Naive Bayesian Classifier Based Semi-supervised Learning for Matching Ontologies
Xingsi Xue, Dequ Chen, Wenyu Liu · 2021 17th International Conference on Computational Intelligence and Security (CIS) · 2021
The evolution of Semantic Web (SW) depends on the increasing number of ontologies it contains. However, the existing ontologies are diverse in structure and content, because their design standards are different. In order to ensure their knowledge sharing, the correspondences between different ontology entities should be determined, which is called ontology matching. Currently, various ontology matching techniques have been proposed, which makes use of different strategies or methods to improve the quality of alignment. Being enlightened by the success of Semi-supervised Learning (SL) in ontology matching domain, this paper further proposes aN aive Bayesian Classifier (NBC) based SL, and use it to obtain high-quality alignment. In particular, our approach first models the ontology matching issue as a binary classification problem; then the Positive and Unlabeled (PU) training set and the negative examples that are determined by it is constructed, which is used to train the NBC; and finally, with the learned knowledge, the complete ontology alignment is obtained. The testing cases used in the experiment are provided by Ontology Alignment Evolution Initiative (OAEI). Comparing with the existing ontology matching techniques, the results show that the effectiveness of our method.