Towards a Semi-Automatic Ontology Mapping - An Approach Using Instance Based Learning and Logic Relation Mining

To share information and knowledge of heterogeneous systems, one of the key issues is the mapping of their ontologies for interoperability. As manual mapping is too tedious, labor intensive, and prone to errors to support complex applications, investigating (semi-) automatic methods for ontology mapping becomes an imperative need. In recent years automatic ontology mapping arouses researchers¿ interests and valuable progresses have been made. However, there are still few reliable methods that have been successfully employed in real applications. In this paper we present a novel approach for systematically integrating all the available information including labels, instances, structures and former mapping experiences for semi-automatic mapping of heterogeneous ontologies. In this approach, we use four steps in making the decisions for mapping. Firstly we adopt an instance-based matching process that is a slight modification of GLUE to find the possible concept mappings between two ontologies. Secondly we use former indexed experiences that embed domain knowledge to modify the result. Thirdly a graph-based iteration process is executed in order to take the different information and knowledge structures into consideration. To eliminate negative efforts of structure heterogeneity we make a preprocessing for the structure reorganization via modifying the structures of ontologies before the iteration. Finally the attribute correspondences are identified through a mining and matching process for the logic relations among ontologies.