Word Semantic Similarity Calculation Based on Domain Knowledge and HowNet

Abstract

Word semantic similarity is the foundation of semantic processing, and is a key issue in many applications. This paper argues that word semantic similarity should associate with domain knowledge, which traditional methods did not take into account. In order to adopt domain knowledge into semantic similarity measurement, this paper proposed a sensitive words sets approach. For this purpose, we also propose a new approach for sememe similarity calculation. This method distinguishes three different positional relationships between two sememes, and the results have shown that our method overperformed than other methods based on a Chinese knowledge base ‘HowNet’.

DOI : http://dx.doi.org/10.11591/telkomnika.v12i2.4416