Extraction of Conceptual Relation Based on HowNet and Concept Graph

Abstract

In view of the low efficiency of depending on one extracting method, this paper proposes a blending extracting method based on the combination of statistics, regulations and managing nature language. By employing template construction to extract conceptual relations, this method adopts transfer learning to obtain concept pairs and by using the advantages of concept graph in knowledge representation, matches templates through conjoining the Hownet in order to gain template set and extract conceptual relations. The experimental results show that this method can raise the accuracy rate in relation extraction.