Semi-supervised Learning Approach for Ontology Mapping Problem

The evolution of the Semantic Web depends on the growing number of ontologies it comprises. However, all ontologies have differences in structure and content because there is no unified standard for their design. To ensure inter-operability and fluent information exchange, the correspondences between entities of different ontologies must be found and mapped. A lot of methods have already been proposed for matching heterogeneous ontologies, but they still have many shortcomings and require improvements. This paper suggests a novel semi-supervised machine learning method, which solves ontology mapping task as a classification problem with training set, comprised only of labeled positive examples. Negative examples are generated artificially using an entropy measure in order to build a more accurate Naive Bayesian classifier.