Combining statistical and logical inference for ontology alignment

TitleCombining statistical and logical inference for ontology alignment
Publication TypeJournal Articles
Year of Publication2007
AuthorsUdrea O, Getoor L
JournalWorkshop on Semantic Web for Collaborative Knowledge Acquisition at the International Joint Conference on Artificial Intelligence
Date Published2007///

In recent years, ontologies have proved a veryattractive medium for storing domain knowl-
edge in a variety of organizations; a few ex-
ample domains are medicine1 and genetics2.
The rapid evolution of representation stan-
dards has led to the creation of distinct on-
tology bases in overlapping domains, making
it paramount for large systems to reconcile on-
tological data from multiple sources. A large
body of work on ontology integration[Euzenat,
2004] has produced algorithms and heuristics
that have proven successful in alleviating the
high computation costs of the process. How-
ever, the effective use of ontology formalisms
– such as rules and axioms – in the integra-
tion process remains an open question. In this
paper, we present a novel ontology integration
method that takes advantage of the logical in-
ference capabilities in OWL Lite to improve the
recall of the merged ontologies. Our CPI (Clus-
tering with Partial Inference) algorithm uses
(i) graph clustering techniques to improve the
process of selecting candidate matching enti-
ties and (ii) a limited, adaptive reasoning pro-
cess to account for the effects of introducing
relationships between such entities. Our initial
experiments, performed on a set of 30 pairs of
publicly available ontologies, show that CPI of-
fers a 30% improvement in recall and a 25% im-
provement in answer quality compared to FCA-
merge [Stumme and Maedche, 2001].