基于HowNet句子相似度的计算

摘要

Chinese sentence similarity computation is a fundamental and important work in the natural language processing. It directly de-termines the status of research and development for certain related fields. Based on the word similarity computing,for the shortcoming of current sentence similarity computing methods,present a method to compute sentence similarity Based on HowNet. Based on the lexical semantic similarity calculation of HowNet,antonyms,negative single sememe,symbol sememe,and definition information are demonstra-ted to calculate word similarity. In this method,word disambiguation is completed before the calculation of sentence similarity. The situa-tion of the similarity between the relation sememe of a word definition and a certain sememe of the given word are considered,and the re-lation sememe weight is added. Under the same test conditions,the experimental results show that the proposed method can improve the computational accuracy of sentence similarity and it is much closer to the people' s comprehension to the meanings of the sentences.%汉语句子的相似度计算在自然语言处理领域中是一项基础而又重要的工作,它直接决定着相关领域的研究发展状况.在词语相似度计算的基础上,针对目前句子相似度计算方法的不足,文中提出一种基于HowNet的计算句子相似度的方法.在《知网》的词汇语义相似度计算基础上,加入了词语定义义原间的反义、对义关系、单义原的否定和符号义原、定义信息来计算词语的相似度.计算句子相似度前加入词语的消歧,在计算句子相似度时考虑了词语定义的关系义原与待比较的词定义的某个义原相等的情况,并加大了关系义原的权重.实验结果表明,在同等的测试条件下,所提出的句子相似度计算方法可以提高句子相似度的计算精度,更符合人的直观感觉.