文献类型：会议论文

英文题名：Where Scene Happened in Text on Relevant Measures

作者：Li, Hanjing[1];Dong, Ruizhi[2]

第一作者：李晗静

通讯作者：Li, HJ[1]

机构：[1]Beijing Union Univ, Specail Eduact Sch, Beijing 100075, Peoples R China;[2]Harbin Inst Technol, MOE MS Key Lab Nat Language Proc & Speech, Harbin 150001, Peoples R China

第一机构：北京联合大学特殊教育学院

通讯机构：[1]corresponding author), Beijing Union Univ, Specail Eduact Sch, Beijing 100075, Peoples R China.|[1141759]北京联合大学特殊教育学院;[11417]北京联合大学;

会议论文集：6th International Conference on Intelligent Systems and Knowledge Engineering (ISKE 2011)

会议日期：DEC 15-17, 2011

会议地点：Shanghai, PEOPLES R CHINA

语种：英文

外文关键词：Dunning's likelihood ratio; text-to-scene; scene identification; scene category

摘要：This paper proposes a method to identify scenes in texts with relevant measures, namely to infer the locations where actions take place without any explicit description in texts. Firstly, 48 scene categories are classified by the sememe hypernym and concept similarity on HowNet. Secondly, the hierarchical scene information topics generated on action tuples and scenes from corpus are extracted by calculating the relevancy for each tuple, and then the knowledge base is generated. Finally, the constructed knowledge base is used to identify scenes in novel texts, and the experimental results on I-best and voting shows the effective of the method.