Heinrich-Heine-Universität Düsseldorf

Ontology Matching is a task needed in various applications, for example for comparison or merging purposes. In literature, many algorithms solving the matching problem can be found, but most of them do not consider instances at all. Mappings are determined by calculating the string-similarity of labels, by recognizing linguistic word relations (synonyms, subsumptions etc.) or by analyzing the (graph) structure. Due to the facts that instances are often modeled within the ontology and that the set of instances describes the meaning of the concepts better than their meta information, instances should definitely be incorporated into the matching process. In this paper several novel instance-based matching algorithms are presented which enhance the quality of matching results obtained with common concept-based methods. Different kinds of formalisms are use to classify concepts on account of their instances and finally to compare the concepts directly.