Investigation of Graph Mining for Business Processes

Azeem Lodhi, Gamal Kassem, Veit Koeppen, Gunter Saake

Abstract—Business process management and business intel- frequently used to accomplish defined tasks? What are the
ligence are fields which gain a lot of attention in recent years. common characteristics and relationships between activities,
These techniques try to improve not only efficiency of processes business objects, and their flow or relation (business process
but also save considerable cost. Graph based representation
of concepts (objects, data) are also used in business domain executions)? What common features can we discover be-
to support aforementioned techniques. Graph mining methods tween successful and unsuccessful scenarios? Prediction on
are successful in many fields for discovery of new relations, a certain business flow whether it will lead to a desired state
knowledge, and visualization. In this paper, we briefly discuss or failure. What commonalities can we find between the
the fields in which graph mining is successfully applied. We executions of business applications by users? How should
also discuss challenges of applying graph mining in business
processes and what are the benefits. an organization be structured to get maximum benefits from
the employees (dealing with social network analysis)? What
Keywords-Graph mining, business processes, graph mining would be the next information request from the user of a
applications, business intelligence, business process analysis
system during business process execution?
Some of these questions are investigated by researchers
I. I NTRODUCTION
and we discuss them in Section V while others are still
Business processes are important elements for the success needed to be investigated. This paper demands further in-
of companies and to perform operations in a organization. vestigation in graphical representation of business process
Due to this reason, enterprises always try to perform business elements (objects, data). Mining the relational knowledge in
processes in an effective and efficient manner. Different business processes requires data to be presented in such a
technologies are used to implement and manage business way that it preserves not only the relational information but
processes successfully. Workflow management systems are also supports mining tasks effectively and efficiently [4].
used to define, manage, execute, and monitor the business The structure of the paper is as follows: we briefly
processes where IT can support in an efficient way. Work- discuss the basics of business processes and graph mining in
flow, itself, is defined as automation of a business process Section II followed by Section III which discusses possible
in a whole or a part during which documents, information, benefits of applying graph mining in the business process
and other elements are passed through resources according to management domain. Challenges of applying graph mining
procedural rules [1]. To understand the execution of business on business processes are discussed in Section IV. Section V,
processes and rules, a considerable amount of research is provides a brief overview of related work of graph mining
carried out in this field. Modeling languages and analytical in business processes and in some other fields followed by
methods are proposed to provide better capabilities for Section VI where we discuss some other issues of graph
understanding and improvement in business processes. mining. Section VII summarizes the paper and provides the
The representation of concepts (objects, data) based on outlook of our work.
graphs is successful in many fields, for example, chem-
ical structural analysis, bioinformatics, predictions like II. G RAPH M INING AND B USINESS P ROCESSES
successful-movie predictions [2], and successful termination A graph consists of nodes and edges which connect nodes.
of processes [3]. In such fields, the relational information Graphical representation is used for the visual representa-
between entities and their attributes is very important as it tion of concepts. A graph provides a convenient way to
helps for discovery of new knowledge. In business process represent relation between entities and respective data. In
management, this new knowledge can give answers to var- graph mining, we exploit relational information to extract
ious questions related to business process improvement as patterns and discover new knowledge. New knowledge can
stated in the following. be a pattern such as sub graph, an undiscovered relation to
Which activities are frequently executed? Which path is other elements, or on a more abstract level expression of
trends in data [4].
Published in ICIIT’10, 28-OCT-2010, Lahore, Pakistan. IEEE c Published by the IEEE Computer Society
 Graph based representation and graph mining have been tation in business processes is the discovery of new relations
used in various domains where visualization helps for better between activities, their attributes, and an overall impact
understanding and management of tasks. For example, in on activities in business processes. Representing a business
communication networks, it is used for monitoring and man- process and its attributes in a graph saves the relationship
agement of complex networks, defining connection policies, between data objects and activities, which is extremely
capturing recurrent patterns in traffic, routing instabilities important for discovering relational knowledge. Discovery
detections, and extrapolation of networks (designing next of relational knowledge using graph mining methods has
generation protocols, etc.). It is also used in the social been discovered by researchers in media industry also [9].
network analysis domain where various applications are de- An often used example of graph mining in media industry
veloped for analyzing the relations between different individ- uses IMDB1 (Internet Movie Database) website as a resource
uals like homeland security (crime and narcotics trafficking website. This website contains information about movies and
connections), computation biology, and web-search [4]. television programs. It provides movies and programs detail
A business process is a set of inter-related activities which information to users through online queries freely. Graph
are executed in a specific order to accomplish a given mining is used on IMDB movie database in which movies
task. It can be as simple as fulfilling an order form or attributes (actors, director, producer, etc.) are represented as
complex such as performing the whole order process from graph nodes, and then graph mining methods are applied to
selecting suppliers up to paying the bill. Business process discover new relations and knowledge. This knowledge is
management is concerned to define, design, execute, and used to make a predictions like how much business will a
evaluate the business activities efficiently and utilizing its certain movie do during a time-span? Will it be nominated
resources effectively such that the organization’s goal is for an award or not? The results of applying graph mining
achieved. A support for understanding this complexity is on IMDB and to some other fields are discussed in [4].
done by modeling business processes. They are graphically Graph mining methods can also be used to predict the
modeled in various languages for communication, analysis, information requirements of a user during the execution
and understanding during the business process management of activities in enterprise applications. Graph mining is
life cycle. explored for user navigation behavior over the web in
Besides different methods to represent business process [10]. Information generated from business intelligence tools
data, we discuss two methods of representation for a formal can be combined with graphical models to provide better
analysis of workflows. One method is used in [3], where analytical facilities for decision making. For example, it
activities are represented as nodes and other notations like can be used to predict whether a certain execution will
bold and dotted edges are used to distinguish between AND lead to achieve the enterprise’s objective or not. Analysis
and XOR structures. The second method is using bipartite can help users to devise company’s best practices that how
graphs called Petri nets [5] in which activities are denoted tasks/activities should be routed based on attribute values.
as places and actions as transition. In this paper we suggest Graph mining methods can be applied to find out the
to use the latter notation for graph mining because of effect of new business collaboration by analyzing relations
mathematical formalism and other reasons discussed in [6]. and collaboration among existing activities, new activities,
One extension of Petri nets is proposed in [7] for analysis and their participants. Similarly, graph based methods can
of business processes with business objects. be used to improve the organizational structure, e.g., which
A Petri nets is bipartite directed graph, which consists employees should work together, see [11]. Application of
of places (denoted by circles), transitions (denoted by graph mining in application usage mining [12] can find the
rectangles), and arcs for connection between them. Places deficiencies between user and system interaction and help
represent buffers, storage places where conditions are eval- to improve business-IT alignment.
uated and decisions are made, and transitions represent
tasks/events/actions. Places can contain tokens which de-
scribe the state of the system and values of related attributes IV. C HALLENGES
at a certain stage [8].
Representation of business processes in graphs is a chal-
III. B ENEFITS OF G RAPH M INING IN B USINESS lenging task as several elements are related with business
P ROCESSES processes like inputs, outputs, resources, rules, and their
Graphs are useful to preserve the structural relationship attributes. Graph mining requires all involved objects to be
between elements, as in the case of chemical compound represented in graph elements. We discuss some of the issues
structure analysis and in bioinformatics, whereas in other related with representation of business processes for graph
conventional approaches, there is a loss of relational infor- mining as follow.
mation between elements. One of the most interesting bene-
fits that can be achieved by applying graph based represen- 1 www.imdb.com
A. Granularity Another challenge which also demands further investiga-
In enterprises, business processes are very large and tion is the transformation of graphs (containing relational
complex, involving hundreds of resources and activities. information of business processes) into other representations
Analysts require a different level of granularity to under- like sequences. However, another question arises how should
stand the business processes. Business process models are the transformation be carried out in such a way that there is
viewed by different stakeholders and so at different levels of no semantic loss (relational information)? How this can be
granularity (abstract or in detail). Executives are interested evaluated whether there is any information loss or not? Such
in an overall picture of business processes, thus they require questions are still needed to be investigated, so mining tech-
a suitable abstraction of BP models, whereas operational niques can be applied efficiently to discover new knowledge
managers look for specific details of processes and activities of process structures.
of their responsibility. V. R ELATED W ORK
Both modeling approaches described in Section II (graph
and Petri nets based) do not provide a suitable flexibility Graph based representation of data gains a lots of attention
in granularity. A new representation of business processes from researchers and is applied to many fields like chemical
should be proposed, which has enough flexibility and se- industry, bioinformatics, and business process management.
mantics to fulfill the granularity demands of an analyst Due to the scope of this paper which is business process
and at the same time it is formal enough for analyses. analysis and graph mining techniques, this section provides
This new representation has to flexible enough that it can a very brief overview of related work in this field.
represent business process data in graphs so that graph Graph based representation is used in social network
mining methods can be applied within the business domain. analysis in [11] where the authors discover the collaboration
patterns and organizational structural information from event
B. Formalism and Structural Relations logs (business process executions). Different questions can
Formal representation of business processes has provided be answered from their method like which persons/groups
various benefits for simulation purposes, non-ambiguous be- are working together to accomplish a certain task in busi-
haviour and conflict resolution, and for other analytical tech- ness processes? How can the efficiency of employees be
niques [6]. Besides formal representation of business pro- improved? Furthermore, social network analysis can be
cesses in modeling languages, structural relationships should extended to other areas like crime investigations and fraud
also be incorporated. In this way, different characteristics of detection. In [13], authors propose an algorithm to transform
activities and processes can be attached to them, and new the transactional data into graph based representation for
relations and knowledge can be extracted. This relational graph mining. Further investigation of this transformation
knowledge may take many forms from periodic patterns of and application of graph mining methods is still missing.
transactions to complicate structural patterns of interrelated In [3], the authors represent activities of business pro-
transactions. Extracting such knowledge requires the data cesses as the graph for pattern mining and propose two
to be represented in a form that it not only captures the algorithms to determine frequent patterns of activities. On
relational information but supports mining of this data and the basis of their proposed method, authors claim that a
comprehensibility of the resulting knowledge. Therefore, user can predict whether a certain execution of the business
before applying graph mining for knowledge discovery and process will successfully terminate or not. This could be
analyzing business processes, modeling languages should be used to determine which tasks need to be executed to provide
extended with structural information of elements. a benefit to the enterprise. Combining this work with social
network analysis and performance analysis, different ques-
C. Sequence Transformation and Semantics tions can be answered like how should we allocate resources
In enterprises where business processes are supported by to business processes for maximum efficiency. Graph mining
information systems that record lots of transactional data has also been used to determine user navigational behavioral
about the execution of activities, applying graph mining patterns on the web [10], where authors used probabilistic
is time consuming, requires high computational effort, and hypertext grammar and N-Grams to predict the likelihood
faces complexity problems. Similarly, graphical business of a web page to be seen based on previous page visits.
models are complex as well. This demands other represen- Graph mining is also applied to extract knowledge from
tation techniques to discover new relations and knowledge. consumer behavior in market [14] by using active mining
Just applying graph mining methods on business processes technique in which dynamic process evolvement is dealt with
and their data sets, i.e., activities/attributes, is not feasible. spiral model and levels of user interests. The result of the
Therefore, it should be investigated how these methods study shows that graph based representation discovers new
should be applied on such large data sets and how much and useful relations between item sets with respect to differ-
would be the information loss if applied on specific data ent constraints, which was not discovered earlier. Similarly,
sets. IMDB (internet movie database) is also investigated by many
researchers [4] for several purposes by representing movies R EFERENCES
in the graph to discover new relations between movies, [1] D. Hollingsworth, “The workflow reference model,”
for movie recommendations to users, and prediction of the Workflow Management Coalition, January 1995. [Online].
movie about its success or failure based on their attributes Available: http://www.wfmc.org/standards/docs/tc003v11.pdf
and relation with other movies.
[2] D. Jensen and J. Neville, “Data mining in social networks,” in
National Academy of Sciences Symposium on Dynamic Social
VI. D ISCUSSION Network Modeling and Analysis, Washington DC, USA, 2002.

Besides the issues we discussed in the earlier sections, [3] G. Greco, A. Guzzo, and G. Manco, “Mining and reasoning
we can also apply graph mining in business processes on workflows,” IEEE Transactions on Knowledge and Data
Engineering, vol. 17, no. 4, pp. 519–534, 2005.
over selected business process executions. For example, it
is useful if we first detect frequent transactions occurring [4] D. J. Cook and L. B. Holder, Mining Graph Data. Wiley &
in the system and then represent the frequent transaction Sons, 2006.
pattern in form of a graph to discover more knowledge
[5] C. A. Petri, “Kommunikation mit automaten,” Ph.D. disserta-
and the relations with other elements which will save some tion, University of Bonn, Bonn, Germany, 1962, (In German).
computational over-head. It should be noted, that frequent
patterns do not represent business process and potentially [6] W. van der Aalst, “Three good reasons for using a Petri net
useful knowledge is not associated with frequent pattern based workflow management system,” in Proceedings of the
International Working Conference on Information and Pro-
tasks. Therefore, domain expert knowledge is required to
cess Integration in Enterprises, S. Navathe and T. Wakayama,
be include to select activities which are not frequent but Eds., Camebridge, Massachusetts, 1996, pp. 179–201.
important for analysis in relation and knowledge discovery.
Despite the benefits of graph based representation of [7] A. Lodhi, G. Kassem, and C. Rautenstrauch, “Modeling and
analysis of business processes using business objects,” in
data, we should be careful in applying graphs on different
Proceedings of The 2nd IEEE International Conference on
applications has to be carefully considered because for Computer, Control, and Communication, Karachi, Pakistan,
some problems its techniques are NP-hard problems. It is Feb. 2009.
not a good idea to use graph data for the problem of
finding whether company’s best practice is followed or not?, [8] W. van der Aalst and K. van Hee, Workflow Management:
Models, Methods, and Systems. MIT Press, 2002.
instead string matching methods are more efficient. The
maximum common subgraph can be used to find similarities [9] “The Internet Movies Database,” http://www.imdb.com.
in executions and patterns. Although this is NP-complete
problem but there are optimal and sub-optimal algorithms [10] J. Borges and M. Levene, “Data mining of user navigation
patterns,” in WEBKDD ’99: Revised Papers from the Interna-
as discussed in [15] where a comparison of algorithms is tional Workshop on Web Usage Analysis and User Profiling.
presented. Sequence mining techniques can also be consid- London, UK: Springer-Verlag, 2000, pp. 92–111.
ered for such problems.
[11] W. van der Aalst, H. A. Reijers, and M. Song, “Discover-
ing social networks from event logs,” Computer Supported
VII. S UMMARY & O UTLOOK Cooperative Work, vol. 14, no. 6, pp. 549–593, 2005.

We provided a brief overview of graph mining in various [12] G. Kassem, “Application usage mining: Grundlagen und
domains. In this paper, we also discuss possible benefits verfahren,” Ph.D. dissertation, Otto von Guericke Universität
of applying graph mining in the business process domain. Magdeburg, 2007.
Challenges of applying graph mining in business processes [13] S. Pradhan, S. Chakravarthy, and A. Telang, “Modeling
and the need of a new representation for this purpose is relational data as graphs for mining,” in Proceedings of 15th
also discussed. Our future work involves the specification International Conference on Management of Data. Mysore,
of new representation which allows the application of graph India: Computer Society of India, Dec. 2009.
mining methods for business process analysis. Using the new
[14] M. Kuroda, K. Yada, H. Motoda, and T. Washio, “Knowledge
representation and graph mining we will able to answer the discovery from consumer behavior in an alcohol market
questions raised in this paper. by using graph mining technique,” IEICE technical report.
Artificial intelligence and knowledge-based processing, vol.
104, no. 486, pp. 21–26, 2004.
ACKNOWLEDGMENT
[15] D. Conte, P. Foggia, and M. Vento, “Challenging complex-
Azeem Lodhi is supported by a grant from the federal ity of maximum common subgraph detection algorithms: A
state of Saxony-Anhalt in Germany. Veit Koeppen is sup- performance analysis of three algorithms on a wide database
ported by the German Ministry of Education and Science of graphs,” Journal of Graph Algorithms and Applications,
(BMBF), within the VIERforES project no. 01IM08003C. vol. 11, no. 1, pp. 99–143, 2007.