面向中文词汇语义相关性计算的ConceptNet与HowNet对比分析  

Systematic Analysis of ConceptNet and HowNet Towards Semantic Relatedness Computing of Chinese

作　　者：曹静雯 王铁鑫 杨志斌[1,2] 李文心 CAO Jing-wen;WANG Tie-xin;YANG Zhi-bin;LI Wen-xin(Nanjing University of Aeronautics and Astronautics,College of Computer Science and Technology,Nanjing 211106,China;Key Laboratory of Safety-Critical Software-Ministry of Industry and Information Technology,Nanjing 211106,China)

机构地区：[1]南京航空航天大学计算机科学与技术学院,南京211106 [2]高安全系统的软件开发与验证技术-工信部重点实验室,南京211106

出　　处：《小型微型计算机系统》2020年第8期1641-1648,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61872182)资助。

摘　　要：作为自然语言处理的关键技术之一,词汇之间的语义相关性计算依赖于语义资源.语义资源可分为知识库和语料库;How Net和ConceptNet是包含中文词汇的两种语义知识库,可为中文词汇语义相关性计算提供依据.本文在分析以上两种语义知识库结构及应用的基础上,提出:面向How Net的基于信息含量的语义相关性计算方法,以及面向ConceptNet的基于路径距离的语义相关性计算方法.通过实验案例分析,得出如下结论:How Net明确定义词汇的概念,适用于检测词对间的语义相似度;ConceptNet定义词对间的多重语义关系,在检测词对语义相似度的基础上,亦能检测词对间的语义相关度.此外,本文探索性地提出结合使用以上两种语义知识库的词对间语义关系计算算法,以期扩大检索词汇的范围并提高效率.Semantic relatedness computing,as a critical technique of natural language processing,relies on semantic resources.Semantic resources can be divided as know ledge bases and corpora.Concerning Chinese lexical semantic relatedness computing,How Net and ConceptNet are employed as the semantic know ledge bases.By analyzing the structures and applications of the two semantic know ledge bases,this paper proposes two semantic relatedness computing methods:an information-content based towards How Net and a path-based towards ConceptNet,respectively.Through a systematic analysis,we drawa conclusion that How Net,which clearly defines the concepts of words,is suitable for detecting the semantic similarity between word pairs and ConceptNet,which defines multiple semantic relations,can be used to detect both the semantic similarity and relatedness between word pairs.In order to expand the scope of search words and improve efficiency,this paper explores an algorithm building on the above two semantic know ledge bases in an exploratory way.

关 键 词：自然语言处理 语义知识库 ConceptNet How Net 语义检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]