DISCUSSION OF THE BENEFIT POTENTIALS OF PROCESS
MINING FOR E-LEARNING PROCESSES
Marianne Holzhüter, Dirk Frosch-Wilke
University of Applied Sciences Kiel, Sokratesplatz 2, 24149 Kiel, Germany
Salvador Sánchez-Alonso
Universidad de Alcalá, Carretera Madrid-Barcelona, C.P. 28871, Alcalá de Henares, Madrid, Spain
407
Holzhüter M., Frosch-Wilke D. and Sánchez-Alonso S. (2010).
DISCUSSION OF THE BENEFIT POTENTIALS OF PROCESS MINING FOR E-LEARNING PROCESSES.
In Proceedings of the 2nd International Conference on Computer Supported Education, pages 407-411
DOI: 10.5220/0002770104070411
Copyright
c
 SciTePress

Figure 1: Process mining phases control (Bensberg & Coners, 2008, p. 271).
2 PROCESS MINING
Process mining – inter alia – aims at building a
process model in order to describe the behavior
contained in event logs of information systems (de
Medeiros et al., 2005). The event logs are produced
by process-aware information systems (e.g.
Workflow Management Systems). Typically, these
event logs contain information about the
start/completion of process steps together with
related context data (e.g. actors and resources).
Furthermore, process mining is a very broad area
both in terms of applications and techniques.
The goal of process mining is to extract
information (e.g., process models) from event logs.
Process mining requires the data categories
represented in figure 1, which should briefly be
explained. First of all process instances, i.e.
processes which have been executed, need to be
identified. It might be useful to enrich these
instances by process objects. In case the process
instance is some „learning process” e.g. it would
possess the process object „learner“. By means of
process outcomes as well as relating to the process
goals, the process instances can possibly be
evaluated. Furthermore the so-called process owner
(e.g. the teacher), the process materials (e.g.
literature) as well as the process context (e.g. start
and end, level of external control) are of interest.
Metadata can be additionally considered for the
documentation of the process execution.
Explicit process knowledge can be generated
from different available process data sources as
mentioned above. In particular, decision tree
induction methods permit the generation of
descriptive rule sets which are able to predict
process quality. These rule sets can be used as
operational knowledge base to ensure effectivity and
efficiency of process executions. (Grob et al., 2008).
3 FRAMEWORK FOR PROCESS
MINING OF E-LEARNING
PROCESSES
In this section we introduce a novel framework for
Process Mining of e-learning-processes. The main
motivation for this approach consists in the process
improvement by integration of rules as control basis
into appropriate e-learning processes. First of all, it
may be of strong interest whether rule-based control
of e-Learning processes makes sense at all. Amidst
several individual learning methods it is possible to
identify processes with a more or less uniform
structure. Analyzing these processes by means of
process mining, conclusions may be drawn for the
optimization of learning modules and online
learning sessions.
3.1 Data Preparation and Application
of the Method
In order to assess and generate rules it is necessary
to analyze the correlations between the process
characteristics, e.g.
• process object = learner
• process object characteristic = level of
knowledge
and the extent to which the process goal has been
reached, e.g. increase of knowledge level.
CSEDU 2010 - 2nd International Conference on Computer Supported Education
408

Figure 2: Integration approach for a rule-based control of e-learning processes.
To effect that the analyzed correlations are made
accessible as rules, rule based data mining methods
may be useful. These methods let the results obtain
the structure of IF-THEN rules. Decision trees
qualify as an appropriate method since the training
period remains short and the results produced
dispose of high comprehensibility. Decision trees
generate classification models out of the set of pre-
classified objects in order to describe the classes as
well as to forecast new objects (Grob et al., 2008,
pp. 270).
Effective and ineffective process instances are
screened. The represented method contributes to
process improvement by deriving a generalized
action statement from a created rule (see figure 2).
For the purpose of improving processes by
integrating rules as a control basis into appropriate
e-learning processes, descriptive process knowledge
is transformed into a normative action statement. An
assessment instruction to analyze the extent, to
which the rule conditions have been met, facilitates
the use of the rule as control element. The purpose
of the assessment is to avoid an ineffective learning
process execution. In case of the necessity to modify
the process flow a learning designer may reconfigure
the manipulable process attributes.
After having identified an effective influence of
variables on the process flow via statistical methods
and having analyzed the correlation between
characteristics and the dependent variables, the
variables need to be approved by a specialist
regarding the respective processes.
In order to use the classification model in a
production rule system within an LMS the
prognostic validity needs to be ensured. This may be
affected by estimating the cross-validity. Ideally,
these kind of approaches are complemented by the
inspection of a process expert. A further important
criteria for prognostic validity is the data quality.
Hereto the different data categories act as toehold.
The following data quality criteria need to be
considered:
• disposability, integrity and consistency of the
respective database
• integration ability in terms of logic (data can
be unified in an overall relational schema)
• appropriate timely reach.
3.2 Selection Model for Appropriate
e-Learning Processes
This section deals with the identification of e-
learning process characteristics for which our
process mining framework can be used. Grob et al.
(2008) have built a model selecting business process
characteristics which facilitate process mining
procedures. This model can be applied to e-learning
processes as well (see figure 3).
The more structured a process the better it is
suited for process mining. Structured Processes can
mainly be found where student performance
requirements to solve a task can clearly be defined
(such as mathematical problems) in contrast to less
definable requirements regarding tasks to which

DISCUSSION OF THE BENEFIT POTENTIALS OF PROCESS MINING FOR E-LEARNING PROCESSES
409

Figure 3: Selection model for choice of e-learning processes for rule based-control.
many possible solutions exist (such as an essay on a
broadly defined topic). Processes with possibly high
contribution to the e-learning goals (e.g. to gain a
higher knowledge level or the reduction of learning
time) as well as high execution frequency are of
special interest regarding process mining. From
these kind of processes, such processes will be
selected, which do not sufficiently reach their
effectiveness (e.g. gain of higher knowledge level)
and efficiency (e.g. reduction of learning time).
Process controlling provides means to identify
improvement potentials for such deficient processes,
which can be used to reconfigure the respective
learning processes.
But, the disadvantage of this method as an a-
posterio method is that the process has already been
executed before evaluation and reaction.
Furthermore certain weaknesses may not be
discovered this way. These two aspects may militate
for process mining.
4 IMPLEMENTATION
In this section we introduce an implementation
model of our process mining framework (see figure
4), which integrates the rules defined by the process
mining system into a Learning Management System.
Cesarini et al. (2004) provide a web-based approach
which we combine with the process mining aspect.
In this implementation model the process mining
system hands over its models via Predictive Model
Markup Language (PMML)-interface to the rule
management system. The rule management system
needs to be able to integrate and interpret these
models as well as to access the necessary data. The
models are leveraged at the point of time they are
evaluated in the course of the process runtime by the
process leading systems. These systems dispose of
interfaces towards the respectively valid process
model and facilitate the determination of actual
parameters of the attributes relevant to the forecast
of the process instance to be executed. The e-
learning application produces and makes use of the
operational data sources. The application is
controlled by the rules generated from the models.
The web server enables dynamic interactions in
the web, whereas the users – the teacher as well as
the student – profit from the possibility to change the
particular data in the database via HTML-pages.
In the course of creating learning objects and
tutoring, teachers profit from information how to
improve the processes involved. A typical procedure
illustrating the benefit of the approach could be the
following:
The effectiveness of e-learning sessions with
different difficulty levels is anticipated by analyzing
historical and actual data. According to the results of
the prognosis, the respective learning objects,
knowledge levels and kinds of external control are
CSEDU 2010 - 2nd International Conference on Computer Supported Education
410

combined in a process model by the teacher. The
time for learning and studying learning objects built
according to this process model can be reduced by
ensuring a higher learning effecivity in contrast to a
scenario which does not employ the process model.
Figure 4: Implementation Model.
5 CONCLUSIONS
The presented approach partly responds to the need
for efficient technical systems as well as innovative
didactical methods to support the knowledge
exchange by enabling the control, analysis and
improvement of already executed e-learning
processes and can even support not yet executed e-
learning processes.
The approach reveals high potentials for use in
companies and educational institutions, e.g. gain of
valuable time and effort eliminating ineffective
learning procedures and accelerating the learning
process of complex topics. Nevertheless, there is still
a need for evaluations in practice. Further studies
could deal with more concrete implementation
scenarios going further into technical detail or a
substantiated cost-benefit-ratio.
To complete the analysis of the potentials of the
approach there is a need to render the application of
process mining in e-learning marketable and ready
for implementation. Furthermore the method can be
analyzed in order to identify possibilities to fill the
gap of missing learning process models and lack of
learning process adaptability.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge comments
received on earlier versions of this paper from
Michael Doktor, Sathya Laufer and Matthias Barz.
REFERENCES
Cesarini, M., Monga, M., & Tedesco, R. (2004). Carrying
on the e-learning process with a workflow
management engine. In Proceedings of the 2004 ACM
symposium on Applied computing (S. 940 - 945).
Nicosia, Cyprus : Association for Computing
Machinery.
Datta, A. (1998). Automating the Discovery of As-Is
Business Process Models: Probabilistic and
Algorithmic Approaches. Information Systems
Research, 9 (3), 275-301.
de Medeiros, A. A., Weijters, T. A., & van der Aalst, W.
(September 2005). Genetic Process Mining: A Basic
Approach and its Challenges. In Proceedings of the
Workshop on Business Process Intelligence. Nancy,
France.
Federal Ministry of Education and Research (Germany)
(2009). bmbf.de – Lifelong Learning. Retrieved
09/09/2009 from bmbf.de:
http://www.bmbf.de/en/411.php.
Franconi, E., & Sattler, U. (1999). A Data Warehouse
Conceptual Data Model for Multidimensional
Aggregation. In Proceedings of the International
Workshop on Design and Management of Data
Warehouses.
Frosch-Wilke, D.; Sànchez-Alonso, S.; Dodero, J.M.
(2008). Evolutionary Design of Collaborative
Learning Processes through Reflective Petri Nets. In
Proceedings of the 8th IEEE International Conference
on Advanced Learning Technologies, Santander
(Spain), pp. 423-427
Grob, H. L., Bensberg, F. & Coners, A. (April 2008).
Regelbasierte Steuerung von Geschäftsprozessen -
Konzeption eines Ansatzes auf Basis des Process
Mining. Die WIRTSCHAFTSINFORMATIK, pp.
268-281.
Helic, D. (2006). Technology-Supported Management of
Collaborative Learning Processes. International
Journal of Learning and Change, Vol. 1, No. 3, pp.
285-298.
Ingvaldsen, J.E., & J.A. Gulla. (2006). Model Based
Business Process Mining. Journal of Information
Systems Management, Vol. 23, No. 1, Special Issue on
Business Intelligence, Auerbach Publications
Palmer, C.-E. (2003). E-Learning - die Revolution des
Lernens gewinnbringend einsetzen. In U. Scheffer, &
M. Charlier (Eds.), E-Learning (pp. 13 ). Stuttgart,
Germany : Klett-Cotta Verlag.
Schönwälder, S. (1997). Portfoliomanagement für
betriebliche Informationssysteme: computergestützter
Ansatz zur partizipativen Einführung und Gestaltung.
Wiesbaden, Germany: Gabler.
DISCUSSION OF THE BENEFIT POTENTIALS OF PROCESS MINING FOR E-LEARNING PROCESSES
411