SEMANTIC DESIGN PATTERNS FOR BUSINESS PROCESSES
Lobna Makni, Nahla Zaaboub Haddar and Hanene Ben-Abdallah
Faculty of Economics and Management Sciences, Mir@cl Laboratory, Sfax University, Sfax, Tunisia
Keywords: Semantic Business Process Patterns, Business process models, Conflicts, Construction rules.
Abstract: Both the academic and industrial communities are increasingly interested in developing methods and tools
for automating the design of business process models. In this context, several approaches were proposed to
make modeling easier and to enhance the quality of the resulting artifacts. To achieve these objectives, these
approaches are based on pattern reuse. Despite the agreed uppon advantages of patterns in accelerating the
design process and improving the produced model quality, a few researchers showed how to construct
business process patterns. In this paper, we describe an approach to construct Semantic Business Process
Patterns (SB2P) from a set of process models. A SB2P is a pattern synthesized from a set of process models
belonging to the same business domain. It is composed of process fragments that are semantically close but
may have structural and/or behavioral differences.
1 INTRODUCTION
Process modeling is considered a labor intensive
task, whose outcome depends on personal domain
expertise. Designers with low modeling competence
or domain expertise may introduce errors or
inconsistencies in the designed model which may
lead to bad performance and high process costs
(Müller et al., 2007). Thus, modeling tools must
incorporate techniques to help inexperienced
designers to work in an efficient manner. In fact,
there is a wide agreement that reuse can accelerate
the design process and produce high quality
solutions by adopting best practices (Buschmann et
al., 2007), (Tran et al., 2007), (Montero et al., 2010).
The various modeling approaches based on reuse can
be calssified into two main classes: reference
modeling and pattern reuse.
Reference modeling aims to increase productivity
by using configurable process models. A
configurable process model is a modeling artifact
that captures a family of process models and allows
analysts to understand what these process models
share, what their differences are, and why and how
these differences occur (Rosa et al., 2010).
Configured models are created for a specific domain
and are meant to be customized in different
application projects. They are constructed by
merging models after detecting similarities and
differences between them (Li et al., 2009) (Dijkman,
2007). On the other hand, a variety of patterns for
business processes have been proposed in the
literature like workflow patterns (der Aalst et al.,
2003), workflow activity patterns (Thom et al., 2009)
and action patterns (Smirnov et al., 2009). Workflow
patterns focus on specific aspects like control flow,
data flow and resource assignments. Workflow
activity patterns refer to the description of a recurrent
business function as it can be frequently found in
business processes (Thom et al., 2009); the authors
in (Thom et al., 2009) propose seven activity patterns
through an extensive literature study of business
process types (e.g., Approval, Question-answer,
Decision Making, …). In contrast to workflow
patterns, action patterns are closely related to the
semantic content of a process model. In addition,
unlike reference models, action patterns are abstract
enough to be applicable in various domains
(Smirnov et al., 2009).
In this paper, we propose a pattern concept that
combines the advantages of both reference models
and action patterns: focusing on structural concepts
specific to business processes, accounting for the
semantic aspects, and ensuring a high level of
abstraction to provide for a wide reuse range. More
specifically, we define the concept of Semantic
Business Process Pattern (SB2P). A SB2P is a
pattern synthesized from a set of “good-quality”
process models belonging to the same business
domain. It is composed of process fragments that are
83
Makni L., Zaaboub Haddar N. and Ben-abdallah H..
SEMANTIC DESIGN PATTERNS FOR BUSINESS PROCESSES.
DOI: 10.5220/0003653500830087
In Proceedings of the 6th International Conference on Software and Database Technologies (ICSOFT-2011), pages 83-87
ISBN: 978-989-8425-76-8
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
common to the source models but it may contain
fragments with different structural and/or behavioral
representations in the source models. It represents a
factorization of constructs common to process
fragments in the source models. The factorization
detects and resolves the semantic, structural and
behavioral conflicts between them.
In the remainder of this paper, we first present
our SB2P construction approach. Secondly, we
highlight conflicts susceptible to exist among process
fragments when trying to factorize them and we
present our factorization/synthesis rules. Finally, we
place the presented work in the context of already
proposed approaches.
2 SB2P CONSTRUCTION
APPROACH
To construct a SB2P, we start from a repository of
BPM of good quality, which are classified into
different business domains. Given a business domain,
our approach proceeds according to the three
following steps:
1. Extraction of process fragments that are
semantically close and frequently present in the
analyzed process models (possibly with
different structures and/or behaviors): The
detection of these fragments relies on label
similarities (see section 3.1) using an ontology
for the analyzed business process domain.
2. Difference/conflict detection: For the extracted
process fragments, this step uses a set of
comparison rules to identify three types of
conflicts: semantic, structural and behavioral
(see section 3.2).
3. SB2P construction: Once the conflicts are
resolved among the semantically-close process
fragments, a set of factorization rules are
applied to construct the SB2P (see section 3.2).
3 CONFLICT DETECTION AND
SEMANTICALLY-CLOSE BP
FRAGMENT IDENTIFICATION
Comparing business processes requires a common
specification notation. For the purposes of this paper,
we abstract away from any specific notation and we
represent BPM as directed graphs with labeled nodes.
Each node has a type that represents the commonly
found types in all process modeling languages:
‘activity’, ‘event’ and ‘connector’. In addition, similar
to most modeling languages, our graph uses three
kinds of connectors: AND, XOR and OR.
3.1 Conflict Detection
When comparing processes, we need to distinguish
between semantic, structural and behavioral
conflicts to carry out the comparison. A semantic
conflict appears when activities use different labels
between which there is a semantic relation
(subsumption, part of…). A structural conflict
emerges when various representations describe
similar behavior. While a behavioral conflict appears
when process fragments are semantically close but
have different behavioral profiles. Table 1 presents
our classification of conflicts inspired from
(Dijkman, 2007).
Table 1: Conflicts between business process fragments.
Semantic
conflicts
Subsumed activity: An activity named a
1
subsumes an activity a
2
, if it represents the same unit of work
as the other activity, but includes another unit of work as well.
Partly corresponding activity: An activity named a
1
partly corresponds to an activity a
2
, if these
activities partly represent equivalent units of work, but both also represent different units of work.
Structural conflicts
Skipped activity: is an activity which exists in one process, but there is neither an equivalent, nor a
subsumed, nor a partly corresponding activity in the other process.
Refined activity: exists if an activity a1 exists in one process, but an equivalent unit of work is only
represented by a collection of activities in the other process. The collection of activities refines the
single activity, because it represents the same unit of work at a different level of granularity.
Additional dependencies: correspond to the case in which one set of activities includes the other. The
set that includes the other has additional dependencies.
Behavioral
conflict
Iterative vs. once-off occurrence: is the case in which an activity is part of a loop in one process while
it is not in the other process. This means that in one process the activity must be performed correctly
in one go, while in the other process it can be performed repeatedly until the result is satisfactory.
Different conditions for occurrence: In case the dependencies for two equivalent activities have
different conditions for their occurrence.
ICSOFT 2011 - 6th International Conference on Software and Data Technologies
84
3.2 Extraction of Similar Business
Process Fragments
To compare elements of a process model with
another one, we use a mapping function
map
f
inspired from (Rosa et al., 2010). With this
function, a mapping between nodes of different
types, or between a split and a join, has a
matching score of 0. The matching score of a
mapping between two activities or between two
events is measured by the similarity of their
labels. Given two activities, their semantic
similarity score is the degree of similarity, based
on equivalence between words in their labels.
Words that are identical are given a score of 1,
while words that are synonymous are given a
score of 0.75, a value that was determined
experimentally by (Dongen et al., 2008). Thus, an
exact match is preferred over a synonym match.
The semantic similarity score of two activities a
1
and a
2
is defined in (1).
12 21
12
(,) \ \
12
12
1* 0,75* ( , )
(, )
max( , )
sl a a a a
map
a a synonym s l
faa
aa
∈×
∩+
=
(1)
In equation (1) synonym is a function that returns 1
if the given words are synonyms and 0 if they are
not. This measure considers a synonym relationship
of two instances, the number of synonyms that are
proposed for one term by the ontology of the domain
and weights the number of synonyms against the
maximum sense cardinality of these two terms.
Frequently occurring words are skipped, such as “a”,
“an” and “for”.
To find similar business process fragments, we
first transform each pair of process models into a
matrix where the lines and columns correspond to
activities of the compared process models. Each
element in the comparison matrix represents the
value of the mapping function of the corresponding
activities.
To illustrate our approach, we consider the two
models in fig. 1describing the process of “obtaining
a loan”. The corresponding comparison matrix is
given in Table 2.
Given a comparison matrix, we consider a block
as a set of adjacent cells. To extract semantic
process fragments from the comparison matrix, we
first permute its lines and columns to form cell
blocks with non zero values. Then, we use the
following rules:
1. If a block consists of only one cell C
i,j
with C
i,j
=1 then activity a
i
in the first process model is
equivalent to activity a
j
in the second model.
We take this activity as it is in the pattern.
2. If a block consists of only one cell
ji,
C
with
[0.89,1[C
ji,
then the label of a
i
in the first
process model is semantically very close to the
label of a
j
in the second one. The cut off value
0.89 is determined experimantelly by (Dongen et
al., 2008). We take one of these activities in the
pattern.
3. If a block has a 1*k or a k*1 (k0) dimension
and consists of cells C
i,j
with
[0.5,0.89[C
ji,
, we
consider the row (or the column) activity as a
refined one. If k=1 then the label of a
i
in the first
process model has the same meaning as the label
of a
k
in the second one. We take one of these
activities in the pattern.
Figure 1: Examples of BPM from the repository.
Model 2
Model 1
SEMANTIC DESIGN PATTERNS FOR BUSINESS PROCESSES
85
Table 2: Comparison matrix for models 1 and 2.
Model 2
Model 1
Revise
fulfillment
of loan
request
Make
client
details
Check
credit
Check
loans
Check
changes
Pass to car
manufacturer
Print
Make
changes
Decide
mandate
Notify
sales
Check
completeness
of loan
request
0,875
0 0
0 0
0 0
0
0 0
Check
existing
client
0,1875
0,583
0,33 0,33 0,33 0 0 0,25 0 0
Add client
information
0
0,583
0 0 0 0 0 0,25
0 0
Produce
approval
0 0 0,325 0 0 0 0 0
0 0
Check credit
0,375
0
1 0,875 0,5 0 0 0 0 0
Checking
module
0,1875
0
0,5
0,5 0,5 0 0 0,325 0 0
Pass to car
manufacturer
0 0 0 0
0
1
0 0
0 0
Make
changes
0 0 0,325 0,325
0,875
0 0
1 0 0
Check
changes
0,1625 0 0,5 0,5
1
0
0
0,325
0 0
Print 0 0 0 0 0 0 1 0 0 0
Plan
appointment
0 0 0 0 0
0
0 0 0 0
Figure 2: The resulting SB2P: Loan pattern.
4. If an activity in a model has an equivalent one
in the other and has also refined activities, then
we consider the refined one as additional
dependencies to the activity.
5. If a block consists of only one cell
ji,
C
with
0C
ji,
=
then the label of a
i
in the first process
model haven’t a corresponding activity in the
second model. If a line or a column contains
only 0s, we consider the corresponding activity
as a skipped one.
We replace a skipped activity by silent one. A
silent activity in the business model is an empty
activity without added value. So, it does not have
label which describes his function in the model. It
can be replaced by the designer during modeling.
By applying construction rules 1-5 on the example
of fig. 1, we obtain the SB2P of fig. 2 which we
baptize “Loan pattern”.
4 CONCLUSIONS
The main contributions of this paper are to propose a
new reuse concept for business processes, called
Semantic Business Process Patterns (SB2P), and an
approach for its construction. The proposed
approach uses semantic relations to compare BPM in
a given domain and to determine common fragments
that are semantically close. In addition, it tolerates
structural and behavioral differences among the
process fragments as long as they are conflict-free.
Our SB2P concept resembles more the reference
modeling: they both offer process fragments. But, in
contrast to the reference model construction
approach which merges process fragments (Rosa et
al., 2010), our SB2P construction approach
factorizes fragments by taking only semantically
close elements in the resulting pattern; this
difference is justified by the genericity/abstraction
property behind an SB2P.
On the other hand, while SB2P accounts for
modeling element semantics, workflow patterns (der
Aalst et al., 2003) and workflow activity patterns
(Rosa et al., 2010) do not, they rather focus on
specific aspects like control and data flows and
resource assignments. Consequently, these patterns
are more appropriate to the development of business
Skipped Activities
Semantically close activities
Refined Activity
Refined Activity
Additional De
p
endencies
Identical Activities
ICSOFT 2011 - 6th International Conference on Software and Data Technologies
86
process modeling tools than the design of business
processes. We are currently automating the presented
SB2P construction approach in order to evaluate its
advantages and limits. In addition, we are examining
the benefits of SB2P in the design of business
processes.
REFERENCES
Müller, D., Reichert, M., Herbst, J.: Data-driven modeling
and coordination of large process structures. LNCS,
4803:131, 2007.
Buschmann, F., Henney, K., Schmidt, D.: Past, present and
future trends in software patterns, IEEE Software, 24(7/8)
31–37, (2007).
Tran, H., Coulette, B., Thuy, D.: Broadening the use of
process patterns for modeling processes. In: Proc. SEKE,
Knowledge Systems Institute Graduate School, 57–62,
(2007).
Thom, L., Reichert, M., Iochpe, C.: On the Support of
Workflow Activity Patterns in Process Modeling Tools:
Purpose and Requirements, (2009).
Smirnov, S., Weidlich, M., Mendling, J., Weske, M.: Action
patterns in business process models. In Baresi, L., Chi,
C.H., Suzuki, J., eds.: ICSOC/ServiceWave. Volume of
5900 LNCS, 115-129, (2009).
Van der Aalst, W. M. P., Ter Hofstede, A. H. M.,
Kiepuszewski, B., Barros, A. P.: Workflow patterns.
Distrib. Parallel Databases, 14(1), (2003).
La Rosa, M., Dumas, M., Uba, R., Dijkman, R.M.: Merging
business process models. In Proc. of OTM, volume 6426
of LNCS, pages 96–113, (2010).
Li, C., Reichert, M., Wombacher, A.: Discovering reference
models by mining process variants using a heuristic
approach. In Proc. of BPM, volume 5701 of LNCS, pages
344-362, Springer, (2009).
Dijkman, R.: A Classification of Differences between Similar
Business Processes, In Proceedings of the 11th IEEE
EDOC Conference (EDOC’07), pages 37–50, (2007).
Montero, I., Pna, J., Ruiz-Cortés, A.: A Methodological
Framework for Obtaining the Core Architecture of
Business Information Systems Families. CAISE
conference. (2010).
van Dongen, B. F., Dijkman, R. M., Mendling, J.: Measuring
similarity between business process models. vol. 5074 of
LNCS, pp. 450–464, Springer, (2008).
SEMANTIC DESIGN PATTERNS FOR BUSINESS PROCESSES
87