An improved algorithm of word similarity computation based on HowNet

Abstract

In this paper, a defect-analysis on computation of concept similarity, sememe similarity and set similarity in previous methods was given first. Next, a dynamic concept decomposition is used to regulate the amount of semantic information in the various parts of concept, and optimized the similarity between concepts; Then, the paper solved the imbalance problem of element match in previous set similarity computation by proposing a new algorithm of set similarity computation from the global perspective of semantic similarity; Finally, by taking into account the depth of the least Common Node reasonably in sememe similarity computation, this paper improved distinction degree of sememe similarity computation. The results in comparative experiment showed that the algorithm proposed by this paper is more close to the manual evaluation.