SOCIALIZATION OF WORK PRACTICE THROUGH BUSINESS
PROCESS ANALYSIS
Mukhammad Andri Setiawan and Shazia Sadiq
School of Information Technology & Electrical Engineering, The University of Queensland, Australia
Keywords: Experience Driven, Business Process Analysis, Work Practice, Socialization.
Abstract: In today’s competitive business era, having the best practice business process is fundamental to the success
of an organisation. Best practice reference models are generally created by experts in the domain, but often
the best practice can be implicitly derived from the work practices of actual workers within the organization.
In this paper, we propose to utilize the experiences and knowledge of previous business process users to
inform and improve the current practices, thereby bringing about a socialization of work practice. We have
developed a recommendation system to assist users to select the best practices of previous users through an
analysis of business process execution logs. Recommendations are generated based on multi criteria
analysis applied to the accumulated process data and the proposed approach is capable of extracting
meaningful recommendations from large data sets in an efficient way.
1 INTRODUCTION
In a very competitive business era, having the best
practice business process is the basis of the success
of an organisation. A valuable and often overlooked
source of best practice is the experiences and
knowledge of individuals who perform various
activities within the business process. This
knowledge constitutes the corporate skill base and is
found in the experiences and practices of
individuals, who are domain experts in a particular
aspect of the overall operations.
Furthermore, business processes often face a
dynamic environment which forces them to have the
characteristic of ad-hocism in order to tailor to
circumstances of individual process cases or
instances. This creates what so called business
process variants (Lu et al., 2009). The variants
include the creativity and individualism of the
knowledge worker, which is generally only tacitly
available. Each variant has the same goal but by
having different approaches, it may have different
time needed, different task set and/or sequence and
most likely different cost.
A traditional Business Process Management
(BPM) System is not generally capable to select best
processes since all instances follow the same process
model, and thus there is hardly any variance that can
reflect individual/unique approaches. However,
some complementary work can be found within the
BPM community that long recognized the need to
provide flexible business process. Some works show
how flexible business process can be achieved by
executing variance with certain selection strategies
(e.g. lowest cost, cycle time) as mentioned by
Vanderfeesten et al. (2008) and Lu & Sadiq (2008).
It is expected by having the flexible business
process, an organisation can rapidly adjust their
business process to suit the environment. But,
having a flexible process is not always a solution to
achieve the most efficient practice for the
organisation. In fact, the more flexible the system,
the more a (inexperienced) user may struggle to find
the best approach to address a particular case. These
users are required to have deep knowledge of the
process they are working on (Helen et al., 2008,
Schonenberg et al., 2008).
In this paper we will present an approach to
providing assistance to users which allows them to
select the best process variants been done by
previous (arguably experienced) users. Rather than
forcing users to make design decisions, we will use
the existing knowledge in the BPM system (through
execution logs) and select the variants that best meet
the required criteria. The approach will guide the
future user to get the benefit from user perspective,
as well as organisational perspective. The remainder
of this paper is organized as follows.
165
Andri Setiawan M. and Sadiq S. (2010).
SOCIALIZATION OF WORK PRACTICE THROUGH BUSINESS PROCESS ANALYSIS.
In Proceedings of the 12th International Conference on Enterprise Information Systems - Information Systems Analysis and Specification, pages
165-170
DOI: 10.5220/0002871801650170
Copyright
c
 SciTePress

In Section 2, we specify the problem background
and related work. Then, in Section 3 we present the
experience driven recommendation service including
the details of the analysis. Finally, in the last section,
Section 4, we provide a summary evaluation of the
proposed approach and conclusions drawn from this
work.
2 PROBLEM BACKGROUND
AND RELATED WORK
Today, many organisations have implemented BPM
system in managing, monitoring, controlling,
analysing and optimizing their business process
(Aalst et al., 2003). BPM allows organisations to
design business process models, execute process
instances in accordance with the models, enable
users/applications to access task lists and execute
task operations (Yujie et al., 2004). The system is
meant to implement business strategies phases by
modelling, developing, deploying, and managing
business process so that organisations can have the
benefit of innovation and optimization.
As the phases work in cycle, the overall business
process can be improved by revamping various
components of the cycle. Business process analysis
is the key means to this end. In business process
analysis, the business process activities are analysed,
mapped, etc (Biazzo, 2000) with the goal of
continuously improving the process and related
practices to create a better quality of the business
process. The efficiency, cost, completeness, and the
confidence level of business process are key to
quality definition. Business process analysis is an
essential prerequisite for organisational change and
is needed to create either gradual change or
incremental change (Biazzo, 2000).
A number of contributions have been made in the
general area of business process analysis. An audit
trail of a BPM system is an example on how it can
be used to find models which describe the process,
organisations, and products. An audit trail contains
information about the events i.e. who executes the
process, what time was taken, which activity and
process instance, etc. All information can then be
analysed in many areas as explained by Bozkaya
(2009), such as to measure the performance of
processes (Hornix, 2007), process discovery
(Günther and Aalst, 2007), process conformance
(Rozinat and Aalst, 2005), and social networks
(Aalst et al., 2005a). Some research also provides
the process model as the output of process discovery
(Aalst et al., 2004) where from a workflow log, a
process model is constructed partially or fully
developed, which later can be used for specific
purposes, such as discovering patterns of execution
(Dubouloz and Toklu, 2005), analyse variance of
process model (Tsai and Chen, 2009). Similarly,
interaction patterns can also be learnt to cover what
social networks exist (Aalst and Song, 2004).
At the same time, business processes are quite
often characterized with variance. Variance itself in
business process execution is the outcome of many
situations, Lu and Sadiq (2008) give examples, such
as the disconnection between documented models
and business operations, the active change and
exception handling, flexible and ad-hoc
requirements, and collaborative and/or knowledge
intensive work. Various work practices are present
in real world, and it incorporates personal
approaches and knowledge of workers of benefit to
the organisation (Lu and Sadiq, 2006). This
especially happens when the organisation have
flexible ways to complete tasks.
Figure 1 presents example process models of
different process variants. The process models show
how same tasks are processed differently in different
variants. Tasks here can represent e.g tests to
diagnose a reported fault in telecommunication, or
investigative activities of an insurance claim etc.
The coordinative nodes Begin, End, Fork,
Synchronizer, Choice, and Merge shown in the
figure are assumed to have typical semantics
(WFMC, 1999).
Consider insurance claim process in a health care
industry as an example. During insurance claiming
process, the same goal could be achieved in multiple
ways. As customers vary, e.g regular or VIP
customer, with single or family type insurance, and
many more criteria, the approaches and steps taken
to complete the claim process will be treated
differently. In addition, a claims officer will have
different approach to handle the claims within the
constraints of the insurance policy. Thus leading to
the creation of process variants.
In the previous work, Lu and Sadiq (2008)
present a facility for discovery of preferred variants
through effective search and retrieval based on the
notion of process similarity, where multiple aspects
of the process variants are compared according to
specific query requirements.
The useful feature of the approach developed
was the ability to provide a quantitative measure for
the similarity between process variants. However,
the problem is much more complex. The value of
a process variant can only be realized if it provides
ICEIS 2010 - 12th International Conference on Enterprise Information Systems
166

Figure 1: Example process model of process variants.
relevant and meaningful recommendations for others
who are working in a similar scenario – so called
socialization of work practice.
We identify the socialization of work practice as
simply providing the best practices which have been
done by previous workers to potential future users.
In this paper, we will investigate and subsequently
define the criteria for identification and ranking of
precedents that working communities may utilize.
The proposed experience driven recommendation
service (see section 3) has the potential to achieve an
effective solution for all stakeholders. This is
achieved by utilizing the experiences built by expert
claim processing officers. We assume that these
experiences are manifested in process variants (as
proposed by Lu and Sadiq (2008), thus process
variants have the capacity to externalize the
previously tacit knowledge found in individual
experiences.
It is also important to understand the role of
experience in organizational learning as experience
is a fundamental notion of our work. Experience has
a potential value to enrich the knowledge sharing
and knowledge transfer between learners, as learning
process is the process of knowledge transfer between
tacit and explicit knowledge.
In our research, we propose a recommendation
service based on the experience of business process
users. The recommendation service is initiated by
the system by analysing the history of executed
activities. The recommendation service then supplies
the system with a recommendation result which is
the best practice among various practices of different
workers.
3 EXPERIENCE DRIVEN
RECOMMENDATION
SERVICE
In this section, we present our approach to
socialization of work practice through a Experience
Driven Recommendation Service (EDRS). The
EDRS is an add-on to the BPM system that provides
the capability to identify and rank previous process
variants against a set of criteria, and thereby assist
current users to deal with specific process cases in
the best possible way as demonstrated by the
practices of previous users.
3.1 EDRS Architecture
Basically, EDRS is an add-on to the BPM system to
provide the information on best past practice. The
experience driven recommendation service relies on
previously recorded execution logs which contain
information on various aspects of the process
including execution times, costs and resources used
etc.
We assume a typical schema of the execution log
(Grigori et al., 2004) typically structured as a set of
timed events. Thus, further information such as the
tasks and structure of various variants, number of
times a variant has been used etc. can also be
extracted.
The EDRS consists of three main components:
the process mining component, experience driven
analysis component, and EDRS recommender
component. The process mining component (similar
to those proposed in Aalst et al. (2005b) is
SOCIALIZATION OF WORK PRACTICE THROUGH BUSINESS PROCESS ANALYSIS
167

responsible for analysing the execution logs and
producing the process models of various variants.
Figure 2: EDRS Architecture.
The experience driven analysis component creates
information on process popularity (number of
instances against a given variant) and respective
weight (where weight is a quantification of time and
resources of a particular process instance) and store
it in the so-called Work Practice Database. The
EDRS recommender will produce a sorted list of
selected processes (variants) from the work practice
database.
In summary, the recommendation setting used in
the EDRS architecture will be based on multi criteria
system which will try to calculate based on the
process structure (task sequence), the time taken,
and the cost of the process instances.
3.2 Analysis and Ranking Procedure
The analysis and ranking procedure commences
once the process mining component has identified
the various process models (variants) from the
execution log. These are first grouped against
behavioural similarity. We do not include this aspect
in this paper due to space limitation and instead rely
on the process proposed by Lu and Sadiq (2008).
Then the popularity of the various variant (models)
is determined by figuring the count of instances
against each. Process popularity forms a benefit
attribute. On the other hand, there is a cost attribute
as well. We calculate this as the weight of the
process based on time and resources utilized. Finally
the cost and benefit attributes are combined through
a multi-criteria decision making approach to identify
the best process instance found from the history of
instances within the execution log.
We first present some basic definitions in order
to explain the analysis and ranking procedures.
Process Model. P is process model variance mined
from the execution logs, where P = {P
1
, P
2
, P
3
, ... ,
P
n
}. The architecture restricts that P are process
models with variance which have the same goal. All
process models are evaluated based on the
behavioural dimension (Lu and Sadiq, 2008), as it
contains the executional information such as the set
of tasks involved in the process execution, the exact
sequence of task execution, the performers and their
roles in executing tasks, the process-relevant data,
execution duration of the process instance and
constituent tasks.
From the reconstructed process model, we have
number of process instances captured by the
execution log. A particular process instance will be
represented as S, where S = {S
1
, S
2
, S
3
, ... , S
m
}.
Definition 1 (Process Model Popularity). Let P
i
denote the set of process model variants and S
j
be
the set of process model instances. Let F(S
j
, P
i
)
denote that “S
j
has the same process structure
(behaviour) as P
i
” Thus process popularity R for a
given variant i is
milarourally siare behavi
PSthatPSFwhereSR
ijijji
and ),,( =
(1)
The popularity of the process model shows how
many times a particular process (variant) models has
been selected by user/used previously. A process
matching on structural similarity (Lu and Sadiq,
2008) of business process model is used to identify
the various (goups of) variant models discovered.
The best practice of business process will show
the best alternatives from selected instances of
process model. As it works with more than one
criterion, a multi criteria decision making approach
has been used to rank the alternatives.
Definition 2 (Process Weighting). Let S be the set
of process instance. The weight ω represents the
value (i.e. cost value) of an activity. Time needed to
complete an activity is ∆t. ω
jl
is a value of an
activity l of process instance S
j
. Y
jl
is the weight of
an activity l of process instance S
j
. ∆t
jl
is the time to
complete an activity l of process instance S
j
. We
found that:
jljljl
tY
ω
×
Δ
=
(2)
Every instance S
j
of process model will then
have a summation of weight Z
j
from activity A to m.
cYZ
m
Al
jlj
+=
∑
=
(3)
with c is some fixed value which is not related to the
execution time (i.e. cost of resources such as to buy
paper, etc).
Selected top-k of process model instances will be
chosen from the set of process model instances with
the least weights, where k is a maximum number of
selected process instance defined by decision maker.
Definition 3 (Multi Attribute Decision Making).
Generally the multi-attribute decision making model
ICEIS 2010 - 12th International Conference on Enterprise Information Systems
168

can be defined as follow (Zimmermann, 2001). Let
C = {c
j
| j = 1, ... , m} the criterion set and let A = {a
i
| i = 1, ..., n} the selected set of process instance. A
multi criteria decision making will evaluate m
alternatives A
i
(i=1,2,...,m) against C
j
(j=1,2,...n)
where every attributes are independent of each other.
A decision matrix, X, given as follows:
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎣
⎡
=
mnmm
n
n
xxx
xxx
xxx
X
L
MMM
21
22221
11211
...
...
(4)
where x
ij
is a rank of a process instance i against
criteria j. The importance factor is given as W to
show the importance relative of each criteria, where
W = {w
1
, w
2
, ..., w
n
}. This importance factor will be
defined by the domain expert.
Definition 4 (Additive Weighting Method). The
concept is to find the weighted summation of
importance factor on each selected instances of
process model (Fishburn, 1967). In order to compare
all criteria, we normalise the decision matrix X into a
comparable scale.
⎪
⎪
⎩
⎪
⎪
⎨
⎧
=
attributecost a is if
Min
attributebenefit a is if
p
x
x
p
xMax
x
r
op
op
op
op
op
(5)
where r
op
is normalised rank of selected instance
alternative A
o
against attribute C
p
. Each selected
instance will have a preferred value V
o
, where
op
n
o
po
rwV
∑
=
=
1
(6)
The preferred value V
o
will indicate how we rank the
selected top-k instances. The higher the V
o
value is,
the higher its rank among others.
Let us consider example from figure 1. From the
execution log, we found variants of sequences <T1,
T2, T6, T5>, <T1, T3, T7, T5>, <T1, T2, T6, T7>,
<T1, T2, T7, T8>, <T1, T2, T8, T7>. Note that for a
given process instance, there is exactly one
execution sequence resulting from the execution,
also having the same sequence does not guarantee
two process instances could complete the process at
the same time. The collection of execution
sequences and counters (popularity) found from the
process is shown in Table 1.
Table 1: List of all execution sequences S and their
counters, from 100 process instances.
Sequences S Count (S)
<T1, T2, T6, T5> 25
<T1, T3, T7, T5> 16
<T1, T2, T6, T7> 18
<T1, T2, T7, T8> 22
<T1, T2, T8, T7> 19
Table 2: Examples of collected instances.
Alternative A
Selected
Instance S
Popularity R Weight Z
A
1
S
3
25 150
⁞ ⁞ ⁞ ⁞
A
8
S
10
18 149
⁞ ⁞ ⁞ ⁞
A
15
S
19
19 146
The selected instances are named as alternative A,
where A = {A
1
, A
2
, ..., A
n
}. These alternatives are
choices to be selected by the additive weighting
method.
Weight is a cost attribute, as system will likely
choose the least weight among all instances, while
popularity is a benefit attribute, as system will prefer
to use the most popular one. To get the decision
matrix, we will normalise all attributes.
967.0
150
145
150
6}149;145;149;148;152;149;158;15
7;148;152;155;156;149;min{150;15
11
===r
1
25
25
}2;19;19;1918;22;22;2
6;18;18;25;16;16;1max{25;25;
25
12
===r
Subsequently we calculate the rests, and develop the
normalised matrix as shown below
)7600,9930( ),7600,0001( ),7600,9730(
),8800,9540( ),8800,9800( ),8800,9120( ),7200,9180(
),7200,9730( ),7200,9240( ),6400,9540( ),6400,9800(
),6400,9730( ),0001,9290( ),0001,9350( ),000.1,967.0(
......
........
........
......
R =
The importance factor given by expert is W =
(6,4). The results are V
1
= 6*0.967 + 1*4 = 9.80;
subsequently we will have V
2
= 9.61; V
3
= 9.58; V
4
= 8.40; V
5
= 8.44; V
6
= 8.28; V
7
= 8.42; V
8
= 8.72;
V
9
= 8.39; V
10
= 8.99; V
11
= 9.40; V
12
= 9.24; V
13
=
8.88; V
14
= 9.04; V
15
= 9.00.
The EDRS output will highly recommend user to
use the instance of alternative A
1
(S
3
) to be used as it
has the highest score relative to others against the
given criteria.
4 METHOD EVALUATION
AND CONCLUSIONS
Our study is not without limitations. Although the
presented method does not seem to pose issues for
efficiency as its computational complexity is low,
the Additive Weighting process will require pair-
wise comparisons which for a large data set can be
prohibitive. Furthermore, when conflicting criteria in
SOCIALIZATION OF WORK PRACTICE THROUGH BUSINESS PROCESS ANALYSIS
169

the decision making framework are extended, it
would lead a computationally hard problem.
In our future work, we hope to reduce the number of
pair wise comparison in SAW method by
incorporating emerging method applied in multi
decision-making problem such as skyline query as
proposed by Börzsönyi et al. (2001). We also plan to
extend the criteria in the decision making framework
with the help of empirical studies on real processes
to gain the complexity faced in the real world
problems.
In summary, this paper has presented a method
for conducting business process analysis that aims at
capitalizing on previous practices and experiences,
thereby bringing about a socialization of work
practice within an organization. The presented
method balances costs (time and resources) and
benefits (process popularity) by utilizing a multi-
criteria decision making approach. Although we
have used specific criteria to demonstrate the
method, these can be extended to include further
criteria to better reflect the requirements of specific
process domains. Output from the recommendation
service of ranked process instances can greatly assist
the inexperienced user to utilize and learn from
previous organizational knowledge and address
specific cases with the knowledge of internal best
practice.
REFERENCES
Aalst, W. M. P. v. d., Hajo, A. R. & Minseok, S. (2005a)
Discovering Social Networks from Event Logs.
Comput. Supported Coop. Work, 14, 549-593.
Aalst, W. M. P. v. d., Hofstede, A. t. & Weske, M. (2003)
Business Process Management: A Survey. Business
Process Management.
Aalst, W. M. P. v. d., Medeiros, A. K. A. d. & Weijters, A.
J. M. M. (2005b) Genetic Process Mining.
Applications and Theory of Petri Nets 2005.
Aalst, W. M. P. v. d. & Song, M. (2004) Mining Social
Networks: Uncovering Interaction Patterns in Business
Processes. Business Process Management.
Aalst, W. M. P. v. d., Ton, W. & Laura, M. (2004)
Workflow Mining: Discovering Process Models from
Event Logs. IEEE Trans. on Knowl. and Data Eng.,
16, 1128-1142.
Biazzo, S. (2000) Approaches to business process
analysis: a review. Business Process Management
Journal, 6, 99-112.
Börzsönyi, S., Kossmann, D. & Stocker, K. (2001) The
skyline operator. ICDE. ICDE.
Boud, D., Cohen, R. & Walker, D. (1993) Using
Experience for Learning, Open University Press.
Bozkaya, M., Gabriels, J. & Werf, J. (2009) Process
Diagnostics: A Method Based on Process Mining.
Information, Process, and Knowledge Management,
2009. eKNOW '09. International Conference on.
Dubouloz, B. & Toklu, C. (2005) Discovering the Most
Frequent Patterns of Executions in Business Processes
Described in BPEL. Web Information Systems
Engineering – WISE 2005.
Fishburn, P. C. (1967) Additive utilities with incomplete
product sets: application to priorities and assignments.
Operations Research, 15, 537-542.
Grigori, D., Casati, F., Castellanos, M., Dayal, U., Sayal,
M. & Shan, M.-C. (2004) Business Process
Intelligence. Computers in Industry, 53, 321-343.
Günther, C. W. & Aalst, W. M. P. v. d. (2007) Fuzzy
Mining - Adaptive Process Simplification Based on
Multi-perspective Metrics. Business Process
Management (BPM) 2007.
Helen, S., Barbara, W., Boudewijn, D. & Wil, A. (2008)
Supporting Flexible Processes through
Recommendations Based on History. Proceedings of
the 6th International Conference on Business Process
Management. Milan, Italy, Springer-Verlag.
Hornix, P. T. G. (2007) P.T.G. Hornix. Department of
Mathematics and Computer Science. Eindhoven,
Technische Universiteit Eindhoven.
Lu, R. & Sadiq, S. (2006) Managing Process Variants as
an Information Resource. Business Process
Management.
Lu, R. & Sadiq, S. (2008) On the Discovery of Preferred
Work Practice Through Business Process Variants.
Conceptual Modeling - ER 2007.
Lu, R., Sadiq, S., Governatori, G. & Yang, X. (2009)
Defining Adaptation Constraints for Business Process
Variants. Business Information Systems.
Rozinat, A. & Aalst, W. M. P. v. d. (2005) Conformance
Testing: Measuring The Fit and Appropriateness of
Event Logs and Process Models.
Business Process
Management (BPM) 2005.
Schonenberg, H., Weber, B., van Dongen, B. & van der
Aalst, W. (2008) Supporting Flexible Processes
through Recommendations Based on History. Business
Process Management.
Tsai, C.-Y. & Chen, I. C. (2009) Using Genetic Process
Mining Technology to Construct a Time-Interval
Process Model. Next-Generation Applied Intelligence.
Vanderfeesten, I., Reijers, H. & Aalst, W. M. P. v. D.
(2008) Product Based Workflow Support: Dynamic
Workflow Execution. Advanced Information Systems
Engineering.
WFMC (1999) Workflow management coalition termi-
nology and glossary. Workflow management coalition.
Yujie, M., Shensheng, Z. & Jian, c. (2004) Supporting
collaborative work with process management
enhanced corporate portal. Systems, Man and
Cybernetics, 2004 IEEE International Conference on.
Zimmermann, H. J. (2001) Fuzzy set theory--and its
applications, Kluwer Academic Pub.
ICEIS 2010 - 12th International Conference on Enterprise Information Systems
170