SEMA: Results for the Ontology Alignment Contest OAEI 2007

In this paper we present SEMA tool for the automatic mapping of ontologies. The main purpose of SEMA is to locate one to one equivalence correspondences (mappings) between elements (i.e., classes and properties) of two input ontologies. Towards this goal, SEMA synthesizes lexical, semantic and structural matching algorithms through their iterative execution. 1 Presentation of the system 1.1 State, purpose, general statement Ontologies have been realized as the key technology to shaping and exploiting information for the effective management of knowledge and for the evolution of the Semantic Web and its applications. In such a distributed setting, ontologies establish a common vocabulary for community members to interlink, combine, and communicate knowledge shaped through practice and interaction, binding the knowledge processes of creating, importing, capturing, retrieving, and using knowledge. However, it seems that there will always be more than one ontology even for the same domain. In such a setting, where different conceptualizations of the same domain exist, information services must effectively answer queries, bridging the gaps between conceptualizations of the same domain. Towards this target, networks of semantically related information must be created at-request. Therefore mapping of ontologies is a major challenge for bridging the gaps between agents (software and human) with different conceptualizations. 1 This work is part of research project ONTOSUM (www.ontosum.org), implemented within the framework of the “Reinforcement Programme of Human Research Manpower” (PENED) and co-financed by E.U.-European Social Fund (75%) and the Greek Ministry of Development-GSRT (25%). Tools for the automated mapping of ontologies have achieved remarkable results but still there is lot of space for improvements when dealing with real world ontologies. Building on our experience in participating in OAEI 2006 with AUTOMS [1], we intent to further increase the precision and recall of our matching methods, and further minimize the efficiency cost, by devising enhanced techniques and combinations of methods. This paper presents the SEMA tool for the mapping of ontologies. SEMA is built on top of AUTOMS-F [2], which a framework implemented as a Java API, aiming to facilitate the rapid development of tools for the automatic mapping of ontologies. AUTOMS-F provides facilities for synthesizing individual ontology matching methods. The main purpose of SEMA is to locate one to one equivalence correspondences (mappings) between the elements (i.e., classes and properties) of two input ontologies, by increasing the recall of the mapping process and achieving a fair balance between precision and recall. SEMA combines lexical, semantic and structural matching algorithms: A semantic matching method exploiting Latent Dirichlet Allocation model (LDA) [3], requiring no external resources, in combination with the lexical matcher COCLU (COmpression-based CLUstering) [4] and a matching method that exploits structural features of the ontologies by means of simple rules. This combination of approaches contributes towards automating the mapping process by exploiting lexical, structural and semantic features of the source ontologies, resulting to increased recall and precision. It must be emphasized that the aggregation of the mappings produced by the individual methods is performed through their iterative execution as described in [5, 6]. It must be pointed that the experience gained by participating in the OAEI contest helped us towards the following aspects: (i) We increased the precision and recall of SEMA by iteratively combining the individual matching methods, (ii) we improved AUTOMS-F framework by adding more facilities towards the synthesis of individual matching methods, (iii) we noticed the fact that tools such as SEMA tend to fail to notice subsumption relations between elements of distinct ontologies, since they assess only equivalences between them, and finally, (vi) we managed to improve the execution time of matching methods, such as the one based on LDA. 1.2 Specific techniques used Fig. 1. Overview of SEMA. SEMA combines six matching methods, executed in a predefined sequence, as depicted in Fig. 1. Each method in sequence exploits the results of the previous methods, aiming to find additional mapping element pairs. This policy is applied as Lexical Matcher Latent Features Matcher VSM Matcher Instance Based Matcher Property Based Matcher Structure Based Matcher Input Ontologies