词语语义相似度在突发事件案例检索中的应用  

Application of Word Semantic Similarity in Emergency Case Retrieval

作　　者：邱俊安 邱奇志[1] 周三三 陈先桥[1] 贺宜[2] QIU Jun′an;QIU Qizhi;ZHOU Sansan;CHEN Xianqiao;HE Yi(School of Computer Science and Technology,WUT,Wuhan 430063,China;不详)

机构地区：[1]武汉理工大学计算机科学与技术学院,湖北武汉430063 [2]武汉理工大学智能交通系统研究中心,湖北武汉430063

出　　处：《武汉理工大学学报（信息与管理工程版）》2020年第3期272-278,共7页Journal of Wuhan University of Technology：Information & Management Engineering

基　　金：国家自然科学基金项目(51605350);安全预警与应急联动技术湖北省协同创新中心开放课题资金项目(JD20150507)。

摘　　要：为解决以往突发事件案例检索中忽视文本属性重要性的问题,提高案例检索的准确度,将词语语义相似度计算应用到文本属性中。因案例文本特别是结构化突发事件案例的文本属性多为短文本,常规的文本分析技术难以获取足够的语义信息,故着重从3个方面研究了短文本的语义:上下文语义、单个汉字的语义和概念的层次关系,提出了基于WNCH的词语语义相似度计算方法,该方法在传统词嵌入模型中增加了Ngram和汉字的语义信息,并与HowNet相融合。在此基础上,给出了基于属性相似度的突发事件案例检索流程,将WNCH方法应用到文本属性相似度的计算。实验结果表明,文本属性的引入使得案例检索匹配更多的关键信息,从而使案例检索更加准确。In order to solve the problem of neglecting the importance of textual attributes in previous emergency case retrieval,improve the accuracy of case retrieval,word semantic similarity calculation of textual attributes was applied.Since the textual attributes of most emergency cases are short texts,traditional text analysis techniques are difficult to obtain enough semantic information.Firstly,the semantics of short texts are studied from three views:context semantics,semantics of individual Chinese characters and hierarchical relationships between concepts.The method of WNCH was proposed to calculate the similarity of Chinese words,which not only added the semantic information of Ngram and Chinese character to the traditional Word Embedding model,but also integrated with HowNet.On this basis,the emergency case retrieval procedure based on attribute similarity was given,and the WNCH was employed to the calculation of the textual attribute similarity.The experimental results show that the introduction of textual attributes reveals semantic information effectively and enhances the accuracy of case retrieval.

关 键 词：案例检索 WNCH 词语相似度 词嵌入 突发事件 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]