A Bipartite Graph Co-Clustering Approach to Ontology Mapping

The necessity of mapping concepts of one ontology to concepts in a second ontology is an important research topic due to the requirements brought by the Semantic Web. Most ontology mapping techniques available today do not allow the existence of many-to-many correspondences among concepts. To overcome this problem we propose to model two ontologies as a weighted bipartite graph. We assign weights to graph edges using current similarity measure techniques and apply graph partitioning techniques in order to co-cluster the vertex sets of the bipartite graph. Then we use the resulting concept clusters to establish mappings between concepts of the two ontologies. The approach combines mapping methods that rely solely on similarity measures with an unsupervised learning technique called bipartite graph co-clustering. The advantage of the combination is that it allows mappings to have many-tomany concept correspondences.