基于概念层次网络的事件抽取研究  

Research on Event Extraction Based on the Hierarchical Network of Concepts

作　　者：韦向峰[1] 缪建明 池哲洁[1] 袁毅[1] WEI Xiangfeng;MIAO Jianming;CHI Zhejie;YUAN Yi(Laboratory of Speech Acoustics and Content Understanding,Institute of Acoustics,Chinese Academy of Sciences,Beijing,100190,China;Information Central of China North Industries Group Corp,Beijing,100089,China)

机构地区：[1]中国科学院声学研究所语言声学与内容理解实验室,北京100190 [2]中国兵器工业信息中心,北京100089

出　　处：《网络新媒体技术》2020年第4期16-21,共6页Network New Media Technology

基　　金：富媒体数字出版内容组织与知识服务重点实验室开放基金项目“面向语音与文本内容的富媒体关联研究与技术实现”;中国科学院信息化专项科学大数据工程(一期)多民族语言资源特色数据库课题(编号:XXH13505-03-203)。

摘　　要：为使计算机能从文本中自动抽取出事件及事件要素,利用概念层次网络理论,对文本进行句类分析得到语义块后,以事件触发概念信息为线索,加入语义块权重,用频数指标提出事件类型判断方法。进一步利用句类知识分析语义块和事件要素的对应关系,并建立对应规则,抽取事件要素。在中文突发事件语料库上进行了实验,事件类型识别和事件要素抽取的F1测度宏平均分别为0.871和0.768,较对比方法分别提升了4.8和6.4个百分点,总体处理效果较好。In order to extract automatically events in text and elements in events by using computer,this paper uses the theory of HNC(Hierarchical Network of Concepts)to obtain the semantic chunks in text after analyzing the sentence categories of sentences in text.An event type recognition method,which regards event-triggered concepts as indicators and considers the weights of semantic chunks,is proposed by using frequency index after conducting sentence category analysis.Event elements extraction is carried out by constructing the corresponding relationship and rules between semantic chunks and event elements.These methods are testified in Chinese Emergency Corpus and the macro F1 of event type recognition and event element extraction achieve 0.871 and 0.768 respectively,which are 4.8 percentage points and 6.4 percentage points higher than the baseline.

关 键 词：事件抽取 概念层次网络 概念基元 句类 语义块 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]