ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
A LEARNER CENTRIC BLENDED E-
LEARNING CONTENT
RECOMMENDATION SYSTEM
Dr. G. Madasamy Raja,
Associate Professor, Department of Information Technology,
Paavai Engineering College, Namakkal, Tamil Nadu, India.
email: madasamyganapathypec@[Link]
Dr. K. Selvi,
Professor, Department of Computer Science and Engineering,
Paavai College of Technology, Namakkal, Tamil Nadu, India.
email: selvimidu@[Link]
Dr. M. Mohammed Thaha,
Assistant Professor, Department of Computer Science and Engineering,
B. S. Abdur Rahman Crescent Institute of Science & Technology, Chennai, Tamil Nadu, India.
email: mohammedthaha@[Link]
M. S. Affia Thabassum,
Research Scholar, Department of Computer Applications,
B. S. Abdur Rahman Crescent Institute of Science & Technology, Chennai, Tamil Nadu, India.
email: affiathabassum98@[Link]
Amena Mahmoud,
Assistant Professor, Department of Computer Science,
KafrElSkeikh University, Egypt.
email: amena_mahmoud@[Link]
Abstract
The growth of personal learning experiences has made it harder for students to find suitable
learning resources. Individual guidelines have been used to facilitate learners' experiences in
personal learning settings, and this technology will provide learners with appropriate learning
content. This algorithm evaluates the multi-dimensional characteristics of the content, student
ranking and the order and temporal patterns of the obtained information in a single model in
order to increase the accuracy of its recommendations. There are two modules in the current
solution. Late trends of obtaining resources are found and presented in two formats with the
balanced association rule mining and a compact tree structure known as the sequence tree in the
sequential-based suggestion module. The learner's preference tree is implemented to consider
multifarious attributes of papers, ranking of learners and order of materials in the attribute-based
module, after clustering learners with latent patterns by K-means algorithm. The hybrid
[10113]
ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
approaches are blended, weighted and cascaded with effective suggestions for implementations.
The tests demonstrate that the solution proposed exceeds evolutionary mechanisms in terms of
accuracy, reminder and intra-list similitude. Improved consistency of the guidelines and a
reduction of the sparseness dilemma are the key contributions by integrating qualitative data
through order and temporal trends of the resources accessed, apprenticeship ratings and
multidimensional materials.
Introduction
Due to the advancement of web-based IT and communication technology, e-learning programs
are increasingly popular in educational institutions [1]. The exponential development of e-
learning programs has transformed conventional learning behaviour and offers learners a
different situation, which significantly facilitates and strengthens online learning habits. Due to
the introduction of several types of teaching practices in the e-learning world, which can include
subjects, teaching activities, tools, etc, it is difficult for learners to choose the learning activities
that better satisfy their requirements. In the Big Data age, the issue of knowledge overload is
more and more severe. E-Learning system is important if individual recommendations are to be
created dynamically to direct the work of a student [2], and as Law [3] has shown, a customized
feedback e-Learner system needs to be put forward. This research motivates us to establish a
suggestion methodology in order to assist students in choosing the most suitable e-learning
services.
In accordance with their applicable contexts, e-learning programs can be categorized into two
categories, namely, formal and informal settings [4]. A structured e-learning framework involves
educational institutions learning offerings in the form of curricula or syllabus. In literature, an
informal environment is defined as a learning stage by so-called lifelong learners who do not
take part in formal education and are accountable for their own speed and course of learning [5].
The method of learning depends largely on personal decisions or desires, and is mostly self-
directed [6].
Unlike the formal environment, the informal setting can provide a range of teaching
opportunities from various suppliers, where students also have diverse backgrounds. Normally,
there is no program or education. Therefore, the choice of proper learning practices in the
informal context is very complicated for students. High drop-out rates and low termination rates
will result. This research focuses on the implementation of a new, tailored suggestion approach
on assisting students in the informal environment.
[10114]
ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
Figure 1. A sample fuzzy tree
Recommendation systems was first introduced and used in the e-commerce field for the
procurement of products as one of the most common frameworks for customization techniques.
Recommend systems can be described as programs which try to recommend items to users by
forecasting the interest of a user in a certain item on the basis of several types of information,
including information on items, users and interactions between users and items. The fundamental
principle of the scheme recommended is that similar people prefer similar objects.
Therefore, in the application of recommending schemes, the similarity measures for users or
articles are significant. Web-based e-commerce, e-business, e-tourism and e-government
platforms is commonly used in a wide variety of web-based applications. This is primarily
because the e-learning practices have unique attributes, specifications and similarities that are
distinct from commercial items of e-commerce and e-business. Both learning and learning
profiles have detailed definitions and characteristics.
A learning process includes many aspects of information such as the definition of content, the
lectures, the preconditions, etc., while a learner profile comprises the learner's context, research
aims, prior experience, learning characteristics etc. With many sub-aspects, each part of the
knowledge may be detailed. The data thus has a hierarchical structure in the e-learning world.
Learning and learning profiles often provide ambiguous details in a real-life setting. There could
be several different classes for a learning activity.
Business intelligence, for example, is mostly concerned with information technology but also
with business. The specifications of the learner typically are defined as highly necessary or very
significant in linguistic terms. Fuzzy set approaches are appropriate for addressing these
ambiguous data categories and language words. Therefore, the tree-structured learning and
learning profiles are represented as floating trees. In the recommendation for instructional
activities, pedagogical problems should be considered. Certain learning practices include
preconditional training.
To research data mining, for example, the data base and algorithms need to be known
beforehand. In addition, students will often want to experience something different or of greater
[10115]
ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
degrees of complexity, so these kinds of preferences between learning tasks must be taken into
consideration. There is no way to vary from one or other learning activities, since learning from
different schools may have different names, for example one subject is called Java and one is
called Fundamental Software, but the same or identical contents are used. The other subject is
called Fundamental Programme.
This paper proposes a fusing tree-based hybrid-recommendation solution to deal with the above
special requirements of e-learning structure. Based on our past findings on the fuzzy preference
table scheme, the student profiles and learning behaviours are outlined in a fluffy tree
hierarchical data model in Fig.1. Fuzzy methods are applied to deal with the unclear problems.
As a key technique of recommendation approaches is the similarity metric, appropriate tree
similitude measures are created. The solution to a recommendation takes advantage all of the
recommendation processes focused upon information and collective filtering and calls similitude
both semantic and collaborative.
The priorities of learning activity are often handled by assessing the study sequences and shaping
the necessary learning activities. The research discussed in this document is ground breaking
since it was the first model to model learning behaviours and profiles using fluffy tree-structured
data model. It contributes in the areas of e-learning as well as recommending systems on both
theoretical and functional aspects. On the theoretical level, fuzzy tree organized data patterns are
formed as well as related tree similarity tests. A fuzzy tree-based hybrid e-learning frameworks
suggestion methodology is built at the functional level.
Existing Work
In the field of e-learning, recommendation mechanisms have recently been implemented.
Felfernig[7] suggested a method that employs techniques of data mining, such as association
rules mining, to model consumer actions, proposing tasks or shortcuts. In [8], the device
framework for customized e-learning was suggested. In this paradigm, a student demand analysis
model of multi-criteria is designed to recognize the needs of a student; a flippant combination
approach is employed to address ambiguous real-life criteria values. By taking the learner's
awareness levels into account, the approach to CF recommendation was tailored for e-learning
purposes.
In the e-learning materials recommendations [9], the attributes of materials are considered, and
the recommendations are produced using CB, CF and some hybrid approaches. A
recommendation solution that blends a fluctuating mutual filtering algorithm with a content-
based algorithm, using learner expectations or information significance, was suggested in order
to minimize the consistency versus the plasticity issue of enhanced learning recommender
systems. To our best understanding, no research has been undertaken in e-learning
recommendations to completely overcome fluid tree-structured data. Much interest was paid to
recommendation methods and several alternatives to recommendations were suggested. Overall,
[10116]
ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
interactive filters (CF), information (CB) and experience and understanding (KB) techniques
[10] are among the three most widely used approaches to recommendations. The CF
methodology assists people in taking actions based on the experiences of other people of
common interests [11]. It can be further broken down into CF methods based on users and
objects. CB strategies suggest that items close to what a single person has traditionally selected.
KB approaches provide users with articles based on user awareness and articles.
Each technology has its own drawbacks, including the problem like material dependence, over-
specification for CB and the problems of cold beginning and scarcity for CF [9]. A hybrid
recommendation approach can be established by incorporating the best characteristics of two or
more recommendation approaches to obtain a higher performance and avoid the pitfalls of
traditional recommendation approaches. A range of advice methods have been developed and
implemented for advisory systems like data processing, agents and reasoning.
Many advanced approaches to recommendation, such as social-network recommendation
systems, fuzzy recommendation systems, context-based recommendation systems and group-
recommendation systems were also recently proposed. The language learning and the sequences
of information access in the learning process typically have a repeatable, periodic time dependent
relationship. Consequently, the relationship between the time dependence of learning materials
may represent the interests of a student and the patterns of content access. The order of access to
the learning content is, for example, {reading notes, references, exercise book}.
If a student has accessed the lecture notes recently, it is most likely that he or she will soon
access the guide. We may then mine the access records of the student to find the sequential
content access trends. We can then estimate the most likely content that a student could access in
the coming years by using this sequential pattern. This helps to increase the consistency of the
suggestions and address the issue of sparsity. A sequential pattern for access is a sequential
pattern that is mostly used by users in a vast number of Access Logs. Sequential pattern mining
has two big algorithms (SPM) and Generalized Sequence models (GSP)[12].
As most Apriori algorithms, these algorithms also face the challenge of scanning multiple
database to assess the frequencies of the applicants. The second group of optimizations to
improve SPM's efficacy are normally based on pattern development, including PrefixSpan. In
comparison with other modules, the PrefixSpan algorithm proved to be more efficient. In this
study the weighted association rules and the PrefixSpan algorithm are used to locate the pattern
structure using two algorithms for sequential pattern mining and the modified Apriori algorithm.
Just whether an object is in a sequence or not is the conventional model of associated laws.
We also applied the weighted association rules mining technique in order to increase the
efficiency of Tao et al. association rules to connect one weight parameter to each learning
material in an association rule. The analysis sample weight in the t = {N1, N2,, ., Nn} series is
determined according to the visiting time of the material. The concept of using a wavy mining
[10117]
ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
algorithm is useful, since we will take account of the needs of the learner and ultimately increase
the suggestion results by examining the duration of the visit of the material in rules mining. The
steps of Apriori algorithm[13] will be amended to represent the weighting method in the
following paragraph. A model tree is used to compactly store sequence access patterns. The tree
of patterns is built by the procedure described in [14].
Due to its superiority over the tree structure historically used, they used a patricia-based data
structure for website recommendations. We just need a search of all consecutive network
conditions to create a pattern tree. It provides the template for tree building. We need to find the
best matching access route in the sequence according to an active user's access sequence to
produce recommendations. A sliding window (w) controls the number of materials to align with
the found sequences to determine a recommendation set for the new active learner sequences.
The matching procedure is carried out as the prefix series on the current materials from the aim
student. The overall access sequence length for an active learner will be less than the length of
the pattern tree and it should be noted. It must be noted that However, given that the suggestions
produced from shorter fit paths normally have lower exactness, it is only possible to process the
material access series with a duration not less than a certain cut-off.
Proposed System
In this sub-section, consistent with earlier studies, a tree-structured matching approach is
proposed (Fig.2). A maximal conceptual similarity tree mapping is developed to distinguish the
sections of two trees that most conceptually match. The mapping takes into account tree
architectures, node definitions and node weights. The conditions for two trees are distinct in
opposing implementation situations. For eg, the weights of both trees should be taken into
consideration when considering two trees. Another example is when a sub-tree matches with a
target tree to decide if the target tree comprises the sub-tree, which mostly weighs the weights of
the sub-tree.
Figure 2. Proposed model
[10118]
ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
The matching approach should then take into consideration the two categories of cases. In the
first case, symmetric matching is called, while the second is called asymmetric matching. During
the estimation of the conceptual similarities between two trees, the full conceptual similarity tree
maps can be built. The logical resemblance, depending on the matching form, is both symmetric
and asymmetric. They are referred to and are defined as the matching type. This section presents
the data framework of the e-learning recommendation system's learning events. A learning
activity can be defined from a number of viewpoints, including criteria, categories, information,
lectures, etc., and some features that create a topographical map can also be described from a
range of subfunctions.
In actual implementations, certain features of a learning experience are unsure. For example, the
subject of market intelligence is primarily in the category of IT but is also in some degree in the
category of enterprise. Therefore, the fuzzy tree-structured data model is used to model our
research tasks in order to cope with the tree-structured data and fluffy category data in the field
of learning. Each node is assigned a mark attribute in the learning activity tree. A group attribute
is allocated to certain nodes. The similitude concept node is dependent on all characteristics. A
valuable feature to the category of learning is to infer semantic relationships between various
learning events. One learning experience will fall into a number of classes of varying degrees in
a real-life setting. The category meaning in our scheme is, thus, a fluctuating category tree. There
are detailed descriptions of the fuzzy group trees and their resemblance measures. The maximum
weighted bipartite matching results can be reported during the calculation step of the conceptual
similarities between two trees. Based on the records, it is possible to identify the most relevant
nodes between two trees. The bases of tree branches are matching pairs of nodes.
The appropriate nodes of two-root children are then classified based on the maximal two-party
weighted matching between the two root children. Additional nodes are similarly defined. The
learning mechanism concerning repetitiveness, periodicity and certain relationships of
dependence should be considered in the guideline for learning behavior [15]. New or marginally
higher than the existing degree of competency for students must be recommended as learning
practices. It is not fair to prescribe elementary tasks to a learner for any similar learning
experiences or for any similar groups if he or she has learned advanced work already. Our
scheme regards two types of primary relationships among learning practices. The first type of
relationships is based on the criteria of learning activity. Most learning experiences are subject to
required courses. The basic learning tasks are defined in the foggy tree-structured learning
experience profiles for the learning activity[16].
The second form of relationship is based on sequences of instruction in the learning experience
of learners. This learning sequences can be used to determine advanced stages of learning tasks
that are hard to recognize in an informal learning context because of the open environment.
When a student chooses a learning experience, the learner is informed by different forms of
[10119]
ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
knowledge, including the learner's context, the purpose of education, needed learning categories
and learning activities.
Both these aspects of knowledge are taken into account in our recommendation scheme. In
general, there are many sub-aspects for each aspect. This creates a foundation of a tree.
Language specifications, such as "very much needed," are more likely to be expressed in actual
applications. These linguistic concepts can be dealt with in fuzzy set techniques. The fluffy tree-
structured data template is then used to model the student profiles. The fluffy tree-structured
learner model is explained in this section. In the same way as the learning operation tree, a name
attribute and group attribute are applied to the learner profile tree nodes to measure the node
definition similitude.
The learning tasks are documented in the learner profile during the learning process by the
system. The learner outlines supplementary details, such as the learner's history, the intended
occupation and the appropriate categories when the student is enrolled. In a predetermined job
list including software engineer, computer programmer, auditor, etc, is chosen to cover the
intended career in particular. Similarity steps are necessary for recognizing similar users or
products and matching relevant items to the criteria of the users during the recommendation
generation process. In an e-learning environment there are different learning experiences in
different categories, but only the learning process under those categories are appropriate for
suggestion for a particular target learner. The specific groups of target students are first
established and learning experiences under the clusters are then chosen to increase the
recommendation's effectiveness.
Results and Discussion
For our studies, we use a real-world dataset. The learning record dataset is based on the use data
of Moodle, the course management system. A Cursual Management System (CMS), a free and
open-source software package that used pedagogical concepts to help educators build productive
online education communities, is described as a modular object-oriented, dynamic teaching
environment. Moodle tracks the student events in depth. In the category defined by the Moodle
method, the instructor may access the summary reports on these events. There are various
immersive learning modules available for Moodle, such as boards, conversations, quizzes and
tasks.
[10120]
ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
Figure 3. Comparison of results
Furthermore, via Moodle, user access identity, IP, time, and actions and information accessed by
users can be recorded and monitored. The device stored the access time, the IP address of the
user's machine, the user ID and the kinds of activities and materials visited in each log. With this
log, Moodle will produce activity reports for each student with charts and data In our tests, a
real-world dataset is used. The data processing of learning records is derived from the user data
of Moodle. In content suggestions, only a limited number of rules with a greater importance of
guidance and trust can be shown to the pupils of learning materials.
Consequently, an acceptable minimum assistance and trust threshold must be chosen. The testing
findings show that a good minimum support level and minimum trust threshold of 0.25 and 0.5
respectively are not achieved in order to achieve a number of laws. We also adjusted window
sizes from 1 to 10 to take into account the effect of window size on the proposed suggestion
algorithm. The framework has more details on the big sliding window, while the occurrences that
appear less often in the dataset allow a wider state space. A size 1 window cannot carry the
advice properly.
The precision thus improves as the screen size increases. But there is no variation in precision
between window size 4 and 6. The truth is that the states comprise the sequences of substance
visits is less common in the dataset. We consider therefore w = 5. Important criteria are the
minimum support and the sliding window scale for Pattern-based tree recommendations. We
take w=5 and min support=0.25 into account in this solution. The numbers of clusters (C), k and
a are important parameters for the attribute-based recommendation method. We have C = 8 for
our dataset according to the measurement of amounts in the cluster.
[10121]
ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
Figure 4. Error comparison
We carry out an experiment to decide the best values for k and a, as shown by figure. We set kn
= 0,65 or a = 0.8 as per the test results. This indicates even the influences of the ABR guideline
on its accuracy, while Q = 8. It suggests that knowledge about materials and students by the
attributes results in better recommendations, and the highest accuracy can be obtained at a = 0.8.
The optimum value is dependent on the data collection and must be calibrated for multiple
datasets. The standard establishes and maximum trust threshold, and the sliding window scale,
are relevant criteria for rules-based recommendations. In content suggestions, only a limited
number of rules with a greater importance of guidance and trust can be shown to the pupils of
learning materials.
Figure 5. The accuracy value of proposed model
Therefore, an acceptable minimum assistance and trust threshold must be chosen. The testing
findings show that a good predefined threshold level and maximum trust threshold of 0.25 and
0.5 respectively are not achieved in order to achieve a number of laws. We also adjusted window
sizes from 1 to 10 to take into account the effect of window size on the proposed suggestion
algorithm. The framework has more details on the big sliding window, while the sequences that
[10122]
ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
appear less often in the dataset allow a wider state space. A size 1 window cannot carry the
advice properly. The precision thus improves as the display size increases. But there is no
variation in precision between window size 5 and 6. That is because the states comprise the
sequence of the substance visits, which is less common in the dataset.
Figure 6. Error function
We consider therefore w = 5. Important criteria are the required support and the local
thresholding scale for Pattern-based tree recommendations. We take w=5 and min support=0.25
into account in this solution. We perform an experiment to decide the best values for k and a that
shows the outcome for a. We set k = 0.65 and a = 0.8 according to the test results. The effect of a
on the ABR recommendation accuracy is also shown in C = 8. It suggests that knowledge about
materials and students by the attributes results in better recommendations, and the highest
accuracy can be obtained at a= 0.8.
The optimum value is dependent on the data collection and must be calibrated for multiple
datasets. teachers, device managers and students are configured for three different groups of
users in the TeLRS e-learning recommendation system. The user functions are outlined below.
The task of the system manager is to preserve the framework of the category of learning
activities and the list of students used to promote the functioning of the system. The teachers
manage the tasks of studying. The learning tasks are incorporated into the framework using
concise explanations.
When a study exercise is entered, the instructor specifies the divisions and the associated
affiliation degrees. The teachers receive input from students on their learning experiences and
communicate with their students during the process of the programme. Students check and
collect feedback for the necessary learning experiences. When registering in the scheme, they
have their history and learning needs. The student will provide input and evaluate the learning
process after a learning activity has stopped. The e-learning suggestion system design is seen.
Conclusion
There are three layers in the web server: the layer of presentation, the business logic layer and
the access layer of data. The display layer shall generate and handle the logic of the user
[10123]
ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
interface and the three user types of events on the requested web pages. The layer of business
logic includes business resources and the central algorithm for suggestion. It comprises four
major sections. The student centre, the teaching centre, the administrative centre and the
Suggestion Engine. The student centre gathers the description and needs of the users, monitors
the actions of the user and makes the quest and feedback for learning opportunities accessible.
The Recommendation motor applies the suggested recommendation method and provides student
users with feedback. Teachers participate and plan the teacher's instructional programs.
Administrators control the users and general data through the Developer Hub. The service layer
addresses the database data processes.
As the recommending programs became effective, research on infrastructure learning began to
work on recommending learning strategies. In order to increase the accuracy of the suggestions,
we suggested the hybrid recommendation method focused on the sequential trends and
characteristics of the resources accessed, taking account of the contextual learner and material
knowledge. The findings of the experiment show that our algorithm will greatly exceed the
standard recommending algorithms in accuracy, reminder, and the intra-list similitude. In
addition, there are a few restrictions which may decide potential avenues for future
investigations. In academic settings students with superior experience must be found to be more
relevant than less knowledgeable students when calculating their recommendations. Thus, a new
methodology should be introduced to consider learner's expertise and improve current algorithms
for recommendations appropriately in order to maximize the outcome of the recommendation.
References
1. Burke, "Knowledge-based recommender systems," Encyclopedia of Library and Information Systems, vol.
69(32), pp. 175-186, 2000.
2. Burke, "Hybrid web recommender systems," in The Adaptive Web. vol. 4321, P. Brusilovsky, A. Kobsa, and
W. Nejdl, Eds., ed: Springer-Verlag, Berlin Heidelberg, 2007, pp. 377-408.
3. Cheung, J. T. Kwok, M. H. Law, and K.-C. Tsui, "Mining customer product ratings for personalized
marketing," Decision Support Systems, vol. 35, pp. 231-243, 2003.
4. Chuan-chang, Telecommunications search recommendation model based on user search behavior and gradual
forgetting collaborative filtering strategy, J. China Univer. Posts Telecommun. 17 (3) (2010) 110–117.
5. De Meo, A. Garro, G. Terracina, D. Ursino, Personalizing learning programs with X-Learn, an XML-based,
‘‘user-device’’ adaptive multi-agent system, Inf. Sci. 177 (8) (2007) 1729–1770.
6. Drachsler, H.G. Hummel, R. Koper, Personal recommender systems for learners in lifelong learning networks:
the requirements, techniques and model, Int. J. Learn. Technol. 3 (4) (2008) 404–423.
7. Felfernig, G. Friedrich, L. Schmidt-Thieme, Introduction to the IEEE intelligent systems special issue:
recommender systems, IEEE Comput. Soc. (2007).
8. García, C. Romero, S. Ventura, C.d Castroa, A collaborative educational association rule mining tool, Int. High.
Educ. 14 (2) (2011) 77–88.
9. García, C. Romero, S. Ventura, C. Castro, An architecture for making recommendations to courseware authors
using association rule mining and collaborative filtering, User Model. User-Adapted Int. 19 (1–2) (2009) 99–
132.
10. Ghauth, N.A. Abdullah, Learning materials recommendation using good learner’s ratings and content-based
filtering, Educ. Technol. Res. Develop. 58(6) (2010) 711–727.
[10124]
ISSN: 0011-9342 | Year 2021
DesignEngineering Issue: 8 | Pages: 10113- 10125
11. Gu, S.-J. Hong, J.-B. Yoo, K.-M. Lee, Personal preference for reliable transaction identification on web service,
in: ICIW ‘08 Proceedings of the 2008 Third International Conference on Internet and Web Applications and
Services, 2008, pp. 150–155.
12. Herlocker, J.A. Konstan, L.G. Terveen, J.T. Riedl, Evaluating collaborative filtering recommender systems,
ACM Trans. Inf. Syst. (TOIS) 22 (1) (2004) 5– 53.
13. Khribi, M. Jemni, O. Nasraoui, Automatic recommendations for e-learning personalization based on web usage
mining techniques and information retrieval, Educ. Technol. Soc. 12 (4) (2009) 30–42.
14. Klašnja-Milic´evic´, B. Vesin, M. Ivanovic´, Z. Budimac, E-Learning personalization based on hybrid
recommendation strategy and learning style identification, Comput. Educ. 56 (3) (2011) 885–899. R. Burke,
"Hybrid recommender systems: survey and experiments," User Modeling and User-Adapted Interaction, vol.
12, pp. 331-370, 2002.
15. Pazzani and D. Billsus, "Content-based recommendation systems," in The Adaptive Web. vol. 4321, P.
Brusilovsky, A. Kobsa, and W. Nejdl, Eds., ed: Springer Berlin Heidelberg, 2007, pp. 325-341.
16. Schafer, D. Frankowski, J. Herlocker, and S. Sen, "Collaborative filtering recommender systems," in The
Adaptive Web. vol. 4321, P. Brusilovsky, A. Kobsa, and W. Nejdl, Eds., ed: Springer Berlin Heidelberg, 2007,
pp. 291-324.
17. Premanand, R.P., Rajaram, A. Enhanced data accuracy based PATH discovery using backing route selection
algorithm in MANET. Peer-to-Peer Netw. Appl. 13, 2089–2098 (2020). [Link]
00824-1
18. Rajaram.A., [Link] . Malicious Node Detection System for Mobile Ad hoc Networks. (IJCSIT)
International Journal of Computer Science and Information Technologies, Vol. 1 (2) , 2010, 77-85
19. [Link], Ayyasamy Rajaram. An Enhanced Distributed Certificate Authority Scheme for
Authentication in Mobile Ad hoc Networks. The International Arab Journal of Information Technology
(IAJIT).vol.9 (3),291-298.
[10125]
