Learning Ontology Automatically Using Topic Model

Abstract

Ontology has been extensively applied in various fields, such as artificial intelligence, information extraction and retrieval et al. In this paper we describe a new approach for automatic learning terminological ontology. The method takes the topics generated by generative topic model as concepts and builds subsumption relationships between such concepts to learn ontology without the existence of seed ontology. The method presents CosTMI measure to compute semantic similarity between topics and to organize these topics into hierarchy structure and form new ontology. We evaluate our method using real world text dataset GENIA corpus which is a collection of biomedical literature. And the experiment results demonstrate the validity and efficiency of proposed method.