An Improved Algorithm of Word Semantic Similarity Based on HowNet

Word semantic similarity calculation has been widely used in information retrieval, text classification, data similarity detection and other fields. However, the commonly used HowNet based word semantic similarity calculation method does not consider the density information of the minimum common parent node and some of the first basic sememes do not participate in the calculation when there is an inclusion or intersection relationship in the semantics, resulting in the calculation results not conforming to the reality and inaccurate, thus limiting its application. To solve this problem, we improve the edge weight function and the word similarity calculation function between nodes. The density of minimum common parent node is considered into the edge weight function. In order to deal with the inclusion and intersection problems, we consider the weight of the number of sememes participating in the calculation and the proportion of the number of first sememes not participating in the calculation to the total number of first basic sememes. The experimental results show that the improved semantic similarity algorithm can effectively improve the accuracy of the sememe similarity calculation, and can solve the problem that the result of the word similarity calculation is too large when the words are included or intersected.