Ontology matching reconciles semantic heterogeneity between ontologies to find correspondence of entities, including classes, properties, and instances. We propose using ProbLog program to tackle this problem. It uses probabilistic facts to encode initial similarities (priors) of candidate matching pairs and exploits definite clauses and annotated disjunctions to respectively express certain and probabilistic influence of matching pairs. We experimentally evaluate our method on the datasets of conference track in Ontology Alignment Evaluation Initiative. Experimental results show: (1) recall is improved compared with priors; (2) our method is better in precision, recall and F_1-measure than the most related Markov logic networks.
展开▼
