Detection of Semantic Relations between Business
Process Activity Labels
Lobna Makni, Nahla Zaaboub Haddar and Hanene Ben Abdallah
Mir@cl Laboratory, University of Sfax, Airport Road Km 4, Sfax, Tunisia
Keywords: Semantic Relation, Subsumed Activity, Part-of Relation Activity, Semantic Relation, Decision Rule.
Abstract: Determining similarities between business process models needs to be resolved in several cases, e.g., when
business processes must be merged, configured or integrated; when reusing fragments of process models,
etc. Similarity can be tackled at the semantic, structural and/or behavioral levels. In this paper, we are
interested in the semantic level similarity between business process models. More specifically, we present a
method for detecting semantic relations between activity labels in process models belonging to the same
business domain.
1 INTRODUCTION
Detecting the similarity between BPMs has recently
gained interest in the business process management
community (Awad et al., 2008). Similarity was
addressed separately either at the semantic,
structural or behavioral level of process models. The
semantic level of a process model is the description
of the meaning of its elements including its activity
and event labels.
Comparing business processes at this level is a
fundamental step for several model management
activities. For example, in a merger scenario, it is
necessary when integrating similar business process
models into one process (La Rosa, 2010). In
addition, this comparison is needed for managing
process model repositories and ensuring the
uniqueness of process models (Dijkman, 2011).
Furthermore, multi-national enterprises need a
similarity measurement to identify specialized
processes of a national branch that no longer comply
with the procedures defined in the company
reference model (Van Dongen, 2008).
The aim of this paper is to use a linguistic
comparison between activity labels to derive
additional semantic relation types such as hyponyms
and meronyms. A hyponym is an activity label
whose semantics is included within another activity
label, its hypernym. A hyponym shares a type-of
semantic relation with its hypernym. On the other
hand, a meronym activity label denotes a constituent
part of another activity label. The detection of such
relations between activity labels helps to identify
relations between process fragments. Indeed, a
process fragment is made up of activities that have
semantic relations and control dependency among
them. Identifying common fragments between
business processes avoids redundancies when
merging them. It is also necessary for establishing
relations between semantically close process
fragments.
In addition to the new semantic relations, the
second contribution of this paper is a relation
detection method that finds semantic relations
among activity labels of process models belonging
to the same business domain. This method can be
used to extract process fragments that are
semantically close and frequently present in the
analyzed process models.
In the remainder of this paper, we first give an
overall view of our relation detection method. Then,
we present its set of decision rules to detect semantic
relations among activity labels of process models
belonging to the same business domain. Finally, we
place the presented work in the context of already
proposed approaches.
2 SEMANTIC RELATION
DETECTION METHOD
A semantic relation between activities appears when
273
Makni L., Zaaboub Haddar N. and Ben Abdallah H..
Detection of Semantic Relations between Business Process Activity Labels.
DOI: 10.5220/0004003102730277
In Proceedings of the 14th International Conference on Enterprise Information Systems (ICEIS-2012), pages 273-277
ISBN: 978-989-8565-12-9
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
activities use different labels among which there is a
semantic relation of equivalence, synonymy,
subsumption or part-of. An activity A in one process
subsumes an activity B in another process if A
represents the same unit of work as B, but includes
other units of work as well (Dijkman et al., 2007).
On the other hand, an activity A in one process
partly corresponds to an activity B in another
process, if A and B partly represent equivalent units
of work, but both represent also different units of
work.
The semantic relation detection method we
propose is composed of two main steps (Figure 1):
label refactoring, and definition of semantic links
between labels.
Figure 1: Semantic relation detection method.
2.1 Label Refactoring
An activity label captures an action and a business
object on which the action is performed. To compare
two activity labels, it is preferable that they have the
same syntactic pattern, e.g., an action followed by a
business object. Unfortunatelly, designers use
different styles when labeling activities. Thus,
before comparing activity labels, a preliminary step
of label refactoring is needed: This step aims at
harmonizing the writing styles of the activity labels
into the following form:
action + Business-Object
Action field BO field
According to (Leopold and al., 2010), in practice,
label writing styles can be classified into three
syntactic patterns: verb-object labels, action-noun
labels, and rarely other irregular forms. (Leopold
and al., 2010) developed an algorithm to detect the
syntactic pattern of an activity label in a BPM. In
addition, this work proposes a derivation algorithm
to rewrite labels from the action-noun pattern to the
verb-object pattern. To refactor the actvity labels
into the above action-business object style, we use
this derivation algorithm to extract from each label
the action and the associated business object. For
example, the label ‘Creation of invoice’ is an action-
noun pattern; after the derivation to the verb-object
label, we get the action ‘create’ and the object
‘invoice’. Thus, after the label refactoring step, each
activity label can be represented as a couple (a, O),
where a is an action and O is a business object. The
third irregular syntactic form is rarely found in
business process models (1.1%). It concerns frozen
forms which can be added as they are to the
ontology used for the detection of semantic
relations.
2.2 Label Semantic Relations
Definition
To identify semantic relations between activity
labels, we proceed in three steps (Figure 1). First,
we compare the business objects specified in the
labels’ object fields; we believe that activities that
do not act on the same object cannot be compared.
For this business objects comparison, we need a
business domain ontology. In our study, we used the
eTVSM ontology (Awad, 2008) which offers a
reference terminology that defines semantic
correspondences between business objects. For
example, using eTVSM, we can detect that the
business object ‘customer order’ is a kind of
business object ‘order’. If two activity labels use the
same (or synonym) business objects, we continue
with the detection of the semantic relation between
their actions. Since we are looking for relations
among verbs in general, in this second step, a
general-purpose dictionary is sufficient. In our
study, we used WordNet to determine semantic
relations between actions (i.e., the verbs in the
labels).
As illustrated in Figure 1, once the semantic
relations between the business objects and the
actions are identified, our semantic relation detection
method applies a set of decision rules that we
describe next.
3 DECISION RULES
To detect semantic relations, we need to construct,
ICEIS2012-14thInternationalConferenceonEnterpriseInformationSystems
274
for each pair of process models belonging to the
same business domain, a matrix to compare the
activity labels. In a comparison matrix, lines and
columns correspond to activity labels of the
compared process models. Each element in the
comparison matrix represents the semantic relation
(synonym, subsumption, part-of, different) according
to the decision rules defined below in this section.
To detect the semantic relations between activities,
our decision rules are based first of all on the
comparison of business objects and then continue
with the comparison of their associated actions.
Given two activities, there are four types of binary
semantic relations between them: identity,
synonymy, subsumption and part-of. These
relations are determined according to the following
rules:
Rule 1 (synonym activities): Let A
1
= (a1, O1) and
A
2
= (a2, O2) be two activities of two process
models M
1
and M
2
respectively. A
1
and A
2
are
synonyms, noted A
1
≡ A
2
, if and only if O
1
≡ O
2
and
a
1
≡ a
2
.
Rule 2 (subsumed activities): Let A
1
= (a1, O1) and
A
2
= (a2, O2) be two activities of two process
models M
1
and M
2
respectively. A
1
subsumes A
2
,
noted A
1
↑ A
2,
if and only if O
1
≡ O
2
and a
1
is an
hypernym of a
2
(a
1
↑a
2
).
Figure 2: Two ordering processes.
Rule 3 (part-of activities): Let A
1
= (a1, O1) and A
2
= (a2, O2) be two activities of two process models
M
1
and M
2
respectively. A
1
is part-of A
2
, noted A
1
−
♦
A
2
, if O
1
is a part-of O
2
and a
1
≡ a
2
.
Rule 4 (different activities): Let A
1
= (a1, O1) and A
2
= (a2, O2) be two activities of two process models
M
1
and M
2
respectively. A
1
and A
2
are different,
noted A
1
Ð
A
2
, if and only if O
1
Ð
O
2
.
To illustrate our approach, we consider the two
models in Figure 2 describing an ordering process.
By applying the above defined decision rules, we
obtain the comparison/decision matrix shown in
Table 1. For example, the semantic relation between
‘confirm order’ and ‘validate order’ is subsumption.
Indeed, after decomposing these activity labels into
the verb-object grammatical structure, we obtain
‘confirm’ and ‘validate’ as verbs in the action field
and ‘order’ as noun in the business object field.
Using the eTVSM ontology, we can detect that these
activity labels have the same object ‘order’. Then,
using WordNet, we find that the action ‘validate’
subsumes ‘confirm’. Therefore, the activity label
‘validate order’ subsumes the activity label ‘confirm
order’.
4 RELATED WORKS
In the business process management domain, there are
many issues and difficulties related to terminology
mismatches and the unstructured and isolated
knowledge representation in process models. To
tackle these weaknesses, several approaches were
proposed in the literature. They are based on solutions
using domain ontology construction, aggregation
techniques or semantic similarity metrics.
Two kinds of ontologies are used to enable the
semantic support of modeling BP activities. The
sBPMN ontology (semantic Business Process
Modeling Notation) represents BPMN process
models, featuring basic concepts and attributes for
Table 1: Decision matrix for BPM 1 and BPM 2.
A
1
A
2
(Receive,
order)
(Validate,
order)
(perform, credit
check)
(Analyze,
order)
(Accept,
order)
(Transfer,
order)
(reject,
order)
(Inform,
customer)
(Receive, customer order) A
1
≡ A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
(Confirm, order) A
1
Ð
A
2
A
1
↑ A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
(Accept, order) A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
≡ A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
(Send, confirmation) A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
(Produce, assemblies) A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
(Create, invoice) A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
(Send, order and invoice) A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
−
♦ A
2
A
1
Ð
A
2
A
1
Ð
A
2
(Send, rejection) A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
A
1
Ð
A
2
DetectionofSemanticRelationsbetweenBusinessProcessActivityLabels
275
standard BPMN elements (Abramowicz et al., 2007).
In (Born et al., 2007), the authors extended sBPMN to
provide for the definition of the states of a business
object before and after the execution of corresponding
activities and to link objects, states and activities to
elements of domain ontologies describing them. With
these extensions, the sBPMN ontology can be used as
an internal and external format for semantically
augmented BPMN process models. The domain
ontology covers information concerning domain
objects and states which help to model business
processes more precisely.
On the other hand, it is common for large
organizations to maintain repositories of BPMs in
order to document and to improve their operations. To
retrieve process models from such BPM repository, a
comparison means is required (Dijkman, 2011)
(Ehrig, 2007) (Van Dongen, 2008). Based on label
similarity, (Dijkman et al., 2011) propose a label
matching similarity metric. The metric definition
depends on the syntactic or semantic similarity
notions or a weighted average of them (Dijkman et
al., 2011). In addition, (Ehrig et al., 2007) also
proposed a combined metric that computes similarity
degrees between a pair of process element names
based on syntactic, linguistic and structural measures.
In (Dijkman et al., 2011) and (Ehrig et al., 2007), the
authors use the WordNet dictionary to detect
synonymous words.
Furthermore, in the context of company mergers,
teams of analysts need to compare similar process
models to identify commonalities and differences, and
to create a configurable process model that captures a
family of process models in an integrated manner (La
Rosa et al., 2010). (La Rosa et al., 2010) used a
matching score of a mapping between two functions
or events based on the similarity between their labels.
The matching score depends on syntactic and
linguistic similarity measures. In (Dijkman et al.,
2011) and (Makni et al., 2011), the authors use the
same mapping functions to calculate the similarity
between activity labels based on synonym words.
A significant point in the design of the
aggregation operation is activity aggregation. Existing
BPM abstraction techniques from the semantics of
activities in business process models. In (Smirnov et
al., 2010), the authors developed an aggregation
technique clustering activities according to their
domain semantics. The technique can guide the user
during a process model abstraction providing
recommendations on related activities. Aggragation of
actions requires them to be related by a part-of or
meronym relation. This work proposed a metric for
comparing activity aggregations and the algorithm for
aggregation mining (Smirnov et al., 2010). The metric
is applied on a meronymy forest represented by the
MIT Process Handbook (Malone et al., 2003). This
latter describes business processes obtained by
researchers with the help of business process experts.
It represents several business domains such as sales,
distribution, and production. The handbook illustrates
about 5000 activities with their semantic relations like
hyponymy and meronymy.
5 CONCLUSIONS
The main contributions of this paper are to propose
decision rules to detect semantic relations between
activity labels and a semantic relations detection
method. The proposed method determines semantic
relations between activity labels such as subsumption
and part-of relation.
We are currently automating the presented method
in order to evaluate its advantages and limits. In
addition, we will validate our relations detection
method by an empirical study on process models to
determine its precision rate.
REFERENCES
Abramowicz W., Filipowska, A., Kaczmarek, M., and
Kaczmarek, T., 2007. Semantically enhanced business
process modelling notation: Workshop on Semantic
Business Process and Product Lifecycle Management
(SBPM 2007).
Awad A., Polyvyanyy, A., Weske M., 2008. Semantic
Querying of BPM: EDOC 2008, - pp. 85-94.
Born M., Dörr, F., Weber, I., 2007. User-friendly Semantic
Annotation in Business Process Modeling: WISE
Workshops, pp-260-271
Dijkman R., Dumas, M., Van Dongen, B., Käärik,R.,
Mendling, J., 2011. Similarity of BPMs: Metrics and
Evaluation: Information System, - Vol. 36(2): pp. 498-
516 .
Ehrig M., Koschmider, A., Oberweis, A., 2007. Measuring
Similarity between Semantic BPMs: APCCM 2007 - pp.
71-80.
La Rosa M., Dumas, M., Uba, R., Dijkman, R. M., 2010.
Merging BPMs: In Proc. of OTM - Vol. 6426: pp. 96–
113.
Leopold H., Smirnov, S., Mendling, J., 2010. Refactoring of
Process Model Activity Labels: NLDB'10- pp. 268-276.
Makni L., Zaaboub, N., Ben Abdallah, H. Semantic Design
Patterns for Business Processes: the 6
th
International
Conference on Software and Data Technologies
(ICSOFT’11), 2011.
Malone T. W., Crowston, K., Herman, G. A., 2003.
Organizing Business Knowledge: The MIT Process
Handbook: The MIT Press, Cambridge.
ICEIS2012-14thInternationalConferenceonEnterpriseInformationSystems
276
Smirnov S., Dijkman, R., Mendling, J., Weske M. , 2010.
Meronymy-Based Aggregation of Activities in BPMs:
ER - pp. 1-14.
Van Dongen, B., Dijkman, R. M., Mendling, J, 2008.
Measuring Similarity between Business Process Models:
In CAiSE, Vol. 5074 Springer, pp. 450-464.
DetectionofSemanticRelationsbetweenBusinessProcessActivityLabels
277
