面向英文句子的框架语义扩展及相似度计算  

Frame Semantics Extension and Similarity Computation for English Sentences

作　　者：王铁鑫 刘文静 杨志斌[1,2] 曹静雯 WANG Tie-xin;LIU Wen-jing;YANG Zhi-bin;CAO Jing-wen(Nanjing University of Aeronautics and Astronautics,College of Computer Science and Technology,Nanjing 210016,China;Key Laboratory of Safety-Critical Software-Ministry of Industry and Information Technology,Nanjing 210016,China)

机构地区：[1]南京航空航天大学计算机科学与技术学院,南京210016 [2]高安全系统的软件开发与验证技术-工信部重点实验室,南京210016

出　　处：《小型微型计算机系统》2021年第10期2059-2064,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金面上项目(61872182)资助。

摘　　要：针对英文句子相似度计算问题,提出基于框架语义(frame semantics)的分析计算方法.对比传统基于关键词及句子结构等计算方法,提出的方法考虑整个句子的语义信息,即关键词的上下文信息,以提高相似度计算结果的准确性.该方法建立在框架语义库FrameNet之上,并结合使用语义框架抽取工具Semafor及命名实体抽取工具NER,以减少人工参与,提高方法的执行效率.通过扩展FrameNet中的语义框架类别,应对部分关键词无法激起框架的问题;通过量化框架间语义关系并定义计算规则,实现句子间相似度的计算与判断.在标准测试集上的实验结果表明,该方法的计算结果在准确率、召回率等评价指标上,优于传统方法.Towards sentence similarity computing,we proposed an analysis and computation method based on frame semantics.Compared with the traditional computation methods based on keywords and sentence structures,the proposed method considers the whole sentence semantic(i.e.,the context of keywords) so as to improve the accuracy of similarity computation results.The method is built on FrameNet and combines with two mature tools Semafor(to extract semantic frames) and NER(to extract named entities).These tools help to reduce human participation and improve the efficiency of the method.By extending newframes to FrameNet,the problem that some keywords can’t arouse frames was solved.By quantifying the semantic relations between frames and defining the computation rules,we realized the computation and judgment of sentence similarity.Comparing with traditional computation methods,the experiment on a standard testing set shows that the proposed method achieved better similarity results on criteria,such as precision and recall.

关 键 词：句子相似度 框架语义 FRAMENET 实体抽取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]