Analyzing Mapping Extraction Approaches

While lots of research in ontology matching is related to the issue of computing and refining similarity measures, only little attention has been paid to question how to extract the final alignment from a matrix of similarity values. In this paper we present a theoretical framework for describing extraction methods and argue that the quality of the final matching result is highly affected by the extraction method. Therefore, we discuss several extraction methods and apply them to some of the results submitted to the OAEI 2006. The results of our experimental study show that the proposed strategies differ with respect to precision and recall. In particular, theoretical considerations as well as emprirical results indicate that methods that additionally make use of information encoded in the ontologies result in better extractions compared to state of the art approaches.