The Research Of Semantic Similarity Computing Algorithm Based On HowNet

With the development of the Internet and the increase of cyber source,providing accurate information to the user from the text is in high demand and we need to improve the ability of computer text information processing.As the ba sic content of text information processing and natural language processing,semantic similarity is widely used in question answering system,example-based machine translation,multi-document summarization,information retrieval and so on.Semantic similari ty computation results directly affect the effect of text information processing,therefore it is necessary to improve the accuracy of semantic similarity computation.The word is the basic unit of semantic and grammar.The word similarity computation is the foundation of semantic similarity computation.We improve the effect of word similarity computation to better serve the application layer.This paper analyses and compares the mainstream methods of word similarity computation and researches on the method based on How Net.Based on the structure and hierarchy model of How Net,this paper proposes an improved method for computating word similarity based on sememe probability density ratio.Compared with the major state-of-the-art methods,the results indicate that the proposed method is much closer to the hand-marked sequences.The sentence is an essential structure of a complete semantic.The sentence similarity computation has a lot to do with lexical,semantic,syntax,context and so on,which is a very challenging problem.The sentence similarity computation method based on one feature has a one-sided disadvantage,this paper research es on the method with multi-feature based on How Net.Based on the abundant semantic information and the unique knowledge str ucture of How Net,integrating morphological features,semantic features and syntactic features,this paper presents an improved method with multi features based on weight of words.Compared with the original method,the experimental results show that the p roposed method improved the effects of sentence similarity computation.