Semantic Approach to Automatically Defined Model Transformation
Tiexin Wang, Sebastien Truptil and Frederick Benaben
Centre Genie Industriel, University de Toulouse - Mines Albi, Campus Jarlard, 81000 Albi, France
Keywords: Model-Driven Engineering, Automatic Model Transformation, Semantic Check, Ontology.
Abstract: Modelling and model transformation are regarded as two pillars of model-driven engineering; they have
been used together to solve practical problems. For instance, since different models (e.g. data model) are
used by heterogeneous partners involved in a specific collaborative situation, there is an urgent need for
model transformations to exchange information among the heterogeneous partners. To quickly define model
transformations, this paper presents an approach, which could replace the users’ effort in making mappings
during the definition of a model transformation process. This approach is based on model transformation
methodology, using syntax and semantic relationship among model elements. For this, a generic meta-meta-
model and semantics checking methodology are proposed, before being illustrated by an example.
1 INTRODUCTION
Nowadays, different kinds of models have been
widely used (store data, simulate industrial processes,
manage information service, etc.) by different
domains, and modelling and model transformation
will play a key role in solving complex, diversity
issues. Problems, which exist in collaborative
situations, are such issues. As explained in (Rajsiri
and al, 2010), more and more collaborative
situations (domains-crossing) are frequently
appearing and disappearing, such as crisis
management, supply chain management, and
enterprise interoperability (Chen and al, 2007), etc.
Consequently, to improve the efficiency of the
collaborative, fast information exchange among
different partners is necessary. Fortunately, model
transformation methods can provide a solution
which aims at solving such issues. There are several
approaches using model transformation methods to
solve practical problems, such as mentioned in
(Castro and al, 2011), and (Grangel and al, 2010).
However, the process of model transformation
definition could still be improved. Indeed,
development of model transformations involves
many repetitive tasks, which are often done
manually (Del Fabro and Valduriez, 2008). These
repetitive tasks are used to define mapping rules
between the elements of source and target models.
These rules are executed during the transformation
process. As the context of model transformation is
different, users should make the mappings manually
(according to the specificity of domains that models
come from, the source and the target models are
distinguished). Therefore, generating a solution to
automatically define model transformations is a
motivating challenge.
Generating automatically model transformation
is closely related to the syntax and semantic
matching between concepts of the source and target
models. The syntax and semantic matching
approaches are defined at the abstract level of model
(meta-model level). Based on this principle, our
main idea consists in automatically defining syntax
and semantic mappings between several meta-
models. Most syntax and semantic matching
approaches cannot be applied to models that
conform to different meta-models (Del Fabro and al,
2005). For this reason, we define a generic meta-
meta-model (we want a generic and simple meta-
meta model that can work along with a specific
ontlogy which provides the data basis for semantic
check). Concerning the syntax matching approach,
existing techniques and methodologies could be
reused such as explained in (Lano and Kolahdouz-
Rahimi, 2013), and (Bollati and al, 2013). Although
semantic check methods have been used in other
research fields as explained in (Ly and al, 2006),
semantic matching approaches for model
transformation are a challenging research goal. To
achieve our aim, we try to use ontology
(McGuinness and Van Harmelen, 2004) and
340
Wang T., Truptil S. and Benaben F..
Semantic Approach to Automatically Defined Model Transformation.
DOI: 10.5220/0004713303400347
In Proceedings of the 2nd International Conference on Model-Driven Engineering and Software Development (MODELSWARD-2014), pages 340-347
ISBN: 978-989-758-007-9
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
semantic check rules.
This paper is divided in four parts. In the first
section, definitions of model, meta-model and model
transformation principles are given. Then an
overview of our solution is proposed in the second
section. The third section makes a focus on the
semantic mapping approach. Before the conclusion,
a case study is presented in the fourth section to
illustrate our solution.
2 MODEL TRANSFORMATION
OVERVIEW
With the wide use of model-driven engineering
theory in many specific domains, more and more
researchers and organizations are becoming
interested in finding solutions to effective model
transformation.
This section is divided into four sub-sections to
give an overview of model transformation
(combined with our proposal) in four aspects,
respectively. First, the definitions of model and
meta-model. Second, the model transformation
theories. Third, the model transformation
approaches. Fourth, the model transformation
techniques.
2.1 Model & Meta-Model
Model transformation is based on two basic and
crucial concepts: model and meta-model (Bézivin,
2006).
A model could be seen as a picture of a system,
depending on a point of view. This picture is a
simplification of this system, which highlights its
characteristics. A meta-model defines the
characteristics of a valid model. A meta-meta-model
for deducing meta-models from input models is
proposed in this article.
2.2 Model Transformation Principles
Figure 1 (Bénaben and al, 2010) illustrates the
model transformation principles. This idea inspired
our work.
The two: “source and target models” are built
according to their meta-models (MM).
The key point is that the source MM shares part
of its concepts with the target MM (the two spaces,
source and target, have to be partially overlapping in
order to allow model transformation). As a
consequence, the source model embeds a shared part
and a specific part. The shared part provides the
extracted knowledge, which may be used for the
model transformation, while the specific part should
be saved as capitalized knowledge in order not to be
lost. Then, mapping rules (built based on the
overlapping conceptual area of MMs) can be applied
onto the extracted knowledge. The transformed
knowledge and an additional knowledge (to fill the
lack of knowledge concerning the non-shared part of
concepts into the target MM) may be finally used to
create the shared part and the specific part of the
target model.
Figure 1: Model transformation framework.
2.3 Model Transformation Approaches
In general, according to (Czarnecki and Helsen,
2003), there are two main kinds of model
transformation approaches. They are: model-to-code
approaches and model-to-model approaches. For the
model-to-code approaches, there are two detailed
categories:
Visitor-based approaches
Template-based approaches
For the model-to-model approaches, there are five
detailed categories:
Direct-Manipulation Approaches
Relational Approaches
Graph-Transformation-Based Approaches
Structure-Driven Approaches
Hybrid Approaches
The method “automatically define model
transformations” could be seen as a complement to
the classification of model transformation
approaches. Automatic model transformation could
also be used in association with other model
transformation approaches; it will become a part of
the whole process of transformation, or complete a
specific function.
SemanticApproachtoAutomaticallyDefinedModelTransformation
341
2.4 Model Transformation Techniques
In practice, a large number of techniques have been
developed to perform model transformation. The
most prevalent model transformation techniques are:
QVT (Query, View, and Transformation language)
(OMG, 2002), ATL (Atlas transformation language)
(Jouault and al, 2007) and some of the graph
rewriting based model transformation languages.
QVT is defined by the “Object Management
Group (OMG)”, and QVT defines three specific
model transformation languages.
The ATL model transformation language is
defined by the “ATLAS Group, (INRIA & LINA)
University of Nantes”. ATL architecture provides a
set of languages: the ATLAS Model Weaving
(AMW), ATL, and the ATL Virtual Machine (ATL
VM).
The model transformation languages (based on
graph rewriting) describe transformations that
operate on a graph by rewriting it. A transformation
is performed in steps operating on a current graph.
There are several graph rewriting languages, such as
“GReAT (Agrawal and al, 2003)” and “AGG
(Taentzer and al, 2009)”, etc.
The “ATL” has been choosed to develop the tests
for our proposal.
3 GENERAL OVERVIEW
OF THE SOLUTION
In this section, an overview of the solution will be
illustrated.
This section is divided into two subsections. In
the first subsection, the main objective of our work
is explained. In the second subsection, an overview
of our solution is given.
3.1 Main Objective
The main objective of this work is to define a
process of automatical model transformation.
In order to achive this objective, there are several
basic function requirements that should be
implemented. They are listed as following:
Define a generic meta-meta-model.
Create an ontology based on the meta-meta-model.
Analysis the input from the users (source model
and target model; source model and target meta-
model; source meta-model and target meta-model)
Deduce our source meta-model and target
meta-model based on the analysis results and the
generic meta-meta-model.
Apply syntax check rules on the definition of
model transformation process
Apply semantic check rules on the definition of
model transformation process.
Here, basic requirements for achieving the final
objective are given. In the next sub-section, these
functions will be added into the architecture of our
solution.
3.2 The Architecture of Theoretical
Solution
Figure 2 illustrates the architecture of the solution.
Figure 2: Theoretical solution architecture.
The purpose of this work is to transform a source
model to a specific target model automatically. The
source and target models could be built in different
modelling languages (“UML (Fowler, 2004)”,
“BPMN (White, 2004)”, etc.). In order to ignore the
modelling language and use the semantic and syntax
check rules on the definition of transforming
process, we suppose to develop several intermediary
models (building with a specific modeling
language). Based on this idea, we also define a meta-
meta-model. In order to add semantic check rules to
our model transformation methods, we need a
specific ontology to provide the data basis. A meta-
meta-model that consistent to this ontology will
greatly simplify the transformation process (this is
the reason that we do not use the existing meta-
meta-models, such as: MOF (OMG,2002)). We
deduce the meta-models for both source and target
models that conform to the meta-meta-model. Then
using the semantic and syntax check rules on the
meta-model level to build transformation mappings.
During the transformation process, the providers of
the source models could check the intermediary
models.
To be efficient, all the semantic and syntax check
MODELSWARD2014-InternationalConferenceonModel-DrivenEngineeringandSoftwareDevelopment
342
rules should be used on the same kind of models
(intermediary models). So, we divide the
transformation process into three steps: from the
source model to the intermediary model, among the
intermediary models and from the intermediary
model to the target model.
The first and third steps just transform the format
of the model (the content and concepts do not
change). The second transformation step contains
three phases: first, using syntax check to change the
syntax part of the source model; next, with the help
of the “ontology” (which contains domains-cross
knowledge), apply the semantic check rules on the
intermediary model to transform the content and
concepts; finally, using the syntax check again to
transform the intermediary model to its final version.
The providers of the source models could check the
intermediary models here and valid the process. The
syntax and semantic check rules are applied during
the transformation process to make the transform
mappings, and all the check rules work on the
intermediary models.
Figure 3: Transformation process.
Figure 3 explains the main steps to define an
automatically model transformation process. In
order to use the syntax and semantic check rules, the
source meta-model and target meta-model should
conform to a generic meta-meta-model. We use the
combination of source model, target model and their
original meta-models, to deduce the meta-models
respecting to our meta-meta-model. (In next
subsection, we will explain the principles of our
meta-meta-model and how can we use it to generate
the other meta-models.) Then, semantic and syntax
check rules are applying several times, to transform
the source model to the target model through
intermediary models. At the end of this loop, the
provider of the source model could check the
intermediary model to see if it is identical to the
source model or not.
4 KEY ISSUES
This section illustrates two of the key aspects within
our solution. They are: the definition of the meta-
meta-model and the semantic check rules used on
the transformation process.
4.1 The Meta-Meta-model
The explanation of the meta-meta-model will be
given with the help of figure 4. This meta-meta-
model works on the top abstract level of all the other
models.
As shown in figure 4, there are ten core elements
in this meta-meta-model.
“Environment”, describes the context of a system
such as crisis situation, supply chain, etc. If two
“Environments” describe the same context of a
specific situation, the relationship between them is
“same”; otherwise, the two “Environments” are
different.
“Model” is the core concept in this meta–meta-
model. In the context of our solution, every source,
target models and their meta-models (deduced or
imported) could be regarded as “Model”. Every
model contains the component: “Element”.
“Element” could contain another “Element” (e.g.
in BPMN (White, 2004) modelling context, a pool
contains a lane). The “Element” has two
inheritance classes: “Concept” and “Link”.
“Concept” stands for an object; it is used to
describe a subject that exists in the world.
“Link” is the relationship between Elements.
Every “Link” has two ends (there are two
relationships between “Link” and “Element”:
“from”, “to”). “Element” contains “Property”.
“Property” is used to identify and explain the
object that contains it. Each “Property” has a “Data
Type”.
SemanticApproachtoAutomaticallyDefinedModelTransformation
343
“Data Type” should be a “Primitive Type” or an
“Enumeration”.
The most important part of this meta-meta-model
is the element “SemanticRelation”. It helps to
express the semantics relations between elements.
“Environment”, “Model”, “Element” and “Property”
inherit from this abstract class. This means that any
items from these class may have “sameAs” or “near”
relation with the other items.
Figure 4: The meta-meta-model overview.
4.2 Semantic Mapping Approach
In practice, semantic check principles have been
used in many different domains to solve real
problems (we explained this point in section 2). For
the model transformation domain, semantic check
methods can also help.
We rely on the existing semantic check rules
defined in (Boissel-Dallier, 2012). Here, only the
core idea of the semantic check rules will be shown.
The detail of the algorithms and dealing process of
the semantic check rules will not be illustrated.
The basic idea is: in order to do semantic
matchmaking between models from different
domains; a common semantic profile (Boissel-
Dallier, 2012) should be defined first. According to
the ontology we created, we define this semantic
profile. Based on the semantic profile, we can
compute the semantic distance measurement
between the elements from the source model the
elements from the target model. After getting the
computed results, we can do the matchmaking
between the two models.
Here, we define the algorithms that are used to
compute the “sameAs” or “near” relationship for the
objects of “semanticRelation” class. In practice, we
compute the average semantic relation value
between source meta-model and target meta-model
within five groups: “Environment”, “Model”,
“Concept”, “Property” and “Link”. The parameters
that we use in these algorithms are assumed for the
first test. The details of calculating these average
values shown as follow:
1) For the “Environment”
This classification depends on the users who provide
the source and target model. The ontology (create
based on the structure of the meta-meta model)
records all the categories of the imported
“Environment”. The semantic relation between two
“Environments” is “sameAs” or “different”.
E_SR_V =
If the source model and the target model come from
the same “Environment”, then this value is “1”. If
not, this value is “0”.
2) For the “Model”
The average semantic relation value between two
models (deduced source meta-model and deduced
target meta-model) could be calculated using the
formula.
M_SR_V = 0.5*E_SR_V + 0.4*SR_Name
+ 0.1*Num_Concept
The “E_SR_V” is the value calculated from the first
formula. The “SR_Name” is the semantic relation
between the names of the two models; it can be
calculated using existing word recognition algorithm.
The “Num_Concept” is the number of “Concept
involved in a model. Using this formula, the
semantic relation value, which for the “Model”
(defined in the meta-meta-model), is computed.
3) For the “Property”
According to our generic meta-meta-model,
“property” is component of “Concept” and “Link”.
Each “property” (in deduced source meta-model)
has a semantic relation value with every “property”
in the target meta-model, respectively. The formula
1 if “sameAs”
0 if “differen
t
(1)
(2)
MODELSWARD2014-InternationalConferenceonModel-DrivenEngineeringandSoftwareDevelopment
344
for this is:
P_SR_V = 0.3*SR_Name + 0.2*type
+ 0.5*value
Here, the value of the “SR_Name” is calculated in
the same way as explained above. If the “type” of
the two “property” is the same, its value is “1”;
otherwise, its value is “0”. The same calculation rule
is used on “value”.
4) For the “Concept”
“Concept” is the core element in the meta-meta
model, the formula for calculating the semantic
relation between two “Concepts” is:
C_SR_V = 0.3*SR_Name + 0.6*SR_Pro
+ 0.1* M_SR_V
In this formula, the “SR_Pro” parameter is
calculated using the following formula:
SR_Pro =
∗
__

__
In this algorithm, the number of properties of both
source model concept “Num_SP” and target model
concept “Num_TP” should be calculated first. Then,
select the max value of each “P_SR_V” (between
the properties of source concept and the target
concept), and add them together. For example, to
calculate “C_SR_V” between “Concepts A” and
“Concept B”; “Concept A” has two properties while
“Concept B” has three properties. Each property of
“Concept A” has a “P_SR_V” with each property of
“Concept B”. The max value of each pair will be
selected and added together.
5) For the “Link”
The semantic relation value also computed for the
“Link”. The formula for this is:
L_SR_V = 0.1*SR_Name + 0.35*SR_FC
+ 0.35* SR_TC + 0.2* P_SR_V
In this formula, the “SR_SC” stands for the semantic
relation value between the two “from concept”
(every link has two concepts as two ends). The
“SR_TC” means semantic relation value between
the two “to concept”.
With the help of these six formulas, the
definition of model transformation process (mapping
rules) could be automatically generated. To explain
this idea more clearly, a case study based on this
algorithm will be illustrated in next section.
5 CASE STUDY
At this moment, we have defined the meta-meta-
model, and illustrated the algorithm used to calculate
the semantic relation value (which can provide help
to the automatically definition of model
transformation process).
In this section, a use case aiming at transforming
the “UML (Fowler, 2004)” model (here, we just use
a UML class model) to the “OWL (McGuinness and
Van Harmelen, 2004)” model, will be shown. This
case concerns a part of the whole transformation
process: using the input models to deduce the meta-
models that conform to the generic meta-meta-
model and calculating the semantic relation value
between the two models. The process of deducing
the meta-models (conform to our meta-meta-model)
from the input models is shown in the following
tables. Two very simple models are created for this
case (using “UML” and “OWL”, respectively).
The “UML” model shows in figure 5.
In this model, there are three classes: “Student”,
“Teacher” and “Course”. The relationship between
“Student” and “Course” is “to_choose”; the
relationship between “Teacher” and Course” is
“to_give”. There is a total of eight properties for the
three classes.
Figure 5: UML class model.
The “OWL” model is shown in figure 6.
According to the meta-meta model, we deduce
the meta-models for both the “UML” model and the
“OWL” model. Both of them are “Model”; all the
classes stand as “Concept”, the properties of these
classes could be regarded as “Property” and the
“Link” in the meta-meta model replaces the
relationships between the classes.
The “Environments” of the two models are
similar. Table 1 shows the E_SR_V.
(3)
(6)
(5)
(4)
SemanticApproachtoAutomaticallyDefinedModelTransformation
345
Figure 6: OWL ontology model.
Table 1: E_SR_V of this case.
Environment UML
OWL 1
Table 1 shows the E_SR_V value between “UML” and
“OWL” environment is “0”.
After calculating the E_SR_V, the next step is to
calculate the M_SR_V, table 2 shows this (the
algorithm used is illustrated above).
Table 2: M _SR_V of this case.
Model UML
OWL 0.42
The most complex step is to calculate the P_SR_V
value, table 3 shows this process.
Table 3: P _SR_V of this case.
Property lid lname pid pname sid sname
courseId 0.26 0.2 0.26 0.2 0.26 0.2
courseName 0.2 0.32 0.2 0.32 0.2 0.32
studentName 0.2 0.32 0.2 0.32 0.22 0.34
studentId 0.26 0.2 0.26 0.2 0.28 0.22
studentAdd 0.2 0.2 0.2 0.2 0.2 0.2
teacherId 0.26 0.2 0.26 0.2 0.26 0.2
teacherName 0.2 0.32 0.2 0.32 0.2 0.32
teacherEmail 0.2 0.2 0.2 0.2 0.2 0.2
In this case study, all the properties’ type is “String”,
and they have no value (just on “class” level, no
objects exist) because the use-case should be as
simple as possible (just illustrating how to use the
algorithm mentioned above). So, the value of P
_SR_V for each pair of property has no practical
significance.
Based on the P _SR_V value (that known from
table 3), the C_SR_V value could be calculated.
Table 4 shows the result of this process.
Table 4: C_SR_V of this case.
Concept Lecture Student Professor
Student 0.232 0.548 0.232
Teacher 0.232 0.232 0.472
Course 0.532 0.232 0.232
According to the records of this table, the mapping
rules for the model transformation (on class level)
could be made. After getting all the C_SR_V values,
the final step is to calculate the L_SR_V value. In
this case, there are two links (just has a name, they
do not have properties).
Table 5: L_SR_V of this case.
Link select teach
to_choose 0.428 0.267
to_give 0.267 0.36
Based on all the values recorded in these five tables
above, the model transformation process could be
automatically defined. We define the mapping rules
between the concepts and links of the source and
target models or on their meta-model levels. We can
search the recorded table above and select the
maximum average semantic value for each pair of
concepts and links (from source model and target
model, respectively). Then, build the mappings
between such kinds of pairs. After getting all the
mapping pairs, the model transformation rules are
automatically defined.
In practical, when creating these five tables, a
specific ontology should provide data basis, and pre-
define rules would be used to judge the semantic
relation.
6 CONCLUSIONS
This paper exposes an approach to automatically
define model transformations. Comparing to the
existing methodologies and principles of this field,
the main contribution of our work is to add the
semantic check rules on the transformation process.
In order to apply semantic check rules on models,
we create a generic meta-meta-model. Furthermore,
based on this meta-meta-model, we build ontology
to provide the data basis for the semantic check rules.
Automatically defining model transformation
process is a big challenge, it will bring great help to
solve the complex practical problems (reduce human
efforts: avoid the repetitive work) quickly. At this
moment, researchers have already been focused on
finding solutions to do the semi-automatic model
MODELSWARD2014-InternationalConferenceonModel-DrivenEngineeringandSoftwareDevelopment
346
transformations for some specific domains (UML
models to Database models, BPMN models to UML
models, etc.). Our proposal aims at solving the
model transformation more efficient and general
(regardless the domains that the models come from
and the modelling languages). The idea that we
proposed above needs to be improved and it can give
some inspirations to the other model transformation
methods at the same time.
The further work of our proposal focuses on two
aspects: fulfilling the ontology and improving the
efficiency of the algorithms (doing the semantic
check). The more valuable information stored in the
ontology, the more precise transformation mapping
rules could be made. More reasonable semantic
detection algorithms can also improve the accuracy
of the mapping rules.
REFERENCES
Agrawal, A., Karsai, G., Shi, F., 2003. Graph
Transformations on Domain-Specific Models. Under
consideration for publication in the Journal on
Software and Systems Modeling.
Bénaben, F., Mu, W., Truptil, S., Pingaud, H., 2010.
Information Systems design for emerging ecosystems.
4th IEEE International Conference on Digital
Ecosystems and Technologies (DEST).
Bézivin, J., 2006. Model Driven Engineering: An
Emerging Technical Space. Generative and
Transformational Techniques in Software Engineering
Lecture Notes in Computer Science Volume 4143, pp
36-64.
Boissel-Dallier, N., 2012. Réconciliation sémantique des
données et des services mis en oeuvre au sein d'une
situation collaborative. Les thèses en ligne de l'INP.
Bollati, V., Vara, J. M., Jimenez, Á., Marcos, E., 2013.
Applying MDE to the (semi-)automatic development of
model transformations. Information and Software
Technology, Volume 55, Issue 4, Pages 699–718.
Castro, D. V., Maros, E., Vara, J. M., 2011. Applying
CIM-to-PIM model transformations for the service-
oriented development of information systems.
Information and Software Technology, Volume 53,
Issue 1, Pages 87–105.
Chen, D., Doumeingtsb, G., Vernadatc, F., 2007.
Architectures for enterprise integration and
interoperability: Past, present and future. Computers
in Industry, Volume 59, Issue 7.
Czarnecki, K., Helsen, S., 2003. Classification of Model
Transformation Approaches. OOPSLA’03 Workshop
on Generative Techniques in the Context of Model-
Driven Architecture.
Del Fabro, M. D., Bézivin, J., Jouault, F., Breton, E., 2005.
AMW: A Generic Model Weaver. 1ère Journées sur
l'Ingénierie Dirigée par les Modèles: Paris.
Del Fabro, M. D., Valduriez, P., 2008. Towards the
efficient development of model transformations using
model weaving and matching transformations.
Software & Systems Modeling, July 2009, Volume 8,
Issue 3, pp 305-324.
Fowler, M., 2004. A brief guide to the standard object
modelling language. UML Distilled Third Edition.
Grangel, R., Bigand, M., Bourey, J. P., 2010.
Transformation of decisional models into UML:
application to GRAI grids. International Journal of
Computer Integrated Manufacturing, Volume 23,
Issue 7.
Jouault, F., Allilaire, F., Bézivin, J., Kurtev, I., 2007. ATL:
A model transformation tool. Science of Computer
Programming, Volume 72, Issues 1–2.
Lano, K., Kolahdouz-Rahimi, S., 2013. Constraint-based
specification of model transformations. Journal of
Systems and Software, Volume 86, Issue 2.
Ly, L. T., Rinderle, S., Dadam, P., 2006. Semantic
Correctness in Adaptive Process Management Systems.
Business Process Management, Lecture Notes in
Computer Science Volume 4102, pp 193-208.
McGuinness, D. L., Van Harmelen, F., 2004. OWL Web
Ontology Language Overview. W3C Recommendation.
Object Management Group, 2002. MOF 2.0 Query /
Views / Transformations RFP. OMG Document.
Rajsiri, V., Lorréa, J. P., Bénaben, F., Pingaud, H., 2010.
Knowledge-based system for collaborative process
specification. Computers in Industry, Volume 61,
Issue 2, Pages 161–175.
Taentzer, G., Ehrig, K., Guerra, E., Lara, D. J., Lengyel,
L., Levendovsky, T., Prange, U., Varro, D., Varro-
Gyapay, S., 2009. Model transformation by graph
transformation: A comparative study. In Proc.
Workshop Model Transformation in Practice.
White, S. A., 2004. Introduction to BPMN. IBM
Cooperation.
SemanticApproachtoAutomaticallyDefinedModelTransformation
347