De ned as a problem-relevant, explicit and formal speci cation of a shared conceptualization, ontologies became a new hype in the context of the Semantic Web. Being a shared knowledge its potential for information integration in the large World Wide Web is promising. But either the reuse of existing ontologies or the matching of di erent ontologies is unavoidable for this integration. Therefore means for analyzing ontologies as well as modularization techniques for partial reuse are very important and a key for the success of information integration based on ontologies. Considering ontologies as networks of concepts connected through properties, this work makes use of network analysis techniques and graph measures. It aims at gaining insight to which extent structure based techniques can be modi ed so they are paying attention to the semantics inherent in ontologies. The expected contribution is a method and tool support for ontology engineers to analyze and modularize ontologies in a (semi-) automatic way. The main goal is to improve the (re-)usability and maintainability by increasing the understandability and allowing ontology engineers to refactor and reuse existing ontologies easily.
During the last two decades the interest in using ontologies has increased.
According to the last few years this trend was mainly driven by the vision of the
Semantic Web [
Even though most of these approaches mention the reuse of existing ontologies as possible starting point, none of them describe in detail how to discover and analyze candidate ontologies. This is very important, because reusing ontologies presumes availability of already existing ontologies and discovery of potential candidates for the particular use case. In this regard Ontolingua and OntoSelect libraries are available and search engines as Swoogle1, Watson2 and Ontosearch3 has been already developed in the context of the Semantic Web. Although the problem of discovering potential candidate ontologies seems to be mainly solved, there is still an issue on selecting appropriate ontologies as well as understanding and analyzing them. Even though the Resource Description Framework (RDF) and the Web Ontology Language (OWL) les are based upon the Extensible Markup Language (XML) syntax, which is declared to be human readable, it takes some time to comprehend the content and the main structure and to understand the main idea and purpose of the model. Even the Friend of a Friend (foaf) vocabulary4 which is rather small shows in its speci cation a grouping of the concepts as illustrated in Figure 1, in order to provide the reader an easier way to understand this vocabulary.
In case of ontologies with hundreds and thousands of concepts (SUMO5: 965 concepts, DBPedia6: 934 concepts ) it is nearly impossible for the human mind to overview the whole model. But this is essential to decide if a candidate ontology is really useful and whether it needs some customization. 2
Ontologies are semantic models with di erent expressiveness levels which are represented in RDF and OWL. A structure-based approach to analyze and modularize these semantic models need to tackle some issues during the development process. This section should provide a brief overview about open research questions which need to be solved during the development process. The rst part of the research questions are derived from the ontologies itself and their properties, while the second part focuses on the ontology engineers which are addressed as the users of this framework. 2.1
RDF allows to create structured information as triples following the form
(Subject, Predicate, Object). The graph syntax of RDF allows to represent triples
as graphs where the subjects and the objects are nodes and the predicates are
directed edges (from subject to object). At this level the inherent semantic of
OWL ontologies are not taken into consideration. Furthermore, the nodes and
edges have di erent types, which are re ected in the labels (namespace and
localname), which is a problem for standard Social Network Analysis approaches
[
Some predicates as \hasLabel" or \hasComment" have an impact on the structural values. That is, their centrality values might be very high. It is very important to lter such concepts, which have an impact on the structural analysis but are not necessary to understand the content of an ontology. Furthermore, it is important to take di erent namespaces into consideration. It might be useful, to consider concepts from one namespace as a class of nodes and to analyze the connectedness of nodes from namespace to nodes of di erent namespaces.
Other open research questions derived from the graph representation of ontologies based on RDF and OWL are: 5 http://www.ontologyportal.org/translations/SUMO.owl 6 http://dbpedia.org/ontology 1. RDF speci cation allows to create blank nodes, which have an in uence on the structure of the graph. How should these blank nodes be handled? 2. Ontologies allow reasoning which leads to a change in the structure of the ontology. Should these changes taken into account. That means, should a reasoning process executed before the structure-analysis process starts? 3. Ontologies might be expressed in di erent expressiveness levels (OWL Lite, OWL DL, OWL FULL). What is the impact of the expressiveness on the structure of an ontology? 4. Besides the schema ontologies represented in OWL may include instances.
Is it important to take them into account? In which cases do they have to be considered and which cases not? 2.2
During the last years there is an increasing interest in the usability aspect of software products. For the success of a framework it is very important to take the addressed users' needs into account during the design process. Because this works addresses ontology engineers, their expectations of an ontology analysis and modularization framework have to be obtained and used as guidelines for the design process. This issue needs further investigation. At this point following questions have been identi ed: 1. How important are realtime and interactivity for ontology engineers? 2. What are the needs of ontology engineers, that have to be taken into account by developing an ontology analysis and modularization framework? 3. Ontology Engineering methodologies are mostly heavyweight. How can the framework be used in di erent methodologies? 3
This work is based on the belief that the utilization of semantic models would
improve the quality of information systems and would enable interoperability
in distributed open systems as the Web. But creating ontologies from scratch
as well as analyzing, reusing and maintaining existing ontologies are complex
tasks which are hindering broad acceptance and application of ontologies. This
is identi ed as the main problem, which this work tries to solve by developing and
implementing methods as well as techniques to analyze and modularize
ontologies. Regarding the de nition "The design science paradigm seeks to extend the
boundaries of human and organizational capabilities by creating new and
innovative artifacts" [
Considering ontologies as networks of concepts connected through properties, network analysis techniques and using network measures (e.g. node centrality, betweenness, density, similarity) are a promising approach to analyze and modularize ontologies. As a very well established discipline in science there are a lot of sophisticated methods and tools for network analysis available. We believe that these methods can be modi ed, extended (in order to take the semantics into consideration) and applied to ontologies, so that the ontology structure can be used to analyze the content and to identify regions, which can be seen as network \communities" and can be extracted as modules. Furthermore, we are convinced that structure analysis enables a rst evaluation of the usability by allowing di erent views, so that existing ontologies can be easier comprehended by ontology engineers. This is very important because refactoring and reusing of existing models assume that these models are understood.
The foundation of this work is the hypothesis, that analyzing and
modularization of an ontology can be done in an e cient manner, by using
structural information about the ontology. Some previously done related work have
shown that this approach is promising. Structural analysis in [
Therefore this work investigates on the application of network analysis techniques and network measures (e.g. node centrality, betweenness, density, similarity) to ontologies and aims at gaining insight to which extent structure based techniques can be modi ed so they are paying attention to the semantics inherent in ontologies. The expected contribution is a method and tool support for ontology engineers to analyze and modularize ontologies in a (semi-) automatic way. The main goal is to improve the usability and maintainability by increasing the understandability and allowing ontology engineers to refactor and reuse existing ontologies easily. 4.1
The current development of the artifact is at a very early state. As a very well known integrated development environment Eclipse allows to implement functional extensions through plugins. In this regard we have identi ed functional components which can be implemented as Eclipse plugins so an Ontology Modularization and Integration framework can be realized. Figure 2 illustrates the architecture of this framework. Based on the developed architecture, the decision was made to reuse the SONIVIS:Tool7 to realize the targeted system. The SONIVIS:Tool is a network analysis software which is based upon Eclipse and allows easy extension through the Eclipse Plugin system. It provides already the Graph Analysis and the Visualization components and makes use of the Eclipse User Interface. Figure 3 illustrates the foaf vocabulary where the node size depends on the node degree.
The biggest nodes in the visualization are \Agent", \Document", \Person", \OnlineAccount", and \Organization". If these concepts are compared with the concept groups (especially the group names) from the speci cation as illustrated in Figure 1 it obvious that there is a similarity. The group names \Personal Info, \Documents and Images" and \Online Accounts"contain some of the concepts which have a high centrality in the structure. This rst insight is an indication for the applicability and usability of the chosen approach and justi es further investigation.
Research activities always have to be validated in order to measure its quality. Design science can make use of di erent evaluation approaches to evaluate the outcome. The most popular approaches are case study, professional review, goalfree evaluation, and goal-based evaluation.
At this stage of this work it has not been clari ed in detail how the
outcomes can be evaluated. The rst ideas are to evaluate the ontology analysis
aspect through professional reviews of di erent ontology engineers. The
important question is whether these ontology engineers gain new insight about their
ontologies when they are using this framework. As their personal opinion
cannot really be quanti ed and objectively compared it is still an open question,
in which degree this is really applicable. For the modularization of ontologies it
is intended to apply di erent ontology evaluation methods as [
The goal-free evaluation is mainly a comparison activity of di erent solutions for the same problem based on some pre-de ned criteria. Based on an in depth literature work about the state-of-the-art these criteria needs to be de ned. In contrary, the goal-based evaluation focuses on the designed artifact itself. It gives a quali ed view on the achievements of the proposed solution. The requirements which have been identi ed by the developer during the problem analysis process are used to evaluate to which extend they have been truly achieved. For this approach the problem to be solved have to be analyzed deeply and the requirements which have to be ful lled by the artifact have to be formulated concretely. 6
The vision of the Semantic Web brought new attention to ontologies by underlining its knowledge sharing aspect. Ontologies are considered as the most important means for information integration in the highly distributed and open Web. But either the reuse of existing ontologies or the matching of di erent ontologies is unavoidable for this integration. Therefore means for analyzing ontologies as well as modularization techniques for partial reuse are very important and a key for the success of information integration based on ontologies. Following the design science research methodology this work is grounded on the hypothesis, that analyzing and modularization of an ontology can be done in an e cient manner, by using structural information about the ontology.
As this work is still in progress the design and development process is ongoing and there are open research questions (see Section 2) which need further investigation. It is also expected that new questions will arise. Additionally, as mentioned in Section 5 it is still an open issue how this work is going to be evaluated. Use cases as well as criteria for goal-based and goal-free evaluation needs to found and de ned.
||||{ Acknowledgements This work has been partially supported by the \InnoPro leCorporate Semantic Web" project funded by the German Federal Ministry of Education and Research (BMBF) and the BMBF Innovation Initiative for the New German Lander - Entrepreneurial Regions.