Grass-roots Class Alignment.pdf

Grass-roots Class Alignment Baoshi Yan Information Sciences Institute, University of Southern California 4676 Admiralty Way, Marina del Rey, California 90292 baoshi@ Abstract Current ontology alignment practices adopt a central- ized approach: the alignment task is carried out by a do- main expert, possibly with the help of some ontology align- ment tools. In a Peer-to-Peer environment, each end-user (peer) may explicitly or implicitly generate ontology align- ments for their own purposes during its own use of the semantic data. This kind of end-user-generated ontology alignments, which we call grass-roots ontology alignments, could be mined and generalized to help other users with their alignment tasks, thus enabling a p2p-style ontology alignment. Grass-roots ontology alignment, often gener- ated as a side effect of other data manipulations, could be peer-specific, task-specific, approximate, or even contradic- tory. Its semantics is not necessarily the equivalency be- tween classes or properties. This paper reports our work on reusing grass-roots class alignment for aligning class hierarchies. A grass-roots class alignment, though approx- imate, still reveals some facts about relationships between different classes. We formalize facts about class relation- ships that can be inferred from an alignment under differ- ent cases. We then apply forward-chaining inference to the facts knowledge base to infer more facts. The facts KB is then leveraged for ontology alignment purposes. To deal with uncertainty and inconsistency, each fact is associated with an evidence that tells how the fact is obtained. The ev- idences are used to select better-supported facts in case of inconsistency. 1 Introduction In a Peer-to-Peer knowledge management environment, each end user (peer) might have access to a different set of ontologies, and might explicitly or implicitly align on- tologies for their own purposes. This kind of end-user- generated ontology alignment, which we call grass