A
PROCESS MINING APPROACH TO ANALYSE USER BEHAVIOUR
Laura M
˘
arus¸ter, Niels R. Faber, Ren
´
e J. Jorna and Rob J. F. van Haren
Faculty of Economics and Business, University of Groningen, PO Box 800, Groningen, The Netherlands
Keywords:
Modelling user behaviour, decision support systems, process mining, web personalization.
Abstract:
Designing and personalising systems for specific user groups encompasses a lot of effort with respect to
analysing and understanding user behaviour. The goal of our paper is to provide a new methodology for
determining navigational patterns of behaviour of specific user groups. We consider agricultural users as a
specific user group, during the usage of a decision support system supporting cultivar selection - OPTIRas
TM
.
Combining process mining techniques with insights from decision making theories, we provide a method
of analysing logs resulted from usage of decision support systems. For instance, farmers show difficulties
in fulfilling the goal of OPTIRas, while other agricultural users seems to manage better. The results of our
analysis can be used to support the redesign and personalization of decision support systems.
1 INTRODUCTION
As the on-line services and Web-based information
systems proliferate in all domains of activities, it
becomes increasingly important to model user be-
haviour and personalization, so that these systems
will appropriately address user characteristics. Par-
ticular topics are addressed by research in human-
computer interaction (HCI) and user-system inter-
action (USI), such as the discovering of user be-
haviour or navigation styles (Herder and Juvina,
2005; Menasalvas et al., 2003; Balajinah and Ragha-
van, 2001), developing metrics involved in modelling
and assessing web navigation (Juvina and Herder,
2005; Herder, 2002; Spiliopoulou and Pohle, 2001),
cognitive models for improving the redesign of infor-
mation systems (Bollini, 2003; Ernst et al., 2005; Lee
and Lee, 2003). Various methods have been devel-
oped to model web navigation in case of generic users
(see for instance (Mobasher, 2006)).
Furthermore, it becomes increasingly important
to address specific user groups (Song and Shepperd,
2006). By investigating navigational patterns of these
groups, the (re)design of the systems used by specific
user groups can be made more effective. Although
various methods have been developed to model user
behaviour of generic users, no research specifically
targeted, as far as we know, the navigational patterns
of agricultural user groups. Our contribution partic-
ularly focuses on agricultural users as a specific user
group. By analysing agricultural users’ patterns of be-
haviour, we aim to support the redesign of web-based
information systems, illustrated in particular for the
redesign of the decision support system OPTIRas
TM
.
The goal of this work is to illustrate a new method-
ology of analysing user behaviour using process min-
ing techniques (van der Aalst and Weijters, 2004) and
insights from decision making theories. By consider-
ing agricultural users as a specific user group, we in-
vestigate agricultural user’s patterns of behaviour dur-
ing the use of a web-based IT system, namely a de-
cision support systems called OPTIRas
TM
, that aids
farmers in their cultivar selection activities (AGRO-
BIOKON, 2006). The results of the analysis provide
recommendations concerning the redesign of the de-
cision support system’s website in order to address
specific agricultural users’ characteristics. Therefore,
in our analysis we use insights from decision making
theories.
The organisation of our paper is as follow. In Sec-
tion 2 we provide an introduction into decision mak-
ing, decision support system OPTIRas
TM
, and data
collection issues. In Section 3 we determine agricul-
tural user’s patterns of behaviour with process mining
techniques. The conclusions and implications for re-
design are presented in Section 4.
208
M
ˇ
aru¸ster L., R. Faber N., J. Jorna R. and J. F. van Haren R. (2008).
A PROCESS MINING APPROACH TO ANALYSE USER BEHAVIOUR.
In Proceedings of the Fourth International Conference on Web Information Systems and Technologies, pages 208-214
DOI: 10.5220/0001526002080214
Copyright
c
SciTePress
2 AGRICULTURAL USERS AND
DECISION MAKING
2.1 Decision Making and Decision
Support Systems
A well-known decision making model was formu-
lated by Simon (Simon, 1977). He explained the
human decision process using three phases, namely
intelligence, design, and choice. In the intelligence
phase, an individual explores the issue about which
he is making a decision, and determines relevant is-
sues. Subsequently, the individual formulates one or
more alternatives solutions to the recognised (sub)-
decisions in the design phase. Eventually, a final solu-
tion is formed in the choice phase. In this final phase
of the decision process, the partial solutions are eval-
uated against criteria of the outcome that have to be
met. Partial solutions that best meet these criteria are
selected. The selected partial solutions are combined
into the overall decision. Because information that is
used in the three phases is not always complete, the
phases do not linearly follow upon each other. Com-
monly, the decision process is characterised by many
iterations in which additional information about the
issue is collected (intelligence) or more alternatives
are explored (design). Mintzberg refers to a similar
trichotomy: identification, development, and selec-
tion (Mintzberg et al., 1976). Specific decision mak-
ing models have been developed for different kinds of
users. For instance, in case of farmers, decision mak-
ing models have been developed by (Johnson et al.,
1961;
¨
Ohlm
´
er et al., 1998) and (Fountas et al., 2006).
Whenever these models are linear (sequential) or
iterative, they all show a common structure: first, in-
formation is collected about the problem at hand, sec-
ond, different candidate solutions are formed, and fi-
nally, a choice is made. It is evident that a decision
support system has to address in one way or another
the basic phases of the decision making process. In
the following, we investigate in which way the deci-
sion support system OPTIRas
TM
has addressed these
phases.
2.2 OPTIRas
TM
A decision support system (DSS) is perceived as a
computer system that aids people in making a deci-
sion regarding a specific domain (Klein and Methlie,
1995). This aid is provided by the DSS by connecting
to the human decision process (Turban and Aronson,
2001).
OPTIRas
TM
system was designed to target low-
yielding potato farmers, attempting to realise an in-
crease in the yield of farmers. OPTIRas
TM
is a de-
cision support system for cultivar selection that sup-
ports a farmer in selecting cultivars relative to culti-
var characteristics. A variety of properties, relating to
yield, resistance against pests and diseases, and stor-
age, characterise a potato cultivar. Potato Cyst Nema-
todes (PCN), a severe potato disease, causes each year
an average loss of 100-150 euros per hectare per year
in the North East region of The Netherlands. This is
about 10-15% of the net revenues of cropping (before
taxation and investment calculations). Reducing the
PCN infestation level to economic acceptable levels
requires the right combination of sampling data, cul-
tivar, growth frequency, field choice, and nematicide
usage. The goal of the decision support system is to
ensure that farmer gains insight into PCN damage lev-
els and the financial consequences of selected cultivar
and the application of pesticides.
The interface of OPTIRas
TM
DSS consists of
seven main pages: Field ID, PCN History, Reorder
Cultivars, Yield Information, Crop Rotation, Report
and Option. Moreover, there are also pages offering
Help, Details about a specific cultivar, but we do not
use them in our analysis.
Web pages do not have the same function, and this
depends on the purpose of the web site. In our case,
the site’s goal is to provide the farmer with informa-
tion about different yield scenarios, given the cho-
sen factors. Therefore, the following types of pages
can be distinguished: (i) the pages reflecting the site’s
goal, e.g. containing information about the yield, (ii)
the pages that can lead to fulfil the site’s goal, and (iii)
the other pages.
In literature, this categorization is called con-
cept hierarchies, or service-based hierarchies
(Spiliopoulou and Pohle, 2001). Concept hierarchies
are used in market-basket analysis or to study the
segmentation of companies clients. Spiliopoulou
and Pohle developed a service-based concept hier-
archy to determine the success of a web site, which
distinguishes between action pages and target pages
(Spiliopoulou and Pohle, 2001). According to their
definition, “an action page is a page whose invocation
indicates that the user is pursuing the site’s goal. A
target page is a page whose invocation indicates that
the user has achieved the site’s goal”.
As we stated in the previous section, a DSS sys-
tem should support the user in all phases of the de-
cision making process. For instance, in terms of Si-
mon’s decision making phases, ’Action’ pages would
correspond to the Intelligence phase, where an indi-
vidual explores the issues about which he is making
a decision, while ’Target’ pages would correspond to
the Design phase, where candidate solutions are de-
A PROCESS MINING APPROACH TO ANALYSE USER BEHAVIOUR
209
Table 1: Categorization of OPTIRas
TM
pages.
’Action’ pages ’Target’ pages Other pages
Field ID Yield Details
PCN information Options
history Crop Help
Reorder rotations Logon
cultivars Report Help
Save
veloped.
We categorize OPTIRas
TM
pages as presented in
Table 1. In the following, we consider that whenever
a user has visited one of the target pages (Yield infor-
mation, Crop rotations, Report), the user has reached
the goal of the web site.
Based on Simon’s decision making phases, we
proceed with the following mapping:
1. ’Action’ pages corresponds to the Intelligence
phase: the user is supported to explore the issue
(the selection of the cultivar), by specifying Field
characteristics and the PCN history;
2. ’Target’ pages corresponds to the Design phase:
the system provides the user with one or more
possible solutions;
3. the Choice phase does not have any correspon-
dence in the DSS; the final decision is actually
taken by the user outside of the system.
Assessing whether the DSS is “really” supporting
the targeted users, or in other words, whether the site
goal is fulfilled, we expect that both decision mak-
ing phases, Intelligence and Design, should be per-
formed, to the same extend. This translates into visi-
tations of both ’Action’ and ’Target’ pages, to a com-
parable extend. In Section 3 we will investigate how
the navigational patterns look like, e.g. how ’Action’
and ’Target’ types of pages are visited.
2.3 Data Collection
According to Herder and Juvina (Herder and Juvina,
2005), user’s navigation behaviour can be modelled
using syntactic information (“e.g. which links are fol-
lowed, what does the navigation graph look like, what
is the time that users spent on each page”), semantic
information (“i.e. what is the meaning of the infor-
mation that the user encountered during navigation”),
and pragmatic information (“i.e. what is the user us-
ing the information for, what are the user’s goals and
tasks”). We choose to record per session syntactic in-
formation, e.g. information concerning the movement
from one source’ page to a destination’ page (time
Table 2: An example of the log file.
Time stamp FromPage ToPage
2004-12-22 22:13:29 field pcn
2004-12-22 22:13:35 pcn order
2004-12-22 22:14:00 order yield
2004-12-22 22:14:26 yield crop
2004-12-22 22:16:16 crop yield
2004-12-22 22:16:25 yield crop
2004-12-22 22:16:53 crop details
2004-12-22 22:17:54 details allcultivars
stamp, name of source page, and name of destination
page, see Table 2).
Table 2 shows an example of an OPTIRas
TM
se-
quence of navigation, where the user sequentially ac-
cesses 7 types of pages. The log file (Table 2) pro-
vides us syntactic information, namely information
needed to build navigation graphs (see Section 3).
Analysing users as a homogeneous group is not
a suitable basis for re-designing decision support sys-
tems that target heterogeneous end users. OPTIRas
TM
system was designed for farmers. However, a
larger group of agricultural users is actually using
OPTIRas
TM
, that can be differentiated into three tar-
get groups, namely farmers (called also growers
1
),
extension workers, and scientists. Extension work-
ers are people with a degree in agricultural sciences,
who help growers in their practical work. Scientists
also use OPTIRas
TM
, but rather for experimental pur-
poses.
Agricultural users can login in OPTIRas
TM
reg-
istering with their e-mails, but they can register also
with an anonymous ID. The system provides this op-
portunity for the sake of user-friendliness, but also to
protect users’ privacy. The anonymous ID takes into
account the IP number. In this way advantages for
the users are provided, but it implies disadvantages
for the analysis. A user may login the first time with
his/her e-mail, the second time anonymously using
computer A, and the third time again anonymously
using computer B; in the analysis, these three ses-
sions will count as sessions belonging to three dis-
tinct users, and we cannot see anymore how a user
eventually changes his behaviour in time. Moreover,
users can login by specifying their role, i.e. grower,
scientist, extension worker, or other. OPTIRas
TM
is
expected to be used in particular within two peak pe-
riods (i) November/December and February (period
for purchasing crops, expected to be the peak period)
and (ii) April and August/September (period when
1
In this article, the terms ‘farmer’ and ‘grower’ are used
interchangeably.
WEBIST 2008 - International Conference on Web Information Systems and Technologies
210
the yield is actually obtained). Farmers may com-
pare the obtained yield with the advice given earlier
by OPTIRas
TM
.
We have inspected only the log files starting
from the 18th of December 2004 (the date when
OPTIRas
TM
became on-line) and stopped at with the
30th of May 2006. Thus we have two peak periods:
(i) December 2004 - February 2005 and (ii) Novem-
ber 2005 - February 2006. In total, 763 user ses-
sions belonging to 501 individual users (see discus-
sion above about distinct users) have been logged and
included in the analyses.
3 MINING NAVIGATIONAL
BEHAVIOUR WITH PROCESS
MINING TECHNIQUES
Developing navigation patterns and performing web
personalization is often done by using data mining
techniques such as Clustering, Classification, Asso-
ciation Rule Discovery, Sequential pattern Discov-
ery, Markov models, hidden (latent) variable models,
and hybrid models (Mobasher, 2006; Herder, 2002;
Juvina and Herder, 2005; Chang et al., 2006). Mostly
of these techniques assume web navigation as a se-
quential process. However, the decision making pro-
cess exhibits a distributed nature (Holsapple, 1988),
which should be captured by the web-mining tech-
nique.
Process mining techniques allow for extracting in-
formation from event logs and can be used to dis-
cover models describing processes, organisations, and
products (van der Aalst et al., 2003). Because these
techniques are able to model parallel events, moni-
tor deviations (e.g., comparing the observed events
with predefined models or business rules), and rely
on robust mathematical formalism, they are a good
candidate for providing insights into navigational be-
haviour for decision making processes. However,
the choice of the most appropriate process mining
algorithm for a certain problem is not straightfor-
ward. A suitable algorithm is the heuristic mining,
implemented in the ProM framework. ProM offers
a wide variety of process mining techniques (avail-
able at www.processmining.org) (ProM, 2007). We
choose heuristic mining because for its robustness for
noise and exceptions (Weijters and Aalst, 2003). The
heuristic mining is based on the frequency of patterns,
therefore it is possible to focus on the main behaviour
in the event log.
We provide insights into two kinds of navigational
patterns: (i) navigation patterns of all agricultural
users show, and (ii) navigational patterns of differen-
tiated target group (e.g. growers, extension workers,
scientists).
3.1 Mining the Navigational Pattern of
All Users
This section concentrates on the navigational patterns
exhibited by all users. In Figure 1 the result of ap-
plying the ProM heuristic mining plug-in is shown.
The focus is to grasp the patterns of navigation from
one page to another (we use the FromPage field of the
log file, see Table 2). The file consists on 763 user
sessions, belonging to 501 individual users. The rect-
angles refer to transitions or activities, in our case to
page names, such as Field, Order, etc. There are two
special activity names, ArtificialStartTask and Artifi-
cialEndTask that refer to a generic start or end page.
The term ’complete’ refers to a finished action (in
terms of a finite-state machine, an activity can be in
the states ’new’, sent’, ’active’, suspended’, ’com-
plete’, etc.). The number inside the rectangle shows
how many times a page has been invoked. The arcs
between two activities A and B are associated with
two numbers
2
: (i) a number between 0 and 1, which
represents the causality/dependency measure between
A and B, and (ii) an integer number that represents
the frequency of occurrences of A and B next to each
other. For instance in Figure 1, Field page is directly
followed 553 times by the Pcn page, with a depen-
dency measure of 0.742, and 374 times by Artificial-
EndTask, with a dependency measure of 0.996.
The complicated picture from Figure 1 can be in-
terpreted as follows. We can determine “the most
common pattern” which consists of ArtificialStart-
Task Field (758), Field Pcn (553), Pcn Yield
(530), and Yield ArtificialEndTask (221). The se-
quence Field Pcn Yield is actually the “pre-
scribed” order of pages in OPTIRas
TM
, e.g. the order
in which pages are presented in OPTIRas
TM
.
We notice the reversed link Yield Pcn, which
may suggest that users are changing PCN values to
observe the impact on calculating the yield (the higher
the values of PCN, the smaller the yield). Notice-
able is also the high frequent self-recurrent link to
Yield(655).
The interpretation of these findings is that farm-
ers in general use the prescribed sequence of page in-
vocation. The directed graphs from Figure 1 reveal
that invocation of ’Action’ pages are predominating.
In Figure 1, the sum of incoming arcs from ’Action’
pages (Field, Pcn, Order) to activity ArtificialEnd-
2
For more details, see (Weijters et al., 2006).
A PROCESS MINING APPROACH TO ANALYSE USER BEHAVIOUR
211
allcultivars
(complete)
99
details
(complete)
279
options
(complete)
64
crop
(complete)
385
yield
(complete)
1207
pcn
(complete)
889
ArtificialEndTask
(complete)
763
order
(complete)
1097
field
(complete)
1494
ArtificialStartTask
(complete)
763
0.981
118
0.8
139
0.998
555
0.998
655
0.996
374
0.75
19
0.742
553
0.979
99
0.991
255
0.875
7
0.923
24
0.5
4
0.993
221
0.957
154
0.999
758
0.958
41
0.983
102
0.938
302
0.938
530
0.192
57
0.995
709
0.992
342
0.958
46
0.909
16
0.989
162
Figure 1: The mined behaviour corresponding to all agri-
cultural users.
Task is 374+162+342=878. This value is much larger
than the sum of incoming arcs from ’Target’ pages
(Yield, Crop) to the activity ArtificialEndTask, which
is 221+99=320.
The first conclusion is that pages from the cat-
egory ’Target’ are visualized significantly less than
’Action’ pages. This is a disappointing result, given
the fact that the main goal of OPTIRas
TM
is to pro-
vide detailed information about yield. Second, after
each page a significant number of sessions stop (in
Figure 1, 374 after page Field, 162 after page Pcn).
However, we also see that whenever users do not ’give
up’ in early stages, and visualize ’Target’ pages, they
show a relevant activity. They recall repeatedly the
Yield page, and they revisit pages which are not di-
rectly linked in the prescribed order (the reverse link
Yield Pcn). This fact illustrates that whenever
users reach a ’Target’ page, their interest may rise.
The Details page is visited unexpectedly often: 279
times, and even forms a path containing only one page
(e.g. the path ArtificialStartTask, Details and Artifi-
cialEndTask (see Figure 1). This suggests that users
are interested in information about different cultivars.
3.2 Mining the Navigational Patterns of
differentiated Target Groups
In this section we investigate the patterns of behaviour
of different target groups. These target groups are
growers, extension workers, and scientists. Whenever
a user logs in into OPTIRas
TM
, he/she has to specify
his/her role, e.g. growers, extension worker, scientist
or others. The distribution of sessions is as follow-
ing: from all 763 sessions, 535 are sessions belonging
to growers, 59 are sessions coming from extension
workers and 72 are results of scientists usage. The
rest of 97 sessions which are not labelled, are left out
from the analysis.
The run of the heuristic algorithm (see Figure 2a.)
depicting the navigation pattern of farmers provide
more insights into the most generic behaviour: there
seems to be differences between all agricultural users
(Figure 1) and farmers (Figure 2a.) For instance, in
case of farmers, the most used path is constituted by
Field and Pcn pages, without the Yield page. This fact
could be interpreted as the majority of farmers do not
reach the Yield page, and thus we may doubt whether
the DSS fulfils its goal. Also, the navigational pattern
of farmers does not contain loops (see in Figure 2a.
the loop between Pcn and Yield), except from self-
loops.
Inspecting further the navigation patterns of the
extension workers, we see in Figure 2b. that the
most common behaviour is to stop after visiting the
Field page. The other path starting with the Pcn page
is less easy to be interpreted (without any incoming
arc); however, we can note the loop between Pcn, Or-
der and Yield, which illustrates a rather complex be-
haviour.
The behaviour of scientists is the most complex
(see Figure 2c.): there are two loops, e.g. (i) Field
and Pcn and (ii) Order and Yield. This illustrates the
fact that scientists, instead of only following the pre-
scribed behaviour, are checking the effects of chang-
ing value parameters. Also, most of users of this
group visit ’Target’ pages (Yield and Crop), which
may suggest that they like to explore DSS possibili-
ties.
4 CONCLUSIONS
In this article we provided a new methodology for
determining navigational behaviour of agricultural
users, by combining process mining techniques and
insights from decision making theories. To our
knowledge, this combination, together with apply-
ing the approach of process mining in case of deci-
sion support systems supporting farming activities are
new. This methodology can be used to redesign the
decision support system, by addressing the character-
istics of agricultural users.
We analysed logs resulting from a decision sup-
port system called OPTIRas
TM
. With respect to all
agricultural users, they use the prescribed order of
pages, e.g., Field, Pcn, Order, Yield. We discovered
that often users visualise just the first page, and then
WEBIST 2008 - International Conference on Web Information Systems and Technologies
212
allcultivars
(complete)
92
details
(complete)
235
options
(complete)
57
crop
(complete)
321
yield
(complete)
980
pcn
(complete)
703
ArtificialEndTask
(complete)
535
order
(complete)
932
field
(complete)
1099
ArtificialStartTask
(complete)
535
0.512
373
0.998
529
0.933
276
0.178
49
0.667
2
0.992
221
0.984
297
0.875
9
0.741
425
0.917
20
0.995
397
0.947
29
0.974
67
0.99
112
0.98
90
0.75
112
0.997
596
0.997
430
0.727
24
0.993
217
0.975
97
0.99
213
0.957
43
crop
(complete)
32
details
(complete)
12
options
(complete)
2
yield
(complete)
77
pcn
(complete)
60
ArtificialEndTask
(complete)
59
order
(complete)
84
field
(complete)
105
ArtificialStartTask
(complete)
59
0.346
16
0.8
8
0.983
58
0.5
1
0.5
1
0.667
3
0.933
16
0.929
20
0.97
47
0.857
4
0.923
23
0.672
45
0.955
45
0.875
16
0.5
50
0.857
1
crop
(complete)
45
details
(complete)
19
ArtificialEndTask
(complete)
72
options
(complete)
11
yield
(complete)
110
order
(complete)
93
pcn
(complete)
107
field
(complete)
129
ArtificialStartTask
(complete)
72
0.833
3
0.933
23
0.986
71
0.958
65
0.646
59
0.95
38
0.929
13
0.857
8
0.5
1
0.967
35
0.341
20
0.933
65
0.947
24
0.958
24
0.933
28
0.8
8
0.667
2
a. growers b. extension workers c. scientists
Figure 2: The mined behaviour of farmers (a.), extension workers (b.) and scientists (c.).
leave the system. With respect to goal fulfilling, the
invocation of ’Action’ pages predominates, while the
invocation of ’Target’ pages have a smaller proportion
than expected. This corroborates with our assump-
tion that users of OPTIRas
TM
spend time in the Intel-
ligence phase of the decision making process, explor-
ing the alternatives in the domain of cultivar selection.
This shows that users do not realise fast enough what
the goal of OPTIRas
TM
is, and what kind of advan-
tages they may have when they use this system. This
leads to a lack of interest, which implies that users
stop using the system. The fact that often the Details
page is visited in the beginning of farmer’s sessions
suggests that in some cases, users are more interested
in information about different cultivars, rather than
cultivar selection based on Yield or Crop figures.
By analysing logs of different target groups, we
found that the most common behaviour of farm-
ers (growers) shows difficulties in fulfilling the goal
of OPTIRas
TM
, e.g. visiting ’Target’ pages, while
the other two target groups, extension workers and
scientists seems to manage better. Given the fact
that OPTIRas
TM
was especially developed to support
farmers (growers), it is a clear sign that redesign ac-
tions are needed.
This methodology can be used to support the re-
design of DSSs in order to address specific agricul-
tural user’s characteristics. First, OPTIRas
TM
should
better support the decision making process, i.e. by
letting users gather information about the various top-
ics involved in cultivar selection, instead of present-
ing itself as an instrument to make optimal choices.
Second, we suggest that in the very first page of
OPTIRas
TM
, a user should be confronted with the
goal of the system. This should be very explicitly and
clearly stated, perhaps illustrated with a short exam-
ple. Third, depending on the target group, hints for
the following steps should be given: e.g. in case of
growers, what path should be followed to get the yield
overview. For this, the use of a sitemap is regarded to
be appropriate.
In future research we plan to investigate in more
detail the navigation behaviour of different farmer
groups, e.g. top-farmers, quality, quantity and nor-
mal farmers (for farmer categories, see (Faber et al.,
2006)), especially to relate their navigational patterns
with their preferred learning styles. Information about
the behavioural patterns of various agricultural user
groups may enhance the successful design and use of
DSSs. Second, we plan to let these users participate
in the redesigning process and to make OPTIRas
TM
more interactive. Third, we intend to incorporate in
our analysis also semantic and pragmatic information.
As a final remark, we would like to state that, al-
though the present research focused on decision sup-
port systems and agricultural users, the methodology
presented in this paper can be used in case of any in-
formation system enhanced with logging functional-
A PROCESS MINING APPROACH TO ANALYSE USER BEHAVIOUR
213
ity, and with groups of users that, based on some cri-
terion, can be differentiated into subgroups.
ACKNOWLEDGEMENTS
We thank Iain Milne from Scottish Crop Re-
search Institute for enabling the logging function of
OPTIRas
TM
. AGROBIOKON is supported by the
Northern Netherlands Provinces (SNN), the Product
Board for Arable Products (HPA), and AVEBE.
REFERENCES
AGROBIOKON (2006). Optiras - rassenkeuze ad-
viessysteem voor zetmeelaardappelen. http:// opti-
ras.agrobiokon.eu/.
Balajinah, B. and Raghavan, S. (2001). Intrusion detection
through learning behavior model. Computer Commu-
nication, 24:1202–1212.
Bollini, L. (2003). Web interface design based on cognitive
maps: Generative dynamics in information architec-
ture. The 6th Generative Art Conference - GA2003.
Chang, S. E., Changchien, S. W., and Huang, R.-H. (2006).
Assessing users’ product-specific knowledge for per-
sonalization in electronic commerce. Expert systems
with Applications, 30:682–693.
Ernst, N. A., Story, M.-A., and Allen, P. (2005). Cognitive
support for ontology modeling. International Journal
of Human-computer Studies, 62(5):553–577.
Faber, N., Jorna, R., van Haren, R., and Maruster, L. (2006).
Knowledge and knoweldge use for sustainable inno-
vation: the case of starch potato production. In Pro-
ceedings of ISMICK’06, Stellenbosch, South Africa.
Fountas, S., Wulfsohn, D., Blackmore, B., Jacobsen, H.,
and Pederson, S. (2006). A model of decision-making
and informations flows for infomation-intensive agri-
culture. Agricultural Systems, 87:192–210.
Herder, E. (2002). Metrics for the adaptation of site struc-
ture. In Proc. of the German Workshop on Adaptivity
and User Modeling in Interactive Systems ABIS’02,
Hannover, pages 22–26.
Herder, E. and Juvina, I. (2005). Discovery of individual
user navigation styles. In Proc. of the Workshop on
Individual Differences - Adaptive Hypermedia, Eind-
hoven 2005.
Holsapple, C., W. A. (1988). Distributed Decision Making:
A Research Agenda. ACM SIGOIS Bullentin.
Johnson, G., Halter, A., Jensen, H., and Thomas, D.
(1961). A Study of Managerial Processes of Midwest-
ern Farmers. Iowa State Press, Ames, Iowa.
Juvina, I. and Herder, E. (2005). Individual differences and
behavioral aspects involved in modeling web naviga-
tion. In Proc. of User-centered interaction paradigms,
volume LNCS 3196, pages 77–95. Springer.
Klein, M. and Methlie, L. (1995). Knowledge-based deci-
sion support systems: with applications in business.
John Wiley & Sons, Chichester.
Lee, K. and Lee, S. (2003). A cognitive map simulation ap-
proach to adjusting the design factors of the electronic
commerce web sites. Expert Systems with Applica-
tions, 24:1–11.
Menasalvas, E., Millan, S., Perez, M., Hochsztain, E., Rob-
les, V., Marban, O., a, J. P., and Tasistro, A. (2003).
Beyond user clicks: an algoritm and an agent-based
arhitecture to discover user behavior. In Proceedings
of ECML/PKDD.
Mintzberg, H., Raisingham, D., and Thor
ˆ
et, A. (1976). The
structure of ’unstructured’ decision process. Adminis-
trative Science Quarterly, 21:246–275.
Mobasher, B. (2006). The Adaptive Web: Methods and
Strategies of Web Personalization, chapter Data Min-
ing for Personalization. Berlin: Springer-Verlag.
¨
Ohlm
´
er, B., Olson, K., and Brehmer, B. (1998). Under-
standing farmers’ decision making process and im-
proving managerial assistance. Agricultural Eco-
nomics, 18:273–290.
ProM (2007). The ProM framework. http://
www.processmining.org.
Simon, H. (1977). The new science of management deci-
sion. Prentice-Hall, Inc., Englewood Cliffs, New Jer-
sey, revised edition.
Song, Q. and Shepperd, M. (2006). Mining web brows-
ing patterns for e-commerce. Computers in Industry,
57:622–630.
Spiliopoulou, M. and Pohle, C. (2001). Data Mining to
Measure and Improve the Success of web Sites. Data
Mining and Knowledge Discovery, 5(1-2):85–114.
Turban, E. and Aronson, J. (2001). Decision Support Sys-
tems and Intelligent Systems. Prentice-Hall, Upper
Saddle River, N.J.
van der Aalst, W., van Dongen, B., Herbst, J., Maruster,
L., Schimm, G., and Weijters, A. (2003). Workflow
Mining: A Survey of Issues and Approaches. Data
and Knowledge Engineering, 47(2):237–267.
van der Aalst, W. and Weijters, A. (2004). Process Mining,
volume 53 of Special Issue in Computers in Industry.
Elsevier Science Publishers,Amsterdam.
Weijters, A. and Aalst, W. (2003). Rediscovering workflow
models from event-based data using Little Thumb. In-
tegrated Computer-Aided Engineering, 10:151–162.
Weijters, A., van der Aalst, W., and de Medeiros, A. A.
(2006). Process mining with the heuristics miner-
algorithm. BETA Working Paper Series WP 166,
Eindhoven University of Technology.
WEBIST 2008 - International Conference on Web Information Systems and Technologies
214