Word Similarity Computing Based on HowNet and Synonymy Thesaurus.

Based on HowNet, some definitions of entity similarity are unreasonable. The synonymy and similar relationships of each entity are analyzed in great detail in Extended Synonym. But Extended Synonym includes only entity words, not other types of words. In view of the deficiencies of the two dictionaries, a word similarity method based on HowNet and Extended Synonym is proposed. Firstly, a synonym expansion figure is established through the existing tree figure in Extended Synonym. Then based on the two types of figures and six types of words, the entity similarity methods in five different cases are established in detail. The important parameters in the five methods are set based on the existing semantic parameter and large-scale experiments. For the other case that one of two words is at least not an entity, the traditional HowNet-based method is used to compute the similarity between two words. The word similarity method proposed in this paper improves the performance of short text similarity.