tions of the student’s learning states and dynamically adapt the presentation of the
learning material accordingly, renders the particular student modeling approach a
novel generic tool for adaptive learning. The gain from the presented approach is
significant as fuzzy logic can be used in combination with overlay and stereotype
models to provide adaptivity and personalization in other interactive systems in
addition to educational applications. The application of this innovative approach is
possible where the user’s changeable state and/or preferences are affected by the
existing dependencies among the system’s elements (like concepts, preferences,
events, choices).

Contribution to Programming Tutoring Systems

Programming tutoring systems teaches computer programming to learners providing

adaptivity. Mainly, these systems adapt the learning process dynamically to the stu-
dent’s knowledge level and needs. However, they do not consider how the learner’s
performance in a domain concept affects the learner’s knowledge level of other related
domain concepts of the learning material. Consequently, the gain of the presented
approach is the modeling of the learning or forgetting process of a student that gives
the ability to the system to adapt dynamically to each individual learner’s needs by
scheduling the sequence of lessons instantly.
The presented approach allows the tutoring system to recognize either the
learner’s knowledge level, or the alterations that occur in the learner’s knowledge
of a domain concept. Then, the system updates the student’s knowledge level of
the domain concepts of the learning material that are related with the concept that
the student has learnt or forgotten. Therefore, the presented novel approach allows
the programming tutoring system to model either the possible increase or decrease
of the learner’s knowledge. Furthermore, the presented approach introduces the
reasoning of errors that are related with other programming languages. In particu-
lar, each time the system checks if the learner’s errors were due to possible con-
fusion with features of another previously-known programming language. In this
way, the system allows each learner to complete the e-learning course at their own
pace, taking decisions about which concepts have to be delivered, which concepts
need revision and which concepts are known and do not need further reading.

Contribution to Fuzzy Logic

The gain from the novel fuzzy student model is that the learner’s knowledge level
is represented in a more realistic way. It considers the fact that the learner’s knowl-
edge level is a moving target and models automatically the learning and forgetting
process. In particular, the presented fuzzy student modeling approach helps the
118 Conclusions and Discussion

tutoring system to recognize when a new domain concept is completely unknown

to the learner, when it is fully learned or assimilated, or when it is partly known due
to the learner having previous related knowledge. This is achieved using fuzzy sets
for describing how well each individual domain concept is known or learned and
a set of fuzzy rules, which are responsible for the update of the overall learner’s
knowledge level after any change of her/his knowledge of a particular domain of
the learning material.
The application of this approach is not limited to adaptive instruction, but it can
also be used in other systems with changeable user states, such as e-shops, where
consumers’ preferences change over the time and affect one another. For example,
it can be used to reduce customer information overload by recommending products
that are likely to be of interest to them, considering their preferences and the depen-
dencies that exist between products’ choices (in accordance to e-learning, users’
preferences correspond to users’ knowledge level and products’ choices correspond
to the domain concepts). Therefore, the particular novel fuzzy approach constitutes
a novel generic fuzzy tool, which offers dynamic adaptation to users’ needs and
preferences of adaptive systems.
Appendix A: The Matrixes of the System’s
FR-CN

Table A.1  The ORDER matrix of the FR-CN of Fig. 2.11

(a)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
C1 C1.1 C1.2 C1.3 C1.4 C1.5 C1.6 C1.7 C2 C2.1 C3 C3.1 C3.2 C3.2.1 C3.3
1 C1 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1
2 C1.1 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1
3 C1.2 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1
4 C1.3 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1
5 C1.4 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1
6 C1.5 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1
7 C1.6 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1
8 C1.7 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1
9 C2 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1
10 C2.1 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1
11 C3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
12 C3.1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0
13 C3.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
14 C3.2.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
15 C3.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
16 C4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
17 C4.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
18 C4.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
19 C4.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
20 C4.4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
(continued)

© Springer International Publishing Switzerland 2015 119

K. Chrysafiadi and M. Virvou, Advances in Personalized Web-Based Education,
Intelligent Systems Reference Library 78, DOI 10.1007/978-3-319-12895-5
120 Appendix A: The Matrixes of the System’s FR-CN

Table A.1  (continued)
(a)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
21 C4.5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
22 C5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
23 C5.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
24 C5.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
25 C5.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
26 C5.4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
27 C5.5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
28 C5.6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
29 C6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
30 C6.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
31 C6.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
32 C6.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
33 C6.4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
34 C6.5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
35 C6.6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
36 C6.7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
37 C7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
38 C7.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

(b)
16 17 18 19 20 21 22 23 24 25 26 27 28
C4 C4.1 C4.2 C4.3 C4.4 C4.5 C5 C5.1 C5.2 C5.3 C5.4 C5.5 C5.6
1 C1 1 1 1 1 1 1 1 1 1 1 1 1 1
2 C1.1 1 1 1 1 1 1 1 1 1 1 1 1 1
3 C1.2 1 1 1 1 1 1 1 1 1 1 1 1 1
4 C1.3 1 1 1 1 1 1 1 1 1 1 1 1 1
5 C1.4 1 1 1 1 1 1 1 1 1 1 1 1 1
6 C1.5 1 1 1 1 1 1 1 1 1 1 1 1 1
7 C1.6 1 1 1 1 1 1 1 1 1 1 1 1 1
8 C1.7 1 1 1 1 1 1 1 1 1 1 1 1 1
9 C2 1 1 1 1 1 1 1 1 1 1 1 1 1
10 C2.1 1 1 1 1 1 1 1 1 1 1 1 1 1
11 C3 1 1 1 1 1 1 1 1 1 1 1 1 1
12 C3.1 1 1 1 1 1 1 1 1 1 1 1 1 1
13 C3.2 1 1 1 1 1 1 1 1 1 1 1 1 1
14 C3.2.1 1 1 1 1 1 1 1 1 1 1 1 1 1
15 C3.3 1 1 1 1 1 1 1 1 1 1 1 1 1
16 C4 0 0 0 0 0 0 1 1 1 1 1 1 1
17 C4.1 0 0 1 0 0 0 1 1 1 1 1 1 1
18 C4.2 0 0 0 0 1 0 1 1 1 1 1 1 1
19 C4.3 0 0 0 0 0 0 1 1 1 1 1 1 1
20 C4.4 0 0 0 0 0 0 1 1 1 1 1 1 1
(continued)
Appendix A: The Matrixes of the System’s FR-CN 121

Table A.1  (continued)
(b)
16 17 18 19 20 21 22 23 24 25 26 27 28
21 C4.5 0 0 0 0 0 0 1 1 1 1 1 1 1
22 C5 0 0 0 0 0 0 0 0 0 0 0 0 0
23 C5.1 0 0 0 0 0 0 0 0 0 0 0 0 0
24 C5.2 0 0 0 0 0 0 0 0 0 0 0 0 0
25 C5.3 0 0 0 0 0 0 0 0 0 0 0 0 0
26 C5.4 0 0 0 0 0 0 0 0 0 0 0 0 0
27 C5.5 0 0 0 0 0 0 0 0 0 0 0 0 0
28 C5.6 0 0 0 0 0 0 0 0 0 0 0 0 0
29 C6 0 0 0 0 0 0 0 0 0 0 0 0 0
30 C6.1 0 0 0 0 0 0 0 0 1 0 0 0 0
31 C6.2 0 0 0 0 0 0 0 0 0 0 1 0 0
32 C6.3 0 0 0 0 0 0 0 0 0 0 1 0 0
33 C6.4 0 0 0 0 0 0 0 0 0 0 0 0 0
34 C6.5 0 0 0 0 0 0 0 0 0 0 0 0 0
35 C6.6 0 0 0 0 0 0 0 0 0 0 0 0 0
36 C6.7 0 0 0 0 0 0 0 0 0 0 0 0 0
37 C7 0 0 0 0 0 0 0 0 0 0 0 0 0
38 C7.1 0 0 0 0 0 0 0 0 0 0 0 0 0

(c)
29 30 31 32 33 34 35 36 37 38
C6 C6.1 C6.2 C6.3 C6.4 C6.5 C6.6 C6.7 C7 C7.1
1 C1 1 1 1 1 1 1 1 1 1 1
2 C1.1 1 1 1 1 1 1 1 1 1 1
3 C1.2 1 1 1 1 1 1 1 1 1 1
4 C1.3 1 1 1 1 1 1 1 1 1 1
5 C1.4 1 1 1 1 1 1 1 1 1 1
6 C1.5 1 1 1 1 1 1 1 1 1 1
7 C1.6 1 1 1 1 1 1 1 1 1 1
8 C1.7 1 1 1 1 1 1 1 1 1 1
9 C2 1 1 1 1 1 1 1 1 1 1
10 C2.1 1 1 1 1 1 1 1 1 1 1
11 C3 1 1 1 1 1 1 1 1 1 1
12 C3.1 1 1 1 1 1 1 1 1 1 1
13 C3.2 1 1 1 1 1 1 1 1 1 1
14 C3.2.1 1 1 1 1 1 1 1 1 1 1
15 C3.3 1 1 1 1 1 1 1 1 1 1
16 C4 1 1 1 1 1 1 1 1 1 1
17 C4.1 1 1 1 1 1 1 1 1 1 1
18 C4.2 1 1 1 1 1 1 1 1 1 1
19 C4.3 1 1 1 1 1 1 1 1 1 1
(continued)
122 Appendix A: The Matrixes of the System’s FR-CN

Table A.1  (continued)
(c)
29 30 31 32 33 34 35 36 37 38
20 C4.4 1 1 1 1 1 1 1 1 1 1
21 C4.5 1 1 1 1 1 1 1 1 1 1
22 C5 1 1 1 1 1 1 1 1 1 1
23 C5.1 1 1 1 1 1 1 1 1 1 1
24 C5.2 1 1 1 1 1 1 1 1 1 1
25 C5.3 1 1 1 1 1 1 1 1 1 1
26 C5.4 1 1 1 1 1 1 1 1 1 1
27 C5.5 1 1 1 1 1 1 1 1 1 1
28 C5.6 1 1 1 1 1 1 1 1 1 1
29 C6 0 0 0 0 0 0 0 0 1 1
30 C6.1 0 0 0 0 0 1 0 0 1 1
31 C6.2 0 0 0 0 0 0 0 0 1 1
32 C6.3 0 0 0 0 0 0 0 0 1 1
33 C6.4 0 0 0 0 0 0 0 0 1 1
34 C6.5 0 0 0 0 0 0 1 1 1 1
35 C6.6 0 0 0 0 0 0 0 0 1 1
36 C6.7 0 0 0 0 0 0 0 0 1 1
37 C7 0 0 0 0 0 0 0 0 0 0
38 C7.1 0 0 0 0 0 0 0 0 0 0

Table A.2  The matrix PART of the FR-CN of Fig. 2.11

(a)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
C1 C1.1 C1.2 C1.3 C1.4 C1.5 C1.6 C1.7 C2 C2.1 C3 C3.1 C3.2 C3.2.1 C3.3
1 C1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 C1.1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
3 C1.2 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
4 C1.3 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
5 C1.4 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
6 C1.5 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
7 C1.6 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
8 C1.7 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
9 C2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
10 C2.1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
11 C3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
12 C3.1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
(continued)
Appendix A: The Matrixes of the System’s FR-CN 123

Table A.2  (continued)
(a)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
13 C3.2 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
14 C3.2.1 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0
15 C3.3 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
16 C4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
17 C4.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
18 C4.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
19 C4.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
20 C4.4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
21 C4.5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
22 C5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
23 C5.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
24 C5.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
25 C5.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
26 C5.4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
27 C5.5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
28 C5.6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
29 C6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
30 C6.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
31 C6.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
32 C6.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
33 C6.4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
34 C6.5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
35 C6.6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
36 C6.7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
37 C7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
38 C7.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

(b)
16 17 18 19 20 21 22 23 24 25 26 27 28
C4 C4.1 C4.2 C4.3 C4.4 C4.5 C5 C5.1 C5.2 C5.3 C5.4 C5.5 C5.6
1 C1 0 0 0 0 0 0 0 0 0 0 0 0 0
2 C1.1 0 0 0 0 0 0 0 0 0 0 0 0 0
3 C1.2 0 0 0 0 0 0 0 0 0 0 0 0 0
4 C1.3 0 0 0 0 0 0 0 0 0 0 0 0 0
5 C1.4 0 0 0 0 0 0 0 0 0 0 0 0 0
6 C1.5 0 0 0 0 0 0 0 0 0 0 0 0 0
7 C1.6 0 0 0 0 0 0 0 0 0 0 0 0 0
8 C1.7 0 0 0 0 0 0 0 0 0 0 0 0 0
9 C2 0 0 0 0 0 0 0 0 0 0 0 0 0
10 C2.1 0 0 0 0 0 0 0 0 0 0 0 0 0
11 C3 0 0 0 0 0 0 0 0 0 0 0 0 0
12 C3.1 0 0 0 0 0 0 0 0 0 0 0 0 0
(continued)
124 Appendix A: The Matrixes of the System’s FR-CN

Table A.2  (continued)
(b)
16 17 18 19 20 21 22 23 24 25 26 27 28
13 C3.2 0 0 0 0 0 0 0 0 0 0 0 0 0
14 C3.2.1 0 0 0 0 0 0 0 0 0 0 0 0 0
15 C3.3 0 0 0 0 0 0 0 0 0 0 0 0 0
16 C4 0 0 0 0 0 0 0 0 0 0 0 0 0
17 C4.1 1 0 0 0 0 0 0 0 0 0 0 0 0
18 C4.2 1 0 0 0 0 0 0 0 0 0 0 0 0
19 C4.3 1 0 0 0 0 0 0 0 0 0 0 0 0
20 C4.4 1 0 0 0 0 0 0 0 0 0 0 0 0
21 C4.5 1 0 0 0 0 0 0 0 0 0 0 0 0
22 C5 0 0 0 0 0 0 0 0 0 0 0 0 0
23 C5.1 0 0 0 0 0 0 1 0 0 0 0 0 0
24 C5.2 0 0 0 0 0 0 1 0 0 0 0 0 0
25 C5.3 0 0 0 0 0 0 1 0 0 0 0 0 0
26 C5.4 0 0 0 0 0 0 1 0 0 0 0 0 0
27 C5.5 0 0 0 0 0 0 1 0 0 0 0 0 0
28 C5.6 0 0 0 0 0 0 1 0 0 0 0 0 0
29 C6 0 0 0 0 0 0 0 0 0 0 0 0 0
30 C6.1 0 0 0 0 0 0 0 0 0 0 0 0 0
31 C6.2 0 0 0 0 0 0 0 0 0 0 0 0 0
32 C6.3 0 0 0 0 0 0 0 0 0 0 0 0 0
33 C6.4 0 0 0 0 0 0 0 0 0 0 0 0 0
34 C6.5 0 0 0 0 0 0 0 0 0 0 0 0 0
35 C6.6 0 0 0 0 0 0 0 0 0 0 0 0 0
36 C6.7 0 0 0 0 0 0 0 0 0 0 0 0 0
37 C7 0 0 0 0 0 0 0 0 0 0 0 0 0
38 C7.1 0 0 0 0 0 0 0 0 0 0 0 0 0

(c)
29 30 31 32 33 34 35 36 37 38
C6 C6.1 C6.2 C6.3 C6.4 C6.5 C6.6 C6.7 C7 C7.1
1 C1 0 0 0 0 0 0 0 0 0 0
2 C1.1 0 0 0 0 0 0 0 0 0 0
3 C1.2 0 0 0 0 0 0 0 0 0 0
4 C1.3 0 0 0 0 0 0 0 0 0 0
5 C1.4 0 0 0 0 0 0 0 0 0 0
6 C1.5 0 0 0 0 0 0 0 0 0 0
7 C1.6 0 0 0 0 0 0 0 0 0 0
8 C1.7 0 0 0 0 0 0 0 0 0 0
9 C2 0 0 0 0 0 0 0 0 0 0
10 C2.1 0 0 0 0 0 0 0 0 0 0
11 C3 0 0 0 0 0 0 0 0 0 0
(continued)
Appendix A: The Matrixes of the System’s FR-CN 125

Table A.2  (continued)
(c)
29 30 31 32 33 34 35 36 37 38
12 C3.1 0 0 0 0 0 0 0 0 0 0
13 C3.2 0 0 0 0 0 0 0 0 0 0
14 C3.2.1 0 0 0 0 0 0 0 0 0 0
15 C3.3 0 0 0 0 0 0 0 0 0 0
16 C4 0 0 0 0 0 0 0 0 0 0
17 C4.1 0 0 0 0 0 0 0 0 0 0
18 C4.2 0 0 0 0 0 0 0 0 0 0
19 C4.3 0 0 0 0 0 0 0 0 0 0
20 C4.4 0 0 0 0 0 0 0 0 0 0
21 C4.5 0 0 0 0 0 0 0 0 0 0
22 C5 0 0 0 0 0 0 0 0 0 0
23 C5.1 0 0 0 0 0 0 0 0 0 0
24 C5.2 0 0 0 0 0 0 0 0 0 0
25 C5.3 0 0 0 0 0 0 0 0 0 0
26 C5.4 0 0 0 0 0 0 0 0 0 0
27 C5.5 0 0 0 0 0 0 0 0 0 0
28 C5.6 0 0 0 0 0 0 0 0 0 0
29 C6 0 0 0 0 0 0 0 0 0 0
30 C6.1 1 0 0 0 0 0 0 0 0 0
31 C6.2 1 0 0 0 0 0 0 0 0 0
32 C6.3 1 0 0 0 0 0 0 0 0 0
33 C6.4 1 0 0 0 0 0 0 0 0 0
34 C6.5 1 0 0 0 0 0 0 0 0 0
35 C6.6 1 0 0 0 0 0 0 0 0 0
36 C6.7 1 0 0 0 0 0 0 0 0 0
37 C7 0 0 0 0 0 0 0 0 0 0
38 C7.1 0 0 0 0 0 0 0 0 1 0

Table A.3  The matrix IMPACT of the FR-CN of Fig. 2.11

(a)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
C1 C1.1 C1.2 C1.3 C1.4 C1.5 C1.6 C1.7 C2 C2.1 C3 C3.1 C3.2 C3.2.1 C3.3
1 C1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 C1.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
3 C1.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
(continued)
126 Appendix A: The Matrixes of the System’s FR-CN

Table A.3  (continued)
(a)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
4 C1.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
5 C1.4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
6 C1.5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
7 C1.6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
8 C1.7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
9 C2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
10 C2.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
11 C3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
12 C3.1 0 0 0 0 0 0 0 0 0 0 0 0 +0.5 0 +0.2
13 C3.2 0 0 0 0 0 0 0 0 0 0 0 −1 0 0 +0.64
14 C3.2.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
15 C3.3 0 0 0 0 0 0 0 0 0 0 0 −1 −1 0 0
16 C4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
17 C4.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
18 C4.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
19 C4.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
20 C4.4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
21 C4.5 0 0 0 0 0 0 0 0 0 0 0 0 0 −0.29 0
22 C5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
23 C5.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
24 C5.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
25 C5.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
26 C5.4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
27 C5.5 0 0 0 0 0 0 0 0 0 0 0 0 0 −0.29 0
28 C5.6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
29 C6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
30 C6.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
31 C6.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
32 C6.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
33 C6.4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
34 C6.5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
35 C6.6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
36 C6.7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
37 C7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
38 C7.1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

(b)
16 17 18 19 20 21 22 23 24 25 26 27 28
C4 C4.1 C4.2 C4.3 C4.4 C4.5 C5 C5.1 C5.2 C5.3 C5.4 C5.5 C5.6
1 C1 0 0 0 0 0 0 0 0 0 0 0 0 0
2 C1.1 0 0 0 0 0 0 0 0 0 0 0 0 0
3 C1.2 0 0 0 0 0 0 0 0 0 0 0 0 0
(continued)
Appendix A: The Matrixes of the System’s FR-CN 127

Table A.3  (continued)
(b)
16 17 18 19 20 21 22 23 24 25 26 27 28
4 C1.3 0 0 0 0 0 0 0 0 0 0 0 0 0
5 C1.4 0 0 0 0 0 0 0 0 0 0 0 0 0
6 C1.5 0 0 0 0 0 0 0 0 0 0 0 0 0
7 C1.6 0 0 0 0 0 0 0 0 0 0 0 0 0
8 C1.7 0 0 0 0 0 0 0 0 0 0 0 0 0
9 C2 0 0 0 0 0 0 0 0 0 0 0 0 0
10 C2.1 0 0 0 0 0 0 0 0 0 0 0 0 0
11 C3 0 0 0 0 0 0 0 0 0 0 0 0 0
12 C3.1 0 0 0 0 0 0 0 0 0 0 0 0 0
13 C3.2 0 0 0 0 0 0 0 0 0 0 0 0 0
14 C3.2.1 0 0 0 0 0 +0.37 0 0 0 0 0 +0.37 0
15 C3.3 0 0 0 0 0 0 0 0 0 0 0 0 0
16 C4 0 0 0 0 0 0 0 0 0 0 0 0 0
17 C4.1 0 0 0 0 0 0 0 0 0 0 0 0 0
18 C4.2 0 0 0 +0.45 +0.81 0 0 0 +1 +0.45 +0.39 0 0
19 C4.3 0 0 −0.42 0 +0.34 0 0 0 +0.42 +1 +0.41 0 0
20 C4.4 0 0 −1 −0.45 0 0 0 0 +1 +0.45 +0.52 0 0
21 C4.5 0 0 0 0 0 0 0 0 0 0 0 +1 0
22 C5 0 0 0 0 0 0 0 0 0 0 0 0 0
23 C5.1 0 0 0 0 0 0 0 0 0 0 0 0 0
24 C5.2 0 0 −1 −0.45 −0.81 0 0 0 0 +0.45 +0.81 0 0
25 C5.3 0 0 −0.42 −1 −0.34 0 0 0 −0.42 0 +0.41 0 0
26 C5.4 0 0 −1 −0.45 −1 0 0 0 −1 −1 0 0 0
27 C5.5 0 0 0 0 0 −1 0 0 0 0 0 0 0
28 C5.6 0 0 0 0 0 0 0 0 0 0 0 0 0
29 C6 0 0 0 0 0 0 0 0 0 0 0 0 0
30 C6.1 0 0 0 0 0 0 0 0 0 0 0 0 0
31 C6.2 0 0 0 0 0 0 0 0 0 0 0 0 0
32 C6.3 0 0 0 0 0 0 0 0 0 0 0 0 0
33 C6.4 0 0 0 0 0 0 0 0 0 0 0 0 0
34 C6.5 0 0 0 0 0 0 0 0 0 0 0 0 0
35 C6.6 0 0 0 0 0 0 0 0 0 0 0 0 0
36 C6.7 0 0 0 0 0 0 0 0 0 0 0 0 0
37 C7 0 0 0 0 0 0 0 0 0 0 0 0 0
38 C7.1 0 0 0 0 0 0 0 0 0 0 0 0 0

(c)
C6 C6.1 C6.2 C6.3 C6.4 C6.5 C6.6 C6.7 C7 C7.1
C1 0 0 0 0 0 0 0 0 0 0
C1.1 0 0 0 0 0 0 0 0 0 0
C1.2 0 0 0 0 0 0 0 0 0 0
(continued)
128 Appendix A: The Matrixes of the System’s FR-CN

Table A.3  (continued)
(c)
C6 C6.1 C6.2 C6.3 C6.4 C6.5 C6.6 C6.7 C7 C7.1
C1.3 0 0 0 0 0 0 0 0 0 0
C1.4 0 0 0 0 0 0 0 0 0 0
C1.5 0 0 0 0 0 0 0 0 0 0
C1.6 0 0 0 0 0 0 0 0 0 0
C1.7 0 0 0 0 0 0 0 0 0 0
C2 0 0 0 0 0 0 0 0 0 0
C2.1 0 0 0 0 0 0 0 0 0 0
C3 0 0 0 0 0 0 0 0 0 0
C3.1 0 0 0 0 0 0 0 0 0 0
C3.2 0 0 0 0 0 0 0 0 0 0
C3.2.1 0 0 0 0 0 0 0 0 0 0
C3.3 0 0 0 0 0 0 0 0 0 0
C4 0 0 0 0 0 0 0 0 0 0
C4.1 0 0 0 0 0 0 0 0 0 0
C4.2 0 0 0 0 0 0 0 0 0 0
C4.3 0 0 0 0 0 0 0 0 0 0
C4.4 0 0 0 0 0 0 0 0 0 0
C4.5 0 0 0 0 0 0 0 0 0 0
C5 0 0 0 0 0 0 0 0 0 0
C5.1 0 0 0 0 0 0 0 0 0 0
C5.2 0 0 0 0 0 0 0 0 0 0
C5.3 0 0 0 0 0 0 0 0 0 0
C5.4 0 0 0 0 0 0 0 0 0 0
C5.5 0 0 0 0 0 0 0 0 0 0
C5.6 0 0 0 0 0 0 0 0 0 0
C6 0 0 0 0 0 0 0 0 0 0
C6.1 0 0 0 0 +0.43 0 0 0 0 0
C6.2 0 0 0 0 0 0 0 0 0 0
C6.3 0 0 0 0 0 0 0 0 0 0
C6.4 0 −0.51 0 0 0 +0.33 +0.33 +0.27 0 0
C6.5 0 0 0 0 −0.99 0 +0.77 0 0 0
C6.6 0 0 0 0 −0.99 −0.77 0 0 0 0
C6.7 1 0 0 0 −0.78 0 0 0 0 0
C7 0 0 0 0 0 0 0 0 0 0
C7.1 0 0 0 0 0 0 0 0 1 0
Appendix B: Questionnaires

Questionnaire A

Table B.1  Questionnaire for measuring learners’ satisfaction

The questionnaire
Questions Answers (circle one for each question)
1 Was the content well 1 2 3 4 5
organized? (not at all) (slightly) (moderately) (very) (absolutely)
2 Did the presentation 1 2 3 4 5
of the content meet (not at all) (slightly) (moderately) (very) (absolutely)
your needs?
3 Was the content practical 1 2 3 4 5
and useful? (not at all) (slightly) (moderately) (very) (absolutely)
4 Were exercises and tests 1 2 3 4 5
useful? (not at all) (slightly) (moderately) (very) (absolutely)
5 Did the educational s­ ystem 1 2 3 4 5
keep your interest alive? (not at all) (slightly) (moderately) (very) (absolutely)
6 What is your opinion about 1 2 3 4 5
the quality of instruction? (very poor) (poor) (fair) (good) (excellent)
7 Did you feel that you had 1 2 3 4 5
assimilated all the ­subjects (not at all) (slightly) (moderately) (very) (absolutely)
that you are taught?
8 Was the user interface 1 2 3 4 5
friendly? (not at all) (slightly) (moderately) (very) (absolutely)
9 Did the educational system 1 2 3 4 5
meet your expectations? (not at all) (slightly) (moderately) (very) (absolutely)
(continued)

© Springer International Publishing Switzerland 2015 129

K. Chrysafiadi and M. Virvou, Advances in Personalized Web-Based Education,
Intelligent Systems Reference Library 78, DOI 10.1007/978-3-319-12895-5
130 Appendix B: Questionnaires

Table B.1  (continued)
The questionnaire
Questions Answers (circle one for each question)
10 Is the educational system 1 2 3 4 5
useful as an ­educational (not at all) (slightly) (moderately) (very) (absolutely)
tool for programming
languages?
11 Do you think that the 1 2 3 4 5
use of the particular (not at all) (slightly) (moderately) (very) (absolutely)
­educational system was
waste of time?
12 Did the educational system 1 2 3 4 5
correspond to your (not at all) (slightly) (moderately) (very) (absolutely)
needs and knowledge level
each time?
13 How time did you spend 1 2 3 4 5
on issues that you already (not at all) (slightly) (moderately) (very) (very much)
known?
14 Did the prompt for 1 2 3 4 5
revision was useful (not at all) (slightly) (moderately) (very) (absolutely)
and appropriate?
15 What is your 1 2 3 4 5
overall rating? (very poor) (poor) (fair) (good) (excellent)
The answers
Questions Evaluation degree
Quality of content 1 Was the content well organized? 4.44
2 Did the presentation of the c­ ontent 4.1
meet your needs?
3 Was the content practical and useful? 4.62
4 Were exercises and tests useful? 4.62
Quality of instruction 5 Did the educational system 4.1
keep your interest alive?
6 What is your opinion about the 4.18
­quality of instruction?
7 Did you feel that you had a­ ssimilated 4.44
all the subjects that you are taught?
Friendliness 8 Was the user interface friendly? 3.97
Usefulness 9 Did the educational system meet 4
your expectations?
10 Is the educational system useful as 4.24
an educational tool for programming
languages?
11 Do you think that the use of the 1
particular educational system was
waste of time?
(continued)
Appendix B: Questionnaires 131

Table B.1  (continued)
The answers
Questions Evaluation degree
Adaptivity 12 Did the educational system 4.15
correspond to your needs and
­knowledge level each time?
13 How time did you spend on issues 1.73
that you already known?
14 Did the prompt for revision 4.06
was ‑useful and appropriate?
Overall rating 15 What is your overall rating? 4.38

Questionnaire B

Table B.2  The questionnaire for measuring the changes on learners’ behavior and thoughts
about computer programming
The questionnaire
Questions Answers (circle one for each question)
1 What is you opinion 1 2 3 4 5
about ­computer (not at all (slightly (moderately (very (extremely
programming? interesting) interesting) interesting) interesting) interesting)
2 Do you want to 1 2 3 4 5
learn a (another) (not at all) (slightly) (moderately) (very) (absolutely)
programming
language?
3 Do you interested 1 2 3 4 5
in participating (not at all) (slightly) (moderately) (very) (absolutely)
in a computer-­
programming
project?
4 Are you ­motivated 1 2 3 4 5
to use ­computer (not at all) (slightly) (moderately) (very) (absolutely)
programming in
your job?
5 Do you think that 1 2 3 4 5
computer program- (strongly (disagree) (neither (agree) (strongly
ming can facilitate disagree) agree or agree)
some everyday disagree)
processes?
(continued)
132 Appendix B: Questionnaires

Table B.2  (continued)
The answers
Questions Evaluation degree
Before After
Arts Science Computer Arts Science Computer
(other than science related (other than science related
computers) computers)
1 What is you 2.7 3.6 4 3.92 4.2 4.42
opinion about
computer
programming?
2 Do you want 3.54 4.2 4.25 3.69 4.5 4.58
to learn a
(another)
programming
language?
3 Do you 2.23 3.3 4.1 3.46 4.1 4.5
interested in
participating
in a computer-
programming
project?
4 Are you 1.77 3.3 4.34 3.39 4.4 4.83
motivated to
use computer
programming
in your job?
5 Do you think 2.69 3.9 4.58 4.08 4.5 4.67
that computer
programming
can facilitate
some everyday
processes?

Questionnaire C

Table B.3  The questionnaire for measuring the changes on learners’ behavior and thoughts
about e-learning
The questionnaire
Questions Answers (circle one for each question)
1 What is you 1 2 3 4 5
­opinion about (not at all (slightly (moderately (very (extremely
­distance interesting) interesting) interesting) interest- interesting)
learning? ing)
2 How useful do 1 2 3 4 5
you think that is (not at all) (slightly) (moderately) (very) (absolutely)
the computer-
based education?
(continued)
Appendix B: Questionnaires 133

Table B.3  (continued)
The questionnaire
Questions Answers (circle one for each question)
3 Do you interested 1 2 3 4 5
in participating (not at all) (slightly) (moderately) (very) (absolutely)
in an e-­learning
training
program?
4 Do you think 1 2 3 4 5
that e-learning (strongly (disagree) (neither agree (agree) (strongly
systems can pro- disagree) or disagree) agree)
vide an effective
education?
5 Do you think 1 2 3 4 5
that e-learning (strongly (disagree) (neither agree (agree) (strongly
systems can disagree) or disagree) agree)
facilitate the
­educational
process?
The answers
Questions Evaluation degree
Before After
Arts Science (other Computer Arts Science Computer
than computers) science (other than science
related computers) related
1 What is you 3 3.4 3.75 3.8 3.8 4
­opinion about
­distance learning?
2 How useful do 2.85 3.4 3.67 3.77 3.6 4
you think that is
the computer-
based education?
3 Do you interested 3.39 3.7 4.08 3.85 4.1 4.42
in participating
in an e-learning
­training program?
4 Do you think 2.46 3.2 3.67 4.08 4.3 4.08
that e-learning
systems can pro-
vide an effective
education?
5 Do you think 2.46 3 3.92 4.23 4.1 4.25
that e-learning
­systems can
­facilitate the edu­
cational process?
134 Appendix B: Questionnaires

Questionnaire D

Table B.4  The questionnaire for measuring the system’s results on learners’ further studies
The questionnaire
Questions Answers (circle one for each question)
1 Did the educational software 1 2 3 4 5
help you to understand better (not at all) (slightly) (moderately) (very) (absolutely)
the logic of programming?
2 Did the educational soft- 1 2 3 4 5
ware help you to learn other (not at all) (slightly) (moderately) (very) (absolutely)
­programming languages?
3 Did the educational software 1 2 3 4 5
help you in your studies? (not at all) (slightly) (moderately) (very) (absolutely)
4 Did the educational software 1 2 3 4 5
help you to understand better (not at all) (slightly) (moderately) (very) (absolutely)
other lessons of computer
science?
5 Did the educational software 1 2 3 4 5
help you in the elaboration of (not at all) (slightly) (moderately) (very) (absolutely)
tasks and activities consider-
ing your studies?
The answers
Questions Evaluation degree
1 Did the educational software help you to understand better the logic 4.4
of programming?
2 Did the educational software help you to learn other programming 3.86
languages?
3 Did the educational software help you in your studies? 3.95
4 Did the educational software help you to understand better other 3.34
­lessons of computer science?
5 Did the educational software help you in the elaboration of tasks 3.71
and activities considering your studies?
Appendix B: Questionnaires 135

Questionnaire E

Table B.5  The questionnaire for measuring learners’ satisfaction about the system’s adaptive
responses to their needs
The questionnaire
Questions Answers (circle one for each question)
1 Did the educational system 1 2 3 4 5
correspond to your needs (not at all) (slightly) (moderately) (very) (absolutely)
and knowledge level each
time?
2 How time did you spend 1 2 3 4 5
on issues that you already (not at all) (slightly) (moderately) (very) (absolutely)
known?
3 Did the prompt for revision 1 2 3 4 5
was useful and appropriate? (not at all) (slightly) (moderately) (very) (absolutely)
4 Did you need to read some 1 2 3 4 5
concepts that the system (not at all) (slightly) (moderately) (very) (absolutely)
considered to be learned?
5 Did the system return you 1 2 3 4 5
to read again a concept that (not at all) (slightly) (moderately) (very) (absolutely)
you knew it?
6 Were the returns to 1 2 3 4 5
already learned concepts (not at all) (slightly) (moderately) (very) (absolutely)
meaningful?
7 Was the system’s decision 1 2 3 4 5
not to read some concepts (not at all) (slightly) (moderately) (very) (absolutely)
meaningful?
8 Did the system’s inferences 1 2 3 4 5
about your knowledge level (not at all) (slightly) (moderately) (very) (absolutely)
on each domain concept
correspond to your actual
needs and level?
The answers
Questions Evaluation degree
1 Did the educational system correspond to your needs and knowledge 4.15
level each time?
2 How time did you spend on issues that you already known? 1.73
3 Did the prompt for revision was useful and appropriate? 4.06
4 Did you need to read some concepts that the system considered 1.44
to be learned?
5 Did the system return you to read again a concept that you knew it? 1.94
6 Were the returns to already learned concepts meaningful? 4.03
7 Was the system’s decision not to read some concepts meaningful? 4.56
8 Did the system’s inferences about your knowledge level on each 4.08
domain concept correspond to your actual needs and level?
136 Appendix B: Questionnaires

Questionnaire F

Table B.6  The questionnaire for measuring the validity of the adaptation decision making
The questionnaire
Questions Answers (circle one for each question)
1 Were the returns to a previous 1 2 3 4 5
learned concept for revision a (not at all) (slightly) (moderately) (very) (absolutely)
waste of time?
2 How many times the returns 1 2 3 4 5
to a previous read concept for (not at all) (slightly) (moderately) (very) (absolutely)
revision concerned concepts
that you actually knew?
3 Did the prompt for revision 1 2 3 4 5
was useful and appropriate? (not at all) (slightly) (moderately) (very) (absolutely)
4 Did the returns to a previous 1 2 3 4 5
domain concept correspond to (not at all) (slightly) (moderately) (very) (absolutely)
your need for revision?
5 Were the returns to already 1 2 3 4 5
learned concepts meaningful? (not at all) (slightly) (moderately) (very) (absolutely)
6 Did the returns to a previous 1 2 3 4 5
domain concept for revision (not at all) (slightly) (moderately) (very) (absolutely)
help you to learn computer
programming better?
The answers
Questions Evaluation degree
1 Were the returns to a previous learned concept for revision a 1.73
waste of time?
2 How many times the returns to a previous read concept for revision 2.06
concerned concepts that you actually knew?
3 Did the prompt for revision was useful and appropriate? 4.06
4 Did the returns to a previous domain concept correspond to your 3.86
need for revision?
5 Were the returns to already learned concepts meaningful? 4.03
6 Did the returns to a previous domain concept for revision help 4
you to learn computer programming better?
Appendix C: Screenshots

• Registration and log-in (Figs. C.1 and C.2).

• Domain concepts (Figs. C.3, C.4 and C.5).
• Exercises and questions of tests (Figs. C.6, C.7, C.8 and C.9).
• Results of the tests (Figs. C.10 and C.11).

Fig. C.1  Log-in form

© Springer International Publishing Switzerland 2015 137

K. Chrysafiadi and M. Virvou, Advances in Personalized Web-Based Education,
Intelligent Systems Reference Library 78, DOI 10.1007/978-3-319-12895-5
138 Appendix C: Screenshots

Fig. C.2  Registration form

Fig. C.3  Domain concepts of knowledge stereotype 1

Appendix C: Screenshots 139

Fig. C.4  Successful completion of the learning process of all the learning material
140 Appendix C: Screenshots

Fig. C.5  Arithmetic operators

Fig. C.6  Fill in the gaps exercise

Appendix C: Screenshots 141

Fig. C.7  Right-wrong exercise

Fig. C.8  Multiple-choice exercise

142 Appendix C: Screenshots

Fig. C.9  Put the algorithm pieces in the right order exercise

Fig. C.10  Overall results-progress

Appendix C: Screenshots 143

Fig. C.11  Results of the test

References

Al-Hmouz, A., Shen, J., Yan, J., & Al-Hmouz, R. (2010). Enhanced learner model for adap-
tive mobile learning. In Proceedings of the 12th International Conference on Information
Integration and Web-based Applications and Services, Paris, France (pp. 783–786).
Al-Hmouz, A., Shen, J., Yan, J., & Al-Hmouz, R. (2011). Modeling mobile learning system using
ANFIS. In Proceedings of the 11th IEEE International Conference on Advanced Learning
Technologies (ICALT 2011), Athens, USA (pp. 32–36).
Albano, G. (2011). Knowledge, skills, competencies: A model for mathematics e-learning. In
Proceedings of the 18th International Conference on Telecommunications, Ayia Napa, Cyprus
(pp. 214–225).
Alepis, E., & Virvou, M. (2006). Emotional intelligence: constructing user stereotypes for affec-
tive bi-modal interaction. In Proc. of the 10th International Conference on Knowledge-Based
& Intelligent Informationa & Engineering Systems (KES2006), Bournemouth International
Conference Centre, United Kingdom (pp. 435–442).
Alepis, E., & Virvou, M. (2011). Automatic generation of emotions in tutoring agents for affec-
tive e-learning in medical education. Expert Systems with Applications, 38, 9840–9847.
Alepis, E., Virvou, M., & Kabassi., K. (2008). Mobile education: Towards affective bi-modal
interaction for adaptivity. In Proceedings of the 3rd International Conference on Digital
Information Management, ICDIM 2008, London (pp. 51–56).
Alfonseca, E., Carro, R.M., Martin, E., Ortigosa, A., & Paredes, P. (2006). The impact of learn-
ing styles on student grouping for collaborative learning: a case study. User Modeling and
User-Adapted Interaction, 16(3–4), 377–401.
Almohammadi, K., & Hagras, H. (2013). An adaptive fuzzy logic based system for improved
knowledge delivery within intelligent e-learning platforms. In Proceedings of the b2013
International Conference on Fuzzy Systems (pp. 1–8). India: Hyderabad International
Convention Center (HICC).
Alves, P., Amaral, L., & Pires, J. (2008). Case-based reasoning approach to adaptive web-based
educational systems. In Proceedings Eighth IEEE International Conference on Advanced
Learning Technologies (ICALT '08), Santander, Cantabria, Spain (pp. 260–261).
Arroyo, I., Meheranian, H., & Woolf, B. P. (2010). Effort-based tutoring: An empirical approach
to intelligent tutoring. In Proceedings of the 3rd International Conference on Educational
Data Mining (pp.1–10). Pittsburgh, PA, USA.
Azadeh, A., Ziaei, B., & Moghaddam, M. (2012). A hybrid fuzzy regression-fuzzy cognitive map
algorithm for forecasting and optimization of housing market fluctuations. Expert Systems
with Applications, 39(1), 298–315.
Badaracco, M., & Martinez, L. (2013). A fuzzy linguistic algorithm for adaptive test in intelligent
tutoring system based on competences. Expert Systems with Applications, 40(8), 3073–3086.

© Springer International Publishing Switzerland 2015 145

K. Chrysafiadi and M. Virvou, Advances in Personalized Web-Based Education,
Intelligent Systems Reference Library 78, DOI 10.1007/978-3-319-12895-5
146 References

Baghaei, N., Mitrovic, A., & Irwin, W. (2005). A constraint-based tutor for learning Object-
Oriented analysis and design using UML. Proceedings. International Conference on
Computers in Education 2005, Singapore (pp. 9–16).
Bai, S. M., & Chen, S. M. (2006a). Automatically constructing grade membership functions for
students’ evaluation for fuzzy grading systems. In Proceedings of the 2006 World Automation
Congress. Budapest, Hungary.
Bai, S. M., & Chen, S. M. (2006b). A new method for students’ learning achievement using
fuzzy membership functions. In Proceedings of the 11th Conference on Artificial Intelligence,
Kaohsiung, Taiwan, Republic of China.
Bai, S. M., & Chen, S. M. (2008). Evaluating students’ learning achievement using fuzzy mem-
bership functions and fuzzy rules. Expert Systems with Applications, 34, 399–410.
Baker, R. S. (2007). Modeling and understanding students’ off-task behavior in intelligent tutor-
ing systems. In Proceedings of the SIGCHI Conf. on Human Factors in Computing Systems,
San Jose, California, USA (pp. 1059–1068).
Baker, R. S., Corbett, A. T., Koedinger, K. R., & Wagner, A. Z. (2004). Off-task behavior in the
cognitive tutor classroom: When students ‘game the system. In Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems (CHI ’04), Vienna, Austria (pp. 383–390).
Baker, R. S., Goldstein, A. B., & Heffernan, N. T. (2010). Detecting the moment of learning.
In Proceedings of the ACM International Conference on Interactive Tabletops and Surfaces,
Saarbrücken, Germany (pp. 25–34).
Barak, M. (2010). Motivating self-regulated learning in technology education. International
Journal of Technology and Design Education, 20(4), 381–401.
Balakrishnan, A. (2011). On Modeling the affective effect on learning. In Proc. of the 5th inter-
national conference on Multi-Disciplinary Trends in Artificial Intelligence, Hyderabald, India
(pp. 225–235).
Baschera, G. M., & Gross, M. (2010). Poisson-based inference for perturbation models in adap-
tive spelling training. International Journal of Artificial Intelligence in Education, 20, 1–3.
Bishop, C. C., & Wheeler, D. (1994). The myers-briggs personality type and its relationship to
computer programming. Journal of Research on Computing in Education, 26, 358–371.
Biswas, R. (1995). An application of fuzzy sets in students’ evaluation. Fuzzy Sets and Systems,
74(2), 187–194.
Bloom, B. S. (1984). The 2 sigma problem: The search for methods of group instruction as effec-
tive as one-to-one tutoring. Educational Researcher, 13(6), 4–16.
Bontcheva, K., & Wilks, Y. (2005). Tailoring automatically generated hypertext. User Modeling
and User-Adapted Interaction, 15(1–2), 135–168.
Boticario, J. G., Santos, O. C., & van Rosmalen, P. (2005). Issues in developing standard-
based adaptive learning management systems. In Proceedings of the EADTU 2005 Working
Conference: Towards Lisbon 2010: Collaboration for Innovative Content in Lifelong Open
and Flexible Learning.
Brusilovsky, P., & Anderson, J. (1998). ACT-R electronic bookshelf: An adaptive system
for learning cognitive psychology on the Web. In Proceedings of the WebNet’98, World
Conference of the WWW, Internet, and Intranet, Orlando, Florida, USA (pp. 92–97).
Brusilovsky, P., & Millán, E. (2007). User models for adaptive hypermedia and adaptive educa-
tional systems. In P. Brusilovsky, A. Kobsa, & W. Neidl (Eds.), The Adaptive Web: Methods
and Strategies of Web Personalization (pp. 3–53).
Bunt, A., & Conati, C. (2003). Probabilistic student modelling to improve exploratory behaviour.
User Modeling and User-Adapted Interaction, 13(3), 269–309.
Burstein, M. H., & Collins, A. M. (1988). Modeling a theory of human plausible reasoning. In
T. O’Shea & V. Sgurev (Eds.), Artificial intelligence iii: methodology, systems, applications
(pp. 21–28). North Holland: Elsevier Science Publishers B.V.
Burstein, M. H., Collins, A., & Baker, M. (1991). Plausible generalisation: extending a model of
human plausible reasoning. Journal of the Learning Sciences, 3, 319–359.
Carmona, C., & Conejo, R. (2004). A learner model in a distributed environment. In Proceedings
of the 3rd International Conference on Adaptive Hypermedia and Adaptive Web-Based
Systems (AH’2004), Eindhoven, the Netherlands (pp. 353–359).
References 147

Carmona, C., Castillo, G., & Millán, E. (2008). Designing a dynamic bayesian network for mod-
eling students’ learning styles. In Proceedings of the 8th IEEE International Conference on
Advanced Learning Technologies (ICALT 2008), Santander, Cantabria, Spain (pp. 346–350).
Carver, R., & Nash, J. G. (2009). Doing data analysis with SPSS. United States: Cengage
Learning Inc.
Castillo, O., & Melin, P. (2008). Intelligent systems with interval type-2 fuzzy logic.
International Journal of Innovative Computing, Information and Control, 4(4), 771–783.
Castillo, G., Gama, J., & Breda, A. M. (2009). An adaptive predictive model for student model-
ing. In Advances in Web-Based Education: Personalized Learning Environments (pp. 70–92).
USA: Information Science Publishing (Chapter IV).
Cetintas, S., Si, L., Xin, Y. P., & Hord C. (2010). Automatic detection of off-task behaviors
in intelligent tutoring systems with machine learning techniques. IEEE Transactions on
Learning Technologies, 3(3), 228–236.
Chang, D. F., & Sun, C. M. (1993). Fuzzy assessment of learning performance of junior high
school students. In Proceedings of the 1993 first national symposium on fuzzy theory and
applications, Hsinchu, Taiwan, Republic of China (pp. 1–10).
Chen, S. M., & Lee, C. H. (1999). New methods for students’ evaluating using fuzzy sets. Fuzzy
Sets and Systems, 104(2), 209–218.
Cheung, R., Wan, C., & Cheng, C. (2010). An ontology-based framework for personalized adap-
tive learning. In Proceedings of the 9th International Conference on Web-based Learning
(ICWL 2010) (pp. 52–61). Shanghai, China.
Chieu, V. M., Luengo, V., Vadcard, L., & Tonetti, J. (2010). Student modeling in orthopedic sur-
gery training: Exploiting symbiosis between temporal bayesian networks and fine-grained
didactic analysis. Journal of Artificial Intelligence in Education, 20(3), 269–301.
Chin, D. N. (2001). Empirical evaluation of the user models and user-adapted systems. User
Modelling and User-Adapted Interaction, 11, 181–194.
Cho, M.-H., & Kin, B. J. (2013). Students’ self-regulation for interaction with others in online
learning environments. Internet and Higher Education, 17, 69–75.
Chrysafiadi, K., & Virvou, M. (2008). Personalized teaching of a programming language over the
web: Stereotypes and rule-based mechanisms. In Proceedings of the 8th Joint Conference on
Knowledge-Based Software Engineering, Piraeus, Greece (pp. 484–492).
Chrysafiadi, K., & Virvou, M. (2012). Evaluating the integration of fuzzy logic into the student
model of a web-based learning environment. Experts Systems with Applications, 39(18),
13127–13134.
Chrysafiadi, K., & Virvou, M. (2013a). PeRSIVA: An empirical evaluation method of a student
model of an intelligent e-learning environment for computer programming. Computers and
Education, 68, 322–333.
Chrysafiadi, K., & Virvou, M. (2013b). Student modeling approaches: A literature review for the
last decade. Expert Systems with Applications, 40(11), 4715–4729.
Chrysafiadi, K., & Virvou, M. (2013c). Dynamically personalized e-training in computer pro-
gramming and the language C. IEEE Transactions on Education, 56(4), 385–392.
Chrysafiadi, K., & Virvou, M. (2014). Fuzzy Logic for adaptive instruction in an e-learning envi-
ronment for computer programming. IEEE Transactions on Fuzzy Systems. d.o.i.:10.1109/
TFU22.2014.2310242.
Clancey, W. (1988). The role of qualitative models in instruction. In J. Self (eds.), Artificial
Intelligence and Human Learning, Chapman and Hall Computing.
Clemente, J., Ramírez, J., & de Antonio, A. (2011). A proposal for student modeling based on
ontologies and diagnosis rules. Expert Systems with Applications, 38(7), 8066–8078.
Codara, L. (1998). Le mappe cognitive. Roma: Carrocci Editore.
Collins, A., & Michalski, R. (1989). The logic of plausible reasoning: A core theory. Cognitive
science (Vol. 13, pp. 1–49). The Netherlands: Elsevier Science.
Conati, C. (2009). Intelligent tutoring systems: New challenges and directions. In Proceedings
of the 21st International Joint Conference on Artificial Intelligence (pp. 2–7). San Francisco,
CA, USA.
148 References

Conati, C., & Maclaren, H. (2009). Empirically building and evaluating a probabilistic model of
user affect. User Modeling and User-Adapted Interaction, 19(3), 267–303.
Conati, C., & Zhou, X. (2002). Modeling students’ emotions from cognitive appraisal in edu-
cational games. In Proceedings of the 6th International Conference on Intelligent Tutoring
Systems, Biarritz, France and San Sebastian, Spain (pp. 944–954).
Conati, C., Gertner, A., & Vanlehn, K. (2002). Using bayesian networks to manage uncertainty in
student modeling. User Modeling and User-Adapted Interaction, 12(4), 371–417.
Craiger, J. P., Goodman, D. F., Weiss, R. J., & Butler, A. (1996). Modeling organizational behav-
ior with fuzzy cognitive maps. Journal of Computational Intelligence and Organizations, 1,
120–123.
Crockett, K., Latham, A., McLean, D., & O’Shea, J. (2013). A fuzzy model for predicting
learning styles using behavioral cues in an conversational intelligent tutoring system. In
Proceedings of the b2013 International Conference on Fuzzy Systems (pp. 1–8). India:
Hyderabad International Convention Center (HICC).
de Raadt, M. (2007). A review of Australasian investigations into problem solving and the novice
programmer. Computer Science Education, 17(3), 201–213.
Del Missier, F., & Ricci, F. (2003). Understanding recommender systems: Experimental evalua-
tion challenges. In Proceedings of the Second Workshop on Empirical Evaluation of Adaptive
Systems, Pittsburgh (pp. 31–40).
Dempster, J. (2004). Evaluating e-learning developments: An overview. Retrieved July 23, 2008
from http://www.warwick.ac.uk/go/cap/resources/eguides.
Desmarais, M. C., & Baker, R. S. (2012). A review of recent advances in learner and skill modeling
in intelligent learning environments. User Modeling and User-Adapted Interaction, 22(1–2),
9–38.
Devedzic, V. (2006). Semantic web and education (Monograph). Berlin Heidelberg, New York:
Springer.
Dodds, P., & Fletcher, J. (2004). Opportunities for new ‘‘smart’’ learning environments enabled
by next-generation web capabilities. Journal of Educational Multimedia and Hypermedia,
13(4), 391–404.
Dolog, P., Henze, N., Nejdl, W., & Sintek, M. (2004). The personal reader: Personalizing and
enriching learning resources using semantic web technologies. In Proceedings of the 3rd
International Conference on Adaptive Hypermedia and Adaptive Web-Based Systems,
Eindhoven, Netherlands (pp. 85–94).
Drigas, A., Argyri, K., & Vrettaros, J. (2009). Decade review (1999–2009): Artificial intelligence
techniques in student modeling. In Proceedings of the 2nd World Summit on the Knowledge
Society (WSKS 2009), Chania, Crete, Greece (pp. 552–564).
Durrani, S., & Durrani, D. S. (2010). Intelligent tutoring systems and cognitive abilities. In
Proceedings of Graduate Colloquium on Computer Sciences (GCCS), Department of
Computer Science (p. 1). FAST-NU Lahore.
Echauz, J. R., & Vachtsevanos, G. J. (1995). Fuzzy grading system. IEEE Transactions on
Education, 38(2), 158–165.
Faraco, R. A., Rosatelli, M. C., & Gauthier, F. A. O. (2004). An approach of student modeling in a
learning companion system. In Proceedings of the IX IBERAMIA, Puebla, Mexico (pp. 891–900).
Felder, R. M., & Silverman, L. K. (1988). Learning and teaching styles. Engineering Education,
78(7), 674–681.
Felder, R. M., & Soloman, B. A. (2003). Learning styles and strategies. Retrieved June 28, 2012,
from http://www.ncsu.edu/felder-public/ILSdir/styles.htm.
Flavell, J. H. (1976) Metacognitive aspects of problem solving. In L. B. Resnick (Ed.), The
nature of intelligence (pp 231–236). Hillsdale: Erlbaum.
Gaudioso, E., Montero, M., & Hernandez-del-Olmo, F. (2012). Supporting teachers in adap-
tive educational system through predictive models: A proof of concept. Expert Systems with
Applications, 39(1), 621–625.
Gena, C. (2005). Methods and techniques for the evaluation of user-adaptive systems. The knowl-
edge Engineering Review, 20(1), 1–37.
References 149

Gena, C., & Weibelzahl, S. (2007). Usability engineering for the adaptive web. In P. Brusilovsky,
P. A. Kobsa, & W. Nejdl (Eds.), The adaptive web: Methods and strategies of web personal-
ization (pp. 720–762). New York: Springer.
Geng, X., Qin, S.,Chang, H., & Yang, Y. (2011). A hybrid knowledge representation for
the domain model of intelligent flight trainer. In Proceedings 2011 IEEE International
Conference on Cloud Computing and Intelligence Systems, Beijing, China (pp. 29–33).
Glaser, R., Lesgold, A., & Lajoie, S. (1987). Toward a cognitive theory for the measurement of
achievement. In R. Ronning, J. Glover, J. C. Conoley, & J. Witt (Eds.), The influence of cog-
nitive psychology on testing and measurement: The Buros-Nebraska symposium on measure-
ment and testing (pp. 41–85). Hillsdale, NJ: Erlbaum.
Glushkova, T. (2008). Adaptive model for user knowledge in the e-learning system. In
Proceedings of the International Conference on Computer Systems and Technologies
(CompSysTech’08).
Goel, G., Lallé, S., & Luengo, V. (2012). Fuzzy logic representation for student modelling
case study on geometry. In Proceedings of the 11th International Conference on Intelligent
Tutoring Systems, Chania, Greece (pp. 428–433).
Goguadze, G., Sosnovsky, S. A., Isotani, S., & McLaren, B. M. (2011a). Evaluating a bayesian
student model of decimal misconceptions. In Proceedings of the 4th International Conference
on Educational Data Mining, Eindhoven, the Netherlands (pp. 301–306).
Goguadze, G., Sosnovsky, S., Isotani, S., & McLaren, B. M. (2011b). Towards a bayesian student
model for detecting decimal misconceptions. In Proceedings 19th International Conference
on Computers in Education, Chiang Mai, Thailand (pp. 34–41).
Graesser, A. C., Conley, M. W., & Olney, A. M. (2012). Intelligent tutoring systems. In S.
Graham & K. Harris (Eds.), APA educational psychology handbook (Vol. 3, pp. 451–473),
Applications to learning and teaching Washington, DC: American Psychological Association.
Grigoriadou, M., Kornilakis, H., Papanikolaou, K. A., & Magoulas, G. D. (2002). Fuzzy infer-
ence for student diagnosis in adaptive educational hypermedia. In Proceedings of the 2nd
Hellenic Conference on AI: Methods and Applications of Artificial Intelligence, Thessaloniki,
Greece (pp. 191–202).
Grubišić, A., Stankov, S., Rosić, M., & Žitko, B. (2009). Controlled experiment replication in
evaluation of e-learning system’s educational influence. Computers and Education,53(3),
591–602.
Grubišić, A., Stankov, S., and Žitko, B. (2013). Stereotype Student Model for an Adaptive
e-Learning System. World Academy of Science, Engineering and Technology, 7, 16–23.
Hernández, Y., Sucar, L., & Arroyo-Figueroa, G. (2010). Evaluating an affective student
model for intelligent learning environments. In Proceedings of the 12th Ibero-American
Conference on Advances in Artificial Intelligence (IBERAMIA’10), Bahía Blanca, Argentina
(pp. 473–482).
Holland, J., Mitrovic, A., & Martin, B. (2009). J-Latte: A constraint-based tutor for java. In
Proceedings of the 17th International Conference on Computers in Education, Hong Kong
(pp. 142–146).
Homsi, M., Lutfi, R., Rosa, M. C., & Bakarat, G. (2008). Student modeling using NN-HMM
for EFL course. In Proceedings of the 3rd International Conference on Information and
Communication Technologies: From Theory to Applications (ICTTA 2008), Umayyad Palace,
Damascus, Syria (pp. 1–6).
Idris, N., Yusof, N., & Saad, P. (2009). Adaptive course sequencing for personalization of learn-
ing path using neural network. International Journal of Advances in Soft Computing and Its
Applications, 1(1), 49–61.
Inventado, P. S., Legaspi, R., Bui, T. D. & Suarez, M. (2010). Predicting student’s appraisal of
feedback in an its using previous affective states and continuous affect labels from EEG data.
In Proceedings of the 18th International Conference on Computers in Education, Putrajaya,
Malaysia (pp. 71–75).
Jameson, A. (1996). Numerical uncertainty management in user and student modeling: An over-
view of systems and issues. User Modeling and User-Adapted Interaction, 5(3–4), 193–251.
150 References

Jaques, N., Conati, C., Harley, J., & Azevedo, R. (2014). Predicting Affect from Gaze Data
During Interaction with an Intelligent Tutoring System. In Proc. of the 12th International on
Intelligent Tutoring Systems (ITS 2014), Honolulu, HI, USA, (pp 29–38).
Jarusek, P., & Pelánek, R. (2012). Modeling and predicting students problem solving times. In
Proceedings of the 38th International Conference on Current Trends in Theory and Practice
of Computer Science (SOFSEM 2012) (pp. 637–648).
Jeremić, Z., Jovanović, J., & Gasĕvić, D. (2009). Evaluating an intelligent tutoring system for
design patterns: The DEPTHS experience. Educational Technology and Society,12(2),
111–130.
Jeremić, Z., Jovanović, J., & Gasĕvić, D. (2012). Student modeling and assessment in intelligent
tutoring of software patterns. Expert Systems with Applications,39(1), 210–222.
Jia, B., Zhong, S., Zheng, T., & Liu, Z. (2010). The study and design of adaptive learning sys-
tem based on fuzzy set theory. In Z. A. D. Cheok, W. Müller, X. Zhang, & K. Wong (Eds.),
Transactions on edutainment IV (pp. 1–11). Berlin, Heidelberg: Springer.
Jili, C., Kebin, H., Feng, W., & Huixia, W. (2009). E-learning behavior analysis based on fuzzy
clustering. In Proceedings Third International Conference on Genetic and Evolutionary
Computing (WGEC'09) (pp. 863–866).
Johnson, W. L., Rickel, J. W., & Lester, J. C. (2000). Animated pedagogical agents: Face-to-face
interaction in interactive learning environment. International Journal of Artificial Intelligence
in Education, 11, 47–78.
Jonassen, D. H., & Grabowski, B. L. (1993). Handbook of individual differences, learning and
instruction. Hillsdale, NJ: Lawrence Erlbaum Associates.
Jurado, F., Santos, O. C., Redondo, M. A., Boticario, J. G., & Ortega, M. (2008). Providing
dynamic instructional adaptation in programming learning. In Proceedings of the 3rd interna-
tional workshop on Hybrid Artificial Intelligence Systems, Burgos, Spain (pp. 329–336).
Karnik, N. N., Mendel, J. M., & Liang, Q. (1999). Type-2 fuzzy logic systems. IEEE
Transactions on Fuzzy Systems, 7(6), 643–658.
Kass, R. (1991). Building a user model implicitly from a cooperative advisory dialog. User
Modeling and User-Adapted Interaction,1, 203–258.
Kassim, A. A., Kazi, S. A., & Ranganath, S. (2004). A Web-based intelligent learning environ-
ment for digital systems. International Journal of Engineering Education, 20(1), 13–23.
Katsionis, G., & Virvou, M. (2004). A cognitive theory for affective user modelling in a virtual
reality educational game. In Proceedings of the IEEE International Conference on Systems,
Man, and Cybernetics (pp. 1209–1213).
Kavčič, A. (2004a). Fuzzy student model in intermediactor platform. In Proceedings of the 26th
International Conference on Information Technology Interfaces, Croatia (pp. 297–302).
Kavčič, A. (2004b). Fuzzy user modeling for adaptation in educational hypermedia. IEEE
Transactions on Systems, Man and Cybernetics. Part C: Applications and Reviews, 34(4),
439–449.
Kay, J. (2000). Stereotypes, student models and scrutability. In Proceedings of the 5th
International Conference on Intelligent Tutoring Systems, Montréal, Canada (pp. 19–30).
Khamis, M. (2011). IDEAL: An intelligent distributed experience-based adaptive learning model.
Journal of Arts and Humanities, 20(1).
Kirkpatrick, D. L. (1979). Techniques for evaluating training programs. Training and
Development Journal, 33(6), 78–92.
Kofod-Petersen, A., Petersen, S. A., Bye, G. G., Kolås, L., & Staupe, A. (2008). Learning in an
ambient intelligent environment—towards modeling learners through stereotypes. Revue
d'Intelligence Artificielle, 22(5), 569–588.
Kosba, E., Dimitrova, V., & Boyle, R. (2003). Using fuzzy techniques to model students in web-
based learning environment. In Proceedings of the 7th International Conference on Knowledge-
Based Intelligent Information and Engineering Systems, United Kingdom (pp. 222–229).
Kosba, E., Dimitrova, V., & Boyle, R. (2005). Using student and group models to support teach-
ers in web-based distance education. In Proceedings of the International Conference on User
Modeling, Edinburgh (pp. 124–133).
References 151

Kumar, A. (2006a). Using enhanced concept map for student modeling in programming tutors.
In Proceedings of the 19th International Florida Artificial Intelligence Research Society
Conference, Melbourne Beach (pp. 527–532).
Kumar, A. (2006b). A scalable solution for adaptive problem sequencing and its evaluation. In
Proceedings of the 4th International Conference on Adaptive Hypermedia and Adaptive Web-
Based Systems (AH'2006), Dublin, Ireland (pp. 161–171).
Kyriacou, D. (2008). A scrutable user modelling infrastructure for enabling life-long user mod-
elling. In Proceedings of the 5th International Conference on Adaptive Hypermedia and
Adaptive Web-Based Systems, Hannover, Germany (pp. 421–425).
Latham, A., Crockett, K., & McLean, D. (2014). An adaptation algorithm for an intelligent natu-
ral language tutoring system. Computers and Education, 71, 97–110.
Lavie, T., Meyer, J., Beugler, K., & Coughlin, J. F. (2005). The evaluation of in-vehicle adap-
tive systems. In Proceedings of the 4th Workshop on the Evaluation of Adaptive Systems,
Edinburgh, UK (pp. 9–18).
Law, C. K. (1996). Using fuzzy numbers in education grading system. Fuzzy Sets and Systems,
83(3), 311–323.
Le, N. T., & Menzel, W. (2009). Using weighted constraints to diagnose errors in logic program-
ming-the case of an Ill-defined domain. Journal on Artificial Intelligence in Education, 19(2),
382–400.
Lehman, B., Matthews. M., D’Mello, S., & Person, N. (2008). What are you feeling? Investigating
student affective states during expert human tutoring sessions. In Proceedings of the 9th
International Conference on Intelligent Tutoring Systems (ITS 2008), Montreal (pp. 50–59).
Leon, M., Napoles, G., Garcia, M., Bello, R., & Vanhoof, K. (2011). Two steps Individuals travel
behavior through fuzzy cognitive maps pre-definition and learning. In Proceedings of the 10th
Mexican International Conference on Artificial Intelligence. Puebla, Mexico.
Li, N., Cohen, W. W., Koedinger, K. R., & Matsuda, N. (2011). A machine learning approach
for automatic student model discovery. In Proceedings of Conference on Educational Data
Mining (EDM 2011), Eindhoven, the Netherlands (pp. 31–40).
Liang, Q., & Mendel, J. M. (2000). Interval type-2 fuzzy logic systems: Theory and design. IEEE
Transactions on Fuzzy Systems, 8(5), 535–550.
Liaw, S.-S., & Huang, H.-M. (2013). Perceived satisfaction, perceived usefulness and interactive
learning environments as predictors to self-regulation in e-learning environments. Computers
and Education, 60, 14–24.
Limongelli, C., Sciarrone, F., Temperini, M., & Vaste, G. (2009). Adaptive learning with the LS-
plan system: A field evaluation. IEEE Transactions on Learning Technologies, 2(3), 203–215.
Lin, C. M. (2007). Combination study of fuzzy cognitive map. International Journal of Energy
and Environment, 1(2), 65–69.
Liu, C. L. (2008). Using bayesian networks for student modeling. In R. M. Viccari, P. Augustin-
Jaques, & R. Verdin (Eds.), Agent-based tutoring systems by cognitive and affective modeling
(pp. 97–113). Hershey: Igi Global.
Liu, Z., & Wang, H. (2007). A modeling method based on bayesian networks in intelligent tutor-
ing system. In Proceedings of the 11th International Conference on Computer Supported
Cooperative Work in Design, Melbourne, Australia (pp. 967–972).
Lo, J.-J., Chan, Y.-C., & Yen, S. W. (2012). Designing an adaptive web-based learning system
based on students’ cognitive styles identified online. Computers and Education, 58, 209–222.
Lu, C. H., Ong, C. S., & Hsu, W. L. (2007). Using an ITS as an arithmetic assistant for teachers
3-year review. Journal of Internet Technology, 8(4), 389–398.
Lu, C. H., Wu, C. W., Wu, S. H., Chiou, G. F., & Hsu, W. L. (2005). Ontological support in mod-
eling learners' problem solving process. Educational Technology and Society, 8(4), 64–74.
Ma, J., & Zhou, D. (2000). Fuzzy set approach to the assessment of student-centered learning.
IEEE Transactions on Education, 43(2), 237–241.
Mahnane, L., Laskri, M. T., & Trigano, P. (2012). An adaptive hypermedia system integrat-
ing thinking style (AHS-TS): Model and experiment. International Journal of Hybrid
Information Technology, 5(1), 11–28.
152 References

Markham, S., Ceddia, J., Sheard, J., Burvill, C., Weir, J., Field, B. et al. (2003). Applying
agent technology to evaluation tasks in e-learning environments. In Exploring Educational
Technologies Conference (pp. 16–17).
Martin, B. (1999). Constraint-based student modeling: Representing student knowledge. In
Proceedings of the 3rd New Zealand Computer Science Research Students’ Conference,
Hamilton NZ (pp. 22–29).
Martins, A. C., Faria, L., Vaz de Carvalho, C., & Carrapatoso, E. (2008). User modeling in adap-
tive hypermedia educational systems. Educational Technology and Society, 11(1), 194–207.
Mayo, M. J. (2001). Bayesian student modelling and decision-theoretic selection of tuto-
rial actions in intelligent tutoring systems. Ph.D. Thesis. Retrieved June 29, 2012, from
http://www.cosc.canterbury.ac.nz/research/reports/PhdTheses/2001/phd_0102.pdf.
McGill, T. J., & Volet, S. E. (1997). A conceptual framework for analyzing students’ knowledge
of programming. Journal of Research on Computing in Education, 29(3), 276–297.
Mendel, J. M. (2001). Uncertain rule-based fuzzy logic systems: Introduction and new directions.
Upper Saddle River, NJ: Prentice-Hall.
Mendel, J. M. (2007). Advances in type-2 fuzzy sets and systems. Information Sciences, 177(1),
84–110.
Miao, Y., & Liu, Z. Q. (2000). On causal inference in fuzzy cognitive maps. IEEE Transactions
on Fuzzy Systems, 8(1), 107–119.
Michaud, L. N., & McCoy, K. F. (2004). Empirical derivation of a sequence of user stereotypes
for language learning. User Modeling and User-Adapted Interaction, 14, 317–350.
Millán, E., & Perez de la Cruz, J.-L. (2002). A bayesian diagnostic algorithm for student model-
ing. User Modeling and User-Adapted Interaction, 12(2/3), 281–330.
Millán, E., Loboda, T., & Pérez-de-la-Cruz, J. L. (2010). Bayesian networks for student model
engineering. Computers and Education, 55(4), 1663–1683.
Mitrovic, A. (2003). An intelligent SQL tutor on the web. International Journal of Artificial
Intelligence in Education, 13(2–4), 173–197.
Mitrovic, A., and Martin, B. (2006). Evaluating the effects of open student models on learning. In
Proceedings of the 2nd International Conference on Adaptive Hypermedia and Adaptive Web-
Based Systems (pp. 296–305).
Mitrovic, A., Marting, B., & Mayo, M. (2002). Using Evaluation to Shape ITS Design: Results
and Experiences with SQL-Tutor. User Modelling and User-Adapted Interaction, 12(2/3),
243–279.
Mitrovic, A., Martin, B., & Suraweera, P. (2007). Intelligent tutors for all: Constraint-based
approach. IEEE Intelligent Systems, 22(4), 38–45.
Mitrovic, A., Mayo, M., Suraweera, P., & Martin, B. (2001). Constraint-based tutors: A suc-
cess story. In Proceedings of 14th International Conference on Industrial and Engineering
Applications of Artificial Intelligence and Expert Systems (IEA/AIE-2001), Budapest
(pp. 931–940).
Mizumoto, M., & Tanaka, K. (1976). Some properties of fuzzy sets of type 2. Information and
Control, 31(4), 312–340.
Moridis, C. N., & Economides, A. A. (2009). Prediction of student’s mood during an online test using
formula-based and neural network-based method. Computers and Education, 53, 644–652.
Mulwa, C., Lawless, S., Sharp, M., & Wade, V. (2011). The evaluation of adaptive and personal-
ized information retrieval systems: A review. International Journal of Knowledge and Web
Intelligence, 2(2/3), 138–156.
Muñoz, K., Mc Kevitt, P., Lunney, T., Noguez, J., & Neri, L. (2010). PlayPhysics: An emotional
game learning environment for teaching physics. In Proceedings of the 4th International
Conference on Knowledge Science, Engineering and Management (KSEM' 10), Belfast,
Northern Ireland, UK (pp. 400–411).
Muñoz, K., Mc Kevitt, P., Lunney, T., Noguez, J., & Neri, L. (2011). An emotional student model
for game-play adaptation. Entertainment Computing, 2(2), 133–141.
Nguyen, L., & Do, P. (2008). Learner model in adaptive learning. In Proceedings of World
Academy of Scıence, Engıneerıng and Technology (pp. 396–401).
References 153

Nguyen, L., & Do, P. (2009). Combination of bayesian network and overlay model in user mod-
eling. In Proceedings of the 9th International Conference on Computational Science, Baton
Rouge, Louisiana, USA (pp. 5–14).
Nguyen, C. D., Vo, K. D., Bui, D. B., & Nguyen, D. T. (2011). An ontology-based IT student
model in an educational social network. In Proceedings of the 13th International Conference
on Information Integration and Web-based Applications and Services (iiWAS '11), Bali,
Indonesia (pp. 379–382).
Norusis, M. J. (2009). SPSS 17.0 statistical procedures companion. United States: Pearson
Education.
Nwana, H. S. (1990). Intelligent tutoring systems: An overview. Artificial Intelligence Review, 4,
251–277.
Nykänen, O. (2006). Inducing fuzzy models for student classification. Educational Technology
and Society, 9(2), 223–234.
Ohlsson, S. (1996). Learning from performance errors. Psychological Review, 103(2), 241–262.
Ortony, A., Clore, G. L., & Collins, A. (1988). The cognitive structure of emotions. Cambridge:
Cambridge University Press.
Papageorgiou, E. (2011). Review study on fuzzy cognitive maps and their applications during the last
decade. In Proceedings of the IEEE International Conference on Fuzzy Systems. Taipei, Taiwan.
Papageorgiou, E. I., & Iakovidis, D. K. (2013). Intuitionistic fuzzy cognitive maps. IEEE
Transactions on Fuzzy Systems, 21(2), 342–354.
Papageorgiou, E. I., & Salmeron, J. L. (2012). Learning fuzzy grey cognitive maps using nonlin-
ear hebbian-based approach. International Journal of Approximate Reasoning,53(1), 54–65.
Papanikolaou, K. A., Grigoriadou, M., Kornilakis, H., & Magoulas, G. D. (2003). Personalizing
the interaction in a web-based educational hypermedia system: The case of INSPIRE. User
Modeling and User-Adapted Interaction, 13(3), 213–267.
Pearl, J. (1988). Probabilistic reasoning in expert systems: Networks of plausible inference. San
Francisco: Morgan Kaufmann Publishers Inc.
Pearl, J. (1996). Decision making under uncertainty. ACM Computing Surveys, 28(1), 89–92.
Pekrun, R., Frenzel, A. C., Goetz, T., & Perry, R. P. (2007). The control value theory of achieve-
ment emotions: An integrative approach to emotions in education. In P. A. Shutz & R. Pekrun
(Eds.), Emotion in education (pp. 13–36). London: Elsevier.
Peylo, C., Teiken, W., Rollinger, C., & Gust, H. (2000). An ontology as domain model in a
web-based educational system for prolog. In Proceedings of the 13th International Florida
Artificial Intelligence Research Society Conference. Orlando, Florida, USA.
Parvez, S. M., & Blank, G. D. (2008). Individualizing tutoring with learning style based feed-
back. In Proceedings of the 9th International Conference on Intelligent Tutoring Systems,
Montreal (pp. 291–301).
Peña, A., & Kayashima, M. (2011). Improving students’ meta-cognitive skills within intelli-
gent educational systems: A review. In Proceedings of the 6th international conference on
Foundations of Augmented Cognition: Directing the Future of Adaptive Systems (pp. 442–451).
Peña, A., & Sossa, H. (2010). Semantic representation and management of student models: An approach
to adapt lecture sequencing to enhance learning. In Proceedings of the 9th Mexican International
Conference on Advances in Artificial Intelligence: Part I, Pachuca, Mexico (pp. 175–186).
Popescu, E. (2009). Diagnosing students’ learning style in an educational hypermedia system. cog-
nitive and emotional processes in web-based education: Integrating human factors and person-
alization. Advances in Web-Based Learning Book Series (pp. 187–208). US: IGI Global.
Popescu, E., Badica, C., & Moraret, L.(2009). WELSA: An intelligent and adaptive web-
based educational system. In Proceedings of the 3rd Symposium on Intelligent Distributed
Computing, Ayia Napa, Cyprus (pp. 175–185).
Popescu, E., Badica, C., & Moraret, L. (2010). Accommodating learning styles in an adaptive
educational system. Informatica, 34, 451–462.
Pramitasari, L., Hidayanto, A. N., Aminah, S., Krisnadhi, A. A., & Ramadhanie, M. A. (2009).
Development of student model ontology for personalization in an e-learning system based on
semantic web. In Proceedings of the International Conference on Advanced Computer Science and
Information Systems (ICACSIS 2009), Universitas Indonesia, Depok, Indonesia (pp. 434–439).
154 References

Rich, E. (1979). User modelling via stereotypes. Cognitive Science, 3(4), 329–354.
Rivers, R. (1989). Embedded user models—where next? Interacting with Computers, 1, 14–30.
Rodrigo, M., Baker, R., Maria, L., Sheryl, L., Alexis, M., Sheila, P., Jerry, S., Leima, S., Jessica, S., &
Sinath, T. (2007). Affect and usage choices in simulation problem solving environments. In
Proceedings of the 13th International Conference on Artificial Intelligence in Education, Marina
Del Ray, CA, USA (pp.145–152).
Rodriguez-Repiso, L., Setchi, R., & Salmeron, J. L. (2007). Modelling IT projects success with
fuzzy cognitive maps. Expert Systems with Applications, 32(2), 543–559.
Sabourin, J., Mott, B., & Lester, J. C. (2011). Modeling learner affect with theoretically
grounded dynamic bayesian networks. In Proceedings of the 4th international conference on
Affective computing and intelligent interaction, Memphis, Tennessee (pp. 286–295).
Salim, N., & Haron, N. (2006). The construction of fuzzy set and fuzzy rulef or mixed approach
in adaptive hypermedia learning system. In Proceedings of the 1st International Conference
on Technologies for E-Learning and Digital Entertainment (Edutainment 2006), Hangzhou,
China (pp. 183–187).
Salmeron, J. L. (2009). Augment fuzzy cognitive maps for modelling LMS critical success fac-
tors. Knowledge-Based Systems, 22(4), 275–278.
Salmeron, J. L., Vidal, R., & Mena, A. (2012). Ranking fuzzy cognitive map based scenarios
with TOPSIS. Expert Systems with Applications, 39(3), 2443–2450.
Salomon, G. (1990). Studying the flute and the orchestra: Controlled vs. classroom research on
computers. International Journal of Educational Research, 14, 521–532.
Sison, R., & Shimura, M. (1998). Student modeling and machine learning. International Journal
of Artificial Intelligence in Education, 9, 128–158.
Schiaffino, S., Garcia, P., & Amandi, A. (2008). eTeacher: Providing personalized assistance to
e-learning students. Computers and Education, 51(4), 1744–1754.
Shakouri, H. G., & Menhaj, M. (2008). A systematic fuzzy decision-making process to
choose the best model among a set of competing models. IEEE Trans on Systems Man and
Cybernetics Part A: Systems and Humans, 38(5), 1118–1128.
Shapiro, J. A. (2005). An algebra subsystem for diagnosing students’ input in a physics tutorin
system. International Journal of Artificial Intelligence in Education, 15(3), 205–228.
Siddapa, M., & Manjunath, A. S. (2007). Knowledge representation using multilevel hierarchical
model in intelligent tutoring system. In Proceedings of the Third International Conference on
Advances in Computer Science and Technology. Thailand.
Siddappa, M., Manjunath, A. S., & Kurian, M. Z. (2009). Design, implementation and evaluation
of intelligent tutoring system for numerical methods (ITNM). International Conference on
Computational Intelligence and Software Engineering, Wuhan, China (pp. 1–7).
Smith, S. (1998). Tutorial on. Retrieved March 15, 2002, from http://www.cs.mdx.ac.
uk/staffpages/serengul/table.of.contents.htm.
Song, H., Miao, C., Roel, W., Shen, Z., & D’Hondt, M. (2011). An extension to fuzzy cognitive
maps for classification and prediction. IEEE Transactions on Fuzzy Systems, 19(1), 116–135.
Spada, H. (1993). How the role of cognitive modeling for computerized instruction is chang-
ing. In Proceedings of AI-ED’93, World Conference on Artificial Intelligence in Education,
Edinburgh, Scotland (pp. 21–25).
Staff, C. (2001). HyperContext: A framework for adaptive and adaptable hypertext. Ph.D. Thesis.
University of Sussex.
Stansfield, J. C., Carr, B., & Goldstein, I. P. (1976). Wumpus advisor I: A first implementa-
tion of a program that tutors logical and probabilistic reasoning skills. At Lab Memo 381.
Cambridge, Massachusetts: Massachusetts Institute of Technology.
Stash, N., Cristea, A., & De Bra, P. (2006). Adaptation to Learning Styles in E-Learning:
Approach Evaluation. In Proc. of World Conference on E-Learning in Corporate,
Government, Healthcare, and Higher Education, Orlando, Florida (pp. 284–291).
Stathacopoulou, R., Magoulas, G. D., Grigoriadou, M., & Samarakou, M. (2005). Neuro-fuzzy
knowledge processing in intelligent learning environments for improved student diagnosis.
Information Sciences, 170(2–4), 273–307.
References 155

Stula, M., Stipanicev, D., & Bodrozic, L. (2010). Intelligent modeling with agent-based fuzzy
cognitive map. International Journal on Intelligent Systems, 25(10), 981–1004.
Stylios, C. D., & Groumpos, P. P. (2004). Modeling complex systems using fuzzy cognitive
maps. IEEE Transactions on Systems Man and Cybernetics: Part A, 34(1), 155–162.
Suarez-Cansino, J., & Hernandez-Gomez, A. (2008). Adaptive testing system modeled through
fuzzy logic. In Proceedings of the Second WSEAS International Conference on Computer
Engineering and Applications (CEA’08), Acapulco, Mexico (pp. 85–89).
Sucar, L. E., & Noguez, J. (2008). Student modeling. In O. Pourret, P. Naοm, & B. Marcot (Eds.),
Bayesian networks: A practical guide to applications (pp. 173–185). West Sussex: Wiley.
Suraweera, P., & Mitrovic, A. (2004). An intelligent tutoring system for entity-relationship mod-
elling. Artificial Intelligence in Education, 14(3–4), 375–417.
Surjono, H., & Maltby, J. (2003). Adaptive educational hypermedia based on multiple student
characteristics. In Proceedings of the 2nd International Conference on Web-based Learning,
Melbourne, Australia (pp. 442–449).
Thomson, D., & Mitrovic, A. (2009). Towards a negotiable student model for constraint-based
ITSs. In Proceedings of the 17th International Conference on Computers in Education, Hong
Kong (pp. 83–90).
Ting, C.-Y., & Phon-Amnuaisuk, S. (2012). Properties of Bayesian student model for INQPRO.
Applied Intelligence, 36(2), 391–406.
Tourtoglou, K., & Virvou, M. (2008). User stereotypes concerning cognitive, personality and per-
formance issues in a collaborative learning environment for UML. In Proceedings of the 1st
International Symposium on Intelligent Interactive Multimedia Systems and Services (KES-
IIMSS 2008), Piraeus, Greece (pp. 385–394).
Tourtoglou, K., & Virvou, M. (2012). An intelligent recommender system for trainers and
trainees in a collaborative learning environment for UML. Journal of Intelligent Decision
Technologies, 6(2), 79–95.
Tretiakov, A., Sridharan, B., & Kinshuk, K. (2005). Conceptual modelling of web-based tutoring
systems. In Proceedings of the International Conference on Computers in Education, Altona
and Melbourne, Australia (pp. 2051–2058).
Tsaganoua, G., Grigoriadou, M., Cavoura, T., & Koutra, D. (2003). Evaluating an intelligent diagno-
sis system of historical text comprehension. Expert Systems with Applications, 25, 493–502.
Tsiriga, V., & Virvou, M. (2002). Initializing the student model using stereotypes and machine
learning. In Proceedings of the 2002 IEEE International Conference on System, Man and
Cybernetics (pp. 404–409).
Tsiriga, V., & Virvou, M. (2003a). Modelling the student to individualise tutoring in a web-based
ICALL. International Journal of Continuing Engineering Education and Lifelong Learning,
13(3–4), 350–365.
Tsiriga, V., & Virvou, M. (2003b). Evaluation of an intelligent web-based language tutor. In
Proceedings of the 7th International Conference on Knowledge-Based Intelligent Information
and Engineering Systems (KES 2003), Oxford, UK (pp. 275–281).
Tsiriga, V., & Virvou, M. (2003c). Initializing student models in web-based ITSs: A generic
approach. In Proceedings of the 3rd International Conference on Advanced Learning
Technologies (ICALT 2003), Athens, Greece (pp. 42–46).
Vasandani, V., & Govindaraj, T. (1995). Knowledge organization in intelligent tutoring systems
for diagnostic problem solving in complex dynamic domains. IEEE Transactions on Systems,
Man and Cybernetics, 25(7), 1076–1096.
Vélez, J., Fabregat, R., Nassiff, S., Petro, J., & Fernandez, A. (2008). User integral model in
adaptive virtual learning environment. In Proceedings of World Conference on E-Learning
in Corporate, Government, Healthcare, and Higher Education, Las Vegas, Nevada, United
States (pp. 3275–3284).
Viccari, R. M., Flores, C. D., Seixas, L., Gluz, J. C., & Coelho, H. (2008). AMPLIA: A probabilistic
learning environment. International Journal of Artificial Intelligence in Education, 18(4), 347–373.
Virvou, M., & Kabassi, K. (2002). F-SMILE: An intelligent multi-agent learning environment.
In Proceedings of IEEE International Conference on Advanced Learning Technologies 2002
(ICALT’02) (pp. 144–149).
156 References

Virvou, M., & Kabassi, K. (2004). Evaluating an intelligent graphical user interface by comparison
with human experts. Knowledge-Based Systems, 17(1), 31–37.
Virvou, M., Katsionis, G., & Manos, K. (2005). Combining software games with education:
Evaluation of its educational effectiveness. Educational Technology and Society, 8(2), 54–65.
Virvou, M., Lampropoulos, A. S. & Tsihrintzis, G. A. (2006). Inma: A knowledge-based authoring
tool for music education. In Proceedings of the 10th International Conference on Knowledge-
Based Intelligent Information and Engineering Systems, United Kingdom (pp. 376–383).
Wang, H. Y., & Chen, S. M. (2006). New methods for evaluating students’ answerscripts using
fuzzy numbers associated with degrees of confidence. In Proceedings of the 2006 IEEE
International Conference on Fuzzy Systems, Vancouver, BC, Canada (pp. 5492–5497).
Wang., X., Yang, Y., & Wen, X. (2009). Study on blended learning approach for english teaching.
In Proceedings of the 2009 IEEE International Conference on Systems, Man, and Cybernetics
(pp. 4641–4644).
Weerasinghe, A., & Mitrovic, A. (2011). Facilitating adaptive tutorial dialogues in EER-tutor.
In Proceedings of the 15th International Conference on Artificial Intelligence in Education,
Auckland, New Zealand (pp. 630–631).
Webb, G. (1998). Preface to UMUAI special issue on machine learning for user modeling. User
Modeling and User-Adapted Interaction, 8, 1–3.
Webb, G., Pazzani, M., & Billsus, D. (2001). Machine learning for user modeling. User
Modeling and User-Adapted Interaction, 11, 19–29.
Weibelzahl, S., & Weber, G. (2003). Evaluation of the inference mechanism of adaptive learning
systems. In Proceedings of the 9th International Conference on User modeling, Johnstown,
PA, USA (pp. 154–168).
Welch, M., & Brownell, K. (2000). The development and evaluation of a multimedia course on
educational collaboration. Journal of Educational Multimedia and Hypermedia, 9(3), 169–194.
Weon, S., & Kim, J. (2001). Learning achievement evaluation strategy using fuzzy membership
function. In Proceedings of the 31st ASEE/IEEE Frontiers In Education Conference, Reno,
NV (pp. 19–24).
Wilson, E., Karr, C. L., & Freeman, L. M. (1998). Flexible, adaptive, automatic fuzzy-based
grade assigning system. In Proceedings of the 1998 north American Fuzzy Information
Processing Society (NAFIPS) Conference (pp. 334–338).
Winter, M., Brooks, C., & Greer, J. (2005).Towards best practices for semantic web student
modeling. In Proceedings of the 12th International Conference on Artificial Intelligence in
Education, Amsterdam, the Netherlands (pp. 694–701).
Wisher, R. A., & Fletcher, J. D. (2004). The case for advanced distributed learning. Information
and Security: An International Journal, 14, 17–25.
Xu, D., Wang, H., & Su, K. (2002). Intelligent student profiling with fuzzy models. In
Proceedings of the 35th Hawaii International Conference on System Sciences.
Yang, G., Kinshuk, K., & Graf, S. (2010). A practical student model for a location-aware and
context-sensitive personalized adaptive learning system. In Proceedings of the IEEE
Technology for Education Conference, Bombay, India (pp. 130–133).
Zadeh, L. A. (1965). Fuzzy sets. Information and Control, 8(3), 338–353.
Zadeh, L. A. (1975). Fuzzy logic and approximate reasoning. Synthese, 30, 407–428.
Zadeh, L. A. (1979). A theory of approximate reasoning. In J. Hayes, D. Michie, &
L. I. Mikulich (Eds.), Machine Intelligence 9 (pp. 149–194). New York: Halstead Press.
Zadeh, L. A. (2001). A new direction in AI—toward a computational theory of perceptions. AI
Magazine, 22(1), 73–84.
Zapata-Rivera, D. (2007). Indirectly visible Bayesian student models. In Proceedings of the 5th
UAI Bayesian Modelling Applications Workshop. Vancouver, BC, Canada.
Zatarain-Cabada, R., Barrón-Estrada M. L., Angulo V. P., García, A. J., & García C. A. R.
(2010). Identification of felder-silverman learning styles with a supervised neural network.
In Proceedings of the 6th International Conference on Intelligent Computing (ICIC 2010),
Changsha, China (pp. 479–486).