改进的基于《知网》的词汇语义相似度计算  

Improved vocabulary semantic similarity calculation based on HowNet

作　　者：朱征宇[1,2] 孙俊华[1,2] 

机构地区：[1]重庆大学计算机学院,重庆400044 [2]软件工程重庆市重点实验室,重庆400044

出　　处：《计算机应用》2013年第8期2276-2279,2288,共5页journal of Computer Applications

基　　金：国家科技支撑计划项目(2011BAH25B04)

摘　　要：针对当前基于《知网》的词汇语义相似度计算方法没有充分考虑知识库描述语言对概念描述的线性特征的情况,提出一种改进的词汇语义相似度计算方法。首先,充分考虑概念描述式中各义原之间的线性关系,提出一种位置相关的权重分配策略;然后,将所提出的策略结合二部图最大权匹配进行概念相似度计算。实验结果表明,采用改进方法得到的聚类结果 F值较对比方法平均提高了5%,从而验证了改进方法的合理性和有效性。The present HowNet-based vocabulary semantic similarity calculation method fails to give due attention to the linear feature of conceptual description in knowledge database mark-up language.To resolve this shortcoming,an improved vocabulary semantic similarity calculation method was proposed.Firstly,fully considering the linear relationship between the sememes in the conceptual description formula,a position-related weight distribution strategy was proposed.Then concept similarity was calculated by combining the strategy above with bigraph maximum weight matching.The experimental results show that,compared with the contrast method,the F-measure of text clustering using improved method increases by 5% on average,thus verifying the rationality and validity of the improved method.

关 键 词：知网 义原 概念 权重 语义相似度 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]