A Semi-supervised Learning Approach for Ontology Matching

Ontology matching is the task of finding correspondences between semantically related entities in different ontologies, which is a key solution to the semantic heterogeneity problem. Recently, several supervised learning approaches for ontology matching have been proposed, which outperform traditional unsupervised approaches. The existing learning based approaches treat the similarity values of matchers as normal numerical features, and need a lot of training examples. In this paper, we propose a semi-supervised learning approach for ontology matching. Our approach needs a small set of training examples, and exploit the dominant relation of similarity metrics to enrich the training examples. A label propagation algorithm is used to determine the matching results. Experimental results show that our approach can achieve good matching results with a few training examples.