A Framework to Combine MDA and Ontology Learning

Regina C. Cantele

, Maria Alice G. V. Ferreira

, Diana F. Adamatti

and Moyses Araujo

InterLab - Laborat´orio de Tecnologias Interativas

Escola Polit´ecnica da Universidade de S˜ao Paulo, S˜ao Paulo, Brazil

Faculdade de Tecnologia - FTEC, Caxias do Sul, Brazil

IBTA, Brazil

Abstract. This paper proposes to join two different approaches: Software Engi-

neering and Semantic Web. The ﬁrst one cames from Model Driven Architecture

(MDA) and the second one, from the Ontology Engineering area, more speciﬁ-

cally Ontology Learning. The main idea of this work is to accelerate the initial

construction of ontologies from knowledge already represented in several differ-

ent models such as CASE repository or database dictionaries and text, based on

standards of MDA for interoperability. In this way, a framework was developed

to join these concepts, where the steps sequence to apply it were deﬁned until an

initial representations of ontology was obtained.

1 Introduction

Nowadays, the Web is an important knowledge source and people use it to ﬁnd relevant

information for their needs. However, Web information is unstructured and placed in

such a way that it is inconvenient for searching. Web pages are simple collections of

characters, and the agents - the knowledge collectors - may be unaware that the infor-

mation they need is available. In the mid 1990s, Berners-Lee and colleagues deﬁned the

Semantic Web as an alternative approach to Web, where information can be structured

and used by the agents. Semantic Web use metadata and ontologies for this task [4].

In Computer Science, ontologies started to be used by the Artiﬁcial Intelligence

(AI) area, mainly in Natural Language Processing (NLP) and Machine Learning (ML).

However, many works in these areas fail to provide interoperability between works.

In Software Engineering, Model-Driven Engineering (MDE) is an approach to the

development of software that separates the speciﬁcation of the system functionality

from its implementation on a particular platform. In this way, it can share models and

concepts between different tools and methods [29]. One of the major initiatives through

which the MDE is accomplishing this goal is by the MDA, deﬁned by Object Man-

agement Group (OMG), as a comprehensive interoperability framework for deﬁning

future interconnected systems. Another important speciﬁcation is Architecture-Driven

Modernization (ADM) - the process of understanding and evolving existing software

C. Cantele R., Alice G. V. Ferreira M., F. Adamatti D. and Araujo M. (2008).

A Framework to Combine MDA and Ontology Lear ning.

In Joint Proceedings of the 5th International Workshop on Ubiquitous Computing (IWUC 2008) 4th International Workshop on Model-Driven Enterprise

Information Systems (MDEIS 2008) 3rd Inter national Workshop on Technologies for Context-Aware Business Process Management (TCoB 2008),

pages 43-51

DOI: 10.5220/0001730400430051

 SciTePress

artifacts and which details the use of the tools of progressive engineering with the tools

of reverse engineering. Among the this speciﬁcations, it is important to point out: Meta

Object Facility (MOF); XML Metadata Interchange (XMI); and ODM [1].

MOF is an extensible model-driven integration framework for deﬁning, manipulat-

ing and integrating metadata and data in a platform independent manner. MOF-based

standards are used for integrating tools, applications and data [26]. In MOF terminol-

ogy, metadata that describes metadata is called meta-metadata, and a model that con-

sists of meta-metadata is called a metamodel. A MOF metamodel deﬁnes the abstract

syntax of the metadata in the MOF representation of a model. The MOF integrates

these metamodels by deﬁning a common abstract syntax for deﬁning metamodels. This

abstract syntax is allied to the MOF Model and is a model for metamodels; i.e. a meta-

metamodel. The classical framework for metamodeling is based on an architecture with

four meta-layers: M3 - meta-metamodel,M2 - metamodel, M1 - model e M0 - instances

of model.

XMI is a model-driven XML Integration framework for deﬁning, interchanging,

manipulating and integrating XML data and objects. XMI-based standards are used for

integrating tools, repositories, applications and data warehouses. XMI provides rules

which can generate a schema for any valid XMI-transmissible MOF-based metamodel.

XMI provides a mapping from MOF to XML [26].

The speciﬁcation Ontology Deﬁnition Metamodel (ODM) joins the Semantic Web

with MDA. ODM is formal grounding for representation of business semantics and de-

ﬁnes independent metamodels, related proﬁles, and mappings among the metamodels

corresponding to several international standards for ontology and Topic Maps deﬁni-

tion, as well as capabilities supporting conventional modeling paradigms for captur-

ing conceptual knowledge, such as Uniﬁed Modeling Language (UML) and Entity-

Relationship (E/R) modeling.

However, ontologies need to be constructed and populated, which is a hard task.

Automatic machines acquisition is an open problem. Therefore, semi-automatic pro-

cesses with human intervention are widely used. A lot of tools as well as standards

allowing systems interoperability have been developed throughout the last years for

the accomplishment of such task. In this approach, Ontology Learning appears as a set

of algorithms and tools to automatically derive knowledge from domain-speciﬁc text

collections or unstructured textual resources [22] [17].

The goal of this paper is to develop a framework to combine the Software Engineer-

ing concepts (based on MDA approach) and Ontology Learning to build new ontologies

for Semantic Web, using knowledge already represented (e.g., legacy systems). This

framework is composed by blocks, where requirements and steps are deﬁned.

The paper is structured in 5 sections. In sections 2 and 3, a brief overview of Ontol-

ogy Deﬁnition Metamodel (ODM) and Ontology Learning are presented, respectively.

Section 4 presents the developed framework to combine Ontology Learning and MDA.

Finally, the conclusions are presented in Section 5.

2 Ontology Deﬁnition Metamodel

According to [16], there are ﬁve components in ontology: concepts, relations, function,

axioms, and instances. Concept can be anything said about something, and therefore,

it could also be the description of a task, function, action, strategy, reasoning process,

properties, etc. Relations represent the type of interaction between domain concepts.

Functions are special case of relations in which the n-th element of the relationship is

unique for the n-1 preceding elements. Axioms are used for modeling sentences are

taken for granted as true. Instances are used for representing elements.

ODM is a standard to support ontology development and conceptual modeling in

several standard representation languages. It provides coherent framework for ontology

creation based on MOF and UML. Knowledge engineering requirements may include

some ontology development for traditional domain, process, or service ontologies, but

they may also include: generation of standard ontology descriptions (e.g., OWL) from

UML models; generation of UML models from standard ontology descriptions (e.g.,

OWL); integration of standard ontology descriptions (e.g., Ontology Web Language

(OWL)) with UML models [26]. Moreover, the metamodel increases with constraints.

These constraints are expressed in the Object Constraint Language(OCL), a declarative

language which provides constraint and object query expressions on object models that

cannot be otherwise expressed by diagrammatic notation.

All MDA speciﬁcations contribute to the interoperability of the constructed dia-

grams already with the necessary new representations. Some researchers joined the

existing ontology representation formalism to the MDA of OMG [2,15,32,20].

3 Ontology Learning

Ontology construction is a complex task and needs some criteria and methodologies

to be developed. Jimenez-Ruiz and Berlanga (2006)[19] studied several methodologies

and classiﬁed them into two groups: centralized or collaborative approach in the de-

velopment of ontologies. The ﬁrst group has the examples Methontology [12], Uniﬁed

Process for Ontology Building (UPON ) [10] and CommonKADS [28]. The second

group has the examples KA2 [3], Human-Centered Ontology Engineering Methodol-

ogy (HCOME)[21] and Diligent [7]. The choice of approach is in accordance with the

type of ontology to be developed. For example, for an ontology of application, the cen-

tralized approach is recommended and for an ontology of domain, the collaborative

development is.

Ontology development can be based on the layer cake built by Cimiano [8], as

shown in Figure 1, where the acquisition of knowledge is executed in subtasks. It is pri-

marily concerned with axiomatizing the deﬁnition of concepts as well as the relation-

ships between them. After that, it is important to connect concepts and relations to the

symbols referring to them. This implies the acquisition of linguistic knowledge of terms

used for referring to a speciﬁc concept and potential synonyms of these terms. More-

over, ontology may consist of some concept hierarchy or any other non-hierarchical re-

lation. In order to constrain interpretations of concepts and relations, axiom schemata,

such as disjointness between concepts as well as symmetry, reﬂexivity, transitivity, etc.,

can be instantiated for relations. Finally, one is also interested in using an ontology to

derive facts that are not explicitly modeled in the knowledge base but can be derived

from it. For this purpose, logical axioms modeling implications between concepts and

relations can be deﬁned.

Fig.1. Ontology Learning Layer Cake[8].

Another way of ontology construction is Ontology Learning. This work considers

the two approaches in Ontology Learning: from texts and from schematas.

The ﬁrst approach shows to be a set of algorithms and tools for automatically deriv-

ing knowledge from domain-speciﬁc text collections or unstructured textual resources

[8]. Pattern-based techniques are used to identify words that are useful inputs for learn-

ing methods because they are likely to represent concepts or relations [20]. Special

syntactic patterns have been developed for populating the instances identiﬁed in the

ontological relations.

The second approach, Ontology Learning from schematas contemplates the data in

the form of schemata with a formal semantics and are typically represented in models

of data such as E/R, UML and XML. Generally, the techniques of Reverse Engineering

with mappings for representation of ontology languages are used.

A comparison between environments of Ontology Learning can be seen in Shams-

fard and Barfoush [30].

Some tools based on these techniques have been developed, such as, ASIUM [11]

of the University of Paris; OntoLT [5] of the German Research Center for Artiﬁcial

Intelligence; Text2Onto and AEON [9] of the University of Karlsruhe; OntoLearn [24]

of the University La Sapienza of Rome; and Ontogen [14] of the Joef Stefan Institute.

One essential requirement for all methods and tools is that a representative input data

for the domain must be built in the ontology.

4 The Proposed Framework

The main goal of the proposed framework is to bring together the concepts of Software

Engineering with MDA and Ontology Engineering to accelerate the construction of on-

tologies from knowledge already represented. This process comprises four blocks and

associated resources, as shown in Figure 2: Selection of Resources, Ontology Learning,

Environment Ontologies and Ontology Engineering.

Fig.2. Framework to combine Ontology Learning and MDA.

A. Selection of resources. It is the part responsible for representing old resources of

an information system. The knowledge of the existing system is present in the in-

put resources (Standards and Texts - (1b)) and in the output resources (database

dictionary RDBMS, UML Diagrams, Programs, Helps, Texts etc. - (1c)) often rep-

resented in Data Repository - metamodels of the CASE tools used in its conception

and implementation (1d). When the existing system (1a) did not use a tool for its

construction (it is very common), the Data Repository (1d) does not exist, and then,

Reverse Engineering (1e) is necessary to populate this repository (1d);for example,

the database dictionary can be used to get the Class Diagram in UML [13]. This is

represented by the dotted lines in Figure 2.

B. Ontology Learning. In this block, the process of models extraction is carried out

by two tasks: extraction of models with standards MOF/XMI (2a) and Transforma-

tion between metamodels (2c). The process of extraction of models with standards

MOF/XMI can be achieved by means of tools implemented with standard MOF

(2a) so that the ﬁnal result is presented in standard XMI in Metadata XMI (2a1) and

(2b1). The (2c) realizes a core transformation by having an input in E/R or UML

models represented in (2a1) and (2b1) and, producing an ontology according to the

OWL metamodel, resulting in an XML document at OWL/XML syntax: Metadata

OWL (2e). The transformation part is dedicated to the mapping out of the UML

model for ontology, i.e. UML classes are mapped into OWL classes, attributes into

datatype property, associations into object property, instances into OWL individu-

als, etc. This process needs to apply rules deﬁned in the ODM to OWL model (2a2)

and (2b2). It is the transformation of a representation form into another, with the

same level of abstraction, preserving the system external behavior (functionality

semantics). This process offers the possibility to manage E/R and UML instances

and populate the ontology with corresponding knowledge. This process pipeline is

represented in gray lines in Figure 2.

The other process is Ontology Learning from text (3a). This process is responsible

for processing text resources (e. g. HTML pages and papers from a given domain)

and creating a very simple ontology from the extracted data. It considers the use

of some tools - based on techniques of natural language and machine learning -

to get concepts, its associations, and hierarchy in metadata (3b). In general, these

tools have implemented many algorithms for this, with incremental executions until

signiﬁcant concepts were reached. After (3a) locks up, this process of extraction

in standard OWL is realized by obtaining the metadata OWL (3d). The extracted

information(3b) is stored in internal metamodel depending on the applied tools and

can be directly produced by applying speciﬁc algorithms. The requirement of tools

must be in compliancewith metamodel OWL for output documents or provide their

proprietary metamodel for the transformation processes.

C. Environment Ontology. After the accomplishment of process 2 or 3, it is necessary

to use the Manipulation Ontology (4a) to visualize the initial ontology representa-

tions - (2e) or (3d). Thus, the ﬁrst conceptual representation of the ontology could

be edited (4b), in a clearly understandable way for ontology engineers, reusing the

ﬁrst model of the ontology in format E/R and UML (1c) or in text format (1b) and

represented on a Ontology Database (5b).

D. Progressive Engineering. (4b) will be used for executing tasks of Development

Method (5a) in order to obtain the ﬁnal ontology representation.The use of method-

ologies assures the Ontology Base (5b) to be consistent and coherent with the rep-

resented domain and with the current system model. The choice of methodology

is linked to the use of ontology; therefore, an ontology of application development

will be centralized as well as a domain ontology of the collaborative development.

One important aspect in this framework is the use of the OMG metamodels stan-

dards to Ontology Learning and ontologies environment according to Figure 3. At this

point, the proposed framework introduces the use of the MOF and their metamodels

already deﬁned to represent the OWL model with ODM. This ensures that all models

(M1) considered by the different tools have their metamodel shared between the tools

through applied rules processing on metamodels with XMI. Also, all instances (M0)

of the models are shared. To really enjoy the results of the various tools for Ontology

Learning from texts, it is necessary to tailor the metamodel to the OWL metamodel.

This can be done with the creation of tools for conversion between the metamodel ex-

isting non-standardized model for the OWL, or a structural change in the architecture

of the tool to implement the standardized metamodel. Text2Onto [18] and Prot´eg´e [25]

are examples of tools with metamodels own disclosed.

Fig.3. Four-Layer Architecture of MOF.

The practical application of this framework was started in a previous work [6],

where the authors tested the schematas approach to generate new ontologies (phases

1, 2 and 4 in the Figure 2). The tests were conducted with a large number of tools, such

as Prot´eg´e [27], an ontology environment and Magic Draw [23], a CASE tool. Regard-

less of the semantics of each of the models being different (E/R, UML and, Ontology),

the represented knowledge is not lost by the transformations. The use of representative

mnemonics from the existent systems helps to generate knowledge-represented. This

process has shown to be effective when there is a large number of entities or classes

from the legacy system. Another test was conducted with UML diagrams provided by

the IMS Learning Design [31], and 192 classes and 390 properties were obtained from

these UML diagrams.

5 Conclusions and Further Works

This paper presented the conceptualviability of different standardsintegration and tech-

niques associated to the metadata to help Ontology Engineering. Many standards of

OMG (MDA, ADM, UML, MOF, XMI, ODM), as well as many standards of the W3C

(XML, and OWL) were recognized, accepted and used by the Information Technology

community.

The success of the Semantic Web depends on the deployment of ontologies. Using

ontology studies from NLP and ML can make the Engineering Ontologies process eas-

ier and faster, because they just share a metamodel of the common representation from

an ontology as proposed by ODM. In the case of Ontology Learning from schematas,

the process is very simple and it presents good results. Therefore, in the case of Ontol-

ogy Learning from texts, it adopts standards of metamodel and it requires efforts from

the researchers’ community. Ontology Engineering can use the initial representations

so that it can be favored by the use of some already represented knowledge. It must be

taken into account that the semantic terms may vary from one context to another, from

a place to another, and from a person to another. Considering these semantic hetero-

geneities, Ontology Engineering is not a trivial activity and requires time, availability

and consensus by specialists. Therefore, the idea of exploring things that have already

been represented in the past has been taken into consideration in [11,5, 15,2] and in

this article and this idea can help in the construction of an initial model of ontology for

a speciﬁc domain.

The proposed framework starts to be the ontology engineers’ convergence point,

allowing the sharing of represented knowledge and the description of the most common

concepts obtained from the representantions initial ontology. As further works, some

tests will be conducted with Ontology Learning from texts tools (phases 1 and 3 in

the Figure 2), transformation tools and repository for XMI ﬁles with the application

of Knowledge Discovery Meta-Model (KDM) and Abstract Syntax Tree Meta-Model

(ASTM) speciﬁcations. The chosen tools must provide metamodels or generate meta-

models to ADM, ODM and OWL standards and control of the ﬁles intermediaries.

References

1. Omg: Object management group - architecture driven modernization. http://adm.omg.org,

2008.

2. K. Baclawski, M. Kokar, P. Kogut, L. Hart, J. Smith, W. Holmes, J. Letkowski, and M. Aros-

ton. UOL: Uniﬁed ontology language. Assorted papers discussed at the DC Ontology SIG

Meeting, 2002. http://www.omg.org/cgi-bin/doc?ontology/2002-11-02.

3. V. R. Benjamins, D. Fensel, S. Decker, and A. Gom´ez-P´erez. (ka)2: building ontologies

for the internet: a mid-term report. International Journal of Human-Computer Studies,

51(3):687–712, 1999.

4. T. Berners-Lee, J. Hendler, and O. Lassila. The semantic web. Scientiﬁc American,

284(5):28–37, 2001.

5. P. Buitelaar, D. Olejnik, and M. Sintek. A prot´eg´e plug-in for ontology extraction from text

based on linguistic analysis. In Proceedings of the 1st European Semantic Web Symposium

(ESWS), Heraklion, Greece, May 2004.

6. R. C. Cantele, D. F. Adamatti, M. Ferreira, and J. S. Sichman. Aplicac˜ao da reengenharia de

software na construc˜ao acelerada de ontologias. In L. F. Friedrich, editor, Simp´osio Brasileiro

de Sistemas de Informac˜ao (SBSI 2005), Florian´opolis, Brasil, Outubro 2005. SBC.

7. P. Casanovas, N. Casellas, C. Tempich, D. Vrandecic, and V. Benjamins. OPJK and diligent:

ontology modeling in a distributed environment. Artiﬁcial Intelligence and Law, 15(2):171–

186, 2007.

8. P. Cimiano. Ontology Learning and Population from Text: Algorithms, Evaluation and Ap-

plications. Springer, Germany, 2006.

9. P. Cimiano and J. Volker. Text2onto - a framework for ontology learning and data-driven

change discovery. In roceedings of the 10th International Conference on Applications of

Natural Language to Information Systems (NLDB’2005), Alicanet, Spain, June 2005.

10. A. De Nicola, M. Missikoff, and R. Navigli. A proposal for a uniﬁed process for pntology

building: Upon. In Proceedings of the 16 International Conference Database and Expert

Systems Applications, pages 655–664, Dinamarca, 2005.

11. D. Faure and T. Poibeau. First experiments of using semantic knowledge learned by asium

for information extraction task using intex. In Proceedings of Ontology Learning ECAI-2000

Workshop, pages 7–12, 2000.

12. M. Fern´andez, G´omez-P´erez, and N. Jurino. Methontology: From ontological art towards

ontological engineering. In Spring Symposium Series, pages 33–40, 1997.

13. M. Ferreira, M. Araujo, and R. C. Cantele. Educac˜ao baseada na Web e Web Semˆantica:

construindo uma nova forma de educac˜ao a distˆancia. In Proceedings of the WCCSETE’2006

- World Congress on Computer Science, Engineering and technology Education, pages 692–

696, Itanham/Santos, 2006.

14. B. Fortuna, M. Grobelnik, and D. Mladenic. Background knowledge for ontology construc-

tion. In Proceedings of the 15th International conference on World Wide Web 2006, pages

949–950, Edinburgh, Scotland, 2006.

15. D. Gasevic, D. Djuric, and D. Devedzic. Model Driven Architecture and Ontology Develop-

ment. Springer, Berlin Heidelberg, Germany, 2006.

16. A. G´omez-P´erez and R. Benjamins. Overview of knowledge sharing and reuse components:

Ontologies and problem solving methods. In V. Benjamins, editor, IJCAI-99 Workshop on

Ontologies and Problem-Solving Methods: Lesson learned and Future Trends, volume 18,

pages 1–15, Estocolmo, 1999. CEUR Publications.

17. A. G´omez-P´erez and D. Manzano-Macho. A survey of ontology learning methods and tech-

niques. http://ontoweb.aifb.uni-karlsruhe.de/Members/ruben/Deliverable, 2003.

18. P. Haase and J. Volker. Ontology learning and reasoning - dealing with uncertainty and

inconsistency. In Proceedings of the 4 International Semantic Web Conference, Ireland,

2005.

19. E. Jimenez-Ruiz and R. Berlanga. A view-based methodology for collaborative ontology

engineering: an approach for complex applications (vimethcoe). In Proceedings of the 15

IEEE International Workshops on Enabling Technologies: Infrastructure for collaborative

Enterprises, pages 376–381, Myanmar, 2006.

20. F. Jouault and I. Kurtev. On the architectural alignment of ATL and QVT. In In Proceedings

of ACM Symposium on Applied Computing (SAC 06), pages 1188–1195, Dijon - France,

2006. ACM Press. Chapter Model Transformation.

21. K. Kotis and G. Vouros. Human-centered ontology engineering: the HCOME methodology.

International Journal of Knowledge and Information Systems, 10(1):109–131, 2006.

22. A. Maedche and S. Staab. Ontology learning for the semantic web. IEEE Intelligent Systems,

16(2):72–79, 2001.

23. Magicdraw. Magicdraw architecture made simple. http://www.magicdraw.com/, 2007.

24. R. Navigli and P. Velardi. Ontology enrichment through automatic semantic annotation of

on-line glossaries. In Proceedings of the 15th International Conference on Knowledge En-

gineering (EKAW 2006), pages 126–140, Podebrady, Czech Republic, 2006. LNAI 4248.

25. N. F. Noy, R. W. Fergerson, and M. A. Musen. The knowledge model of prot´eg´e-2000:

combining interoperability and ﬂexibility. In Proceedings of the 2 International Conference

on Knowledge Engineering and Knowledge Management, pages 69–82, France, 2000.

26. OMG. Object Management Group - ontology deﬁnition metamodel speciﬁcation.

http://www.omg.org/, 2008.

27. Prot´eg´e. The prot´eg´e ontology editor and knowledge acquisition system.

http://protege.stanford.edu/, 2007.

28. G. Schereiber, M. Crubezy, and M. Musen. A case study in using prot´eg´e-2000 as a case tool

for commonkads. In Proceedings of the 12 International Conference Knowledge Engineering

and Knowledge Management, volume 1937, pages 33–48, Berlin, 2000.

29. D. Schimt. Model driven-engineering. In IEEE Computer Society, pages 25–31, 2006.

30. M. Shamsfard and A. Barforoush. Learning ontologies from natural language texts. Inter-

national Journal of Human-Computer Studies, 60(1):17–63, 2004.

31. Van der Vegt, Wim. The poseidon ﬁle containing the uml proﬁle of ims-ld level c.

http://dspace.ou.nl/handle/1820/659, 2006.

32. T. Win, H. M. Aung, and N. L. Thein. An mda based approach for facilitating representa-

tion of semantic web service technology. In Proceedings of 6th Asia-Paciﬁc Symposium on

Information and Telecommunication Technologies, pages 260–265, Myanmar, 2005.