面向图注意力网络的突发热点事件联合抽取  

Joint Extraction of Sudden Hot Events Based on Graph Attention Network

作　　者：徐子路 朱睿莎 余敦辉 邢赛楠 XU Zi-lu;ZHU Rui-sha;YU Dun-hui;XING Sai-nan(College of Computer and Information Engineering,Hubei University,Wuhan Hubei 430062,China;Education Informationization Engineering and Technology Center,Hubei 430062,China)

机构地区：[1]湖北大学计算机与信息工程学院,武汉430062 [2]湖北省教育信息化工程技术中心,武汉430062

出　　处：《小型微型计算机系统》2023年第5期902-909,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61977021)资助;湖北省技术创新专项(重大项目)(2020AEA008)资助。

摘　　要：当前,突发热点事件的传播日益迅猛与广泛.如何通过事件抽取准确快速地抽取出事件触发词及其事件元素,有助于决策者分析舆情态势、引导社会舆论.针对现有事件抽取方法多是从单个句子中抽取事件元素,而突发热点事件的事件元素往往分布在多个句子当中的问题,提出了一种基于图注意力网络的突发热点事件联合抽取方法,该方法分为三个阶段:基于TextRank的事件句抽取、基于图注意力网络的篇章级事件联合抽取、突发热点事件补全.在抽取出新闻主旨事件以后对整篇新闻做事件抽取,利用候选事件与新闻主旨事件的事件向量相似度以及事件论元相似度对该新闻主旨事件进行补全.实验结果表明,该方法在DUEE1.0数据集上进行触发词抽取和论元角色抽取任务时的F1指标分别达到83.2%、59.1%;在中文突发事件语料库上进行触发词抽取和论元角色抽取任务时的F1指标分别达到82.7%、58.7%,验证了模型的合理性和有效性.Recently,hot events spread more rapidly and extensively.The method of extracting event triggers and the elements accurately and quickly by event extraction is conducive to decision makers,for it helps them to analyze the trends of public opinions and to guide public opinions.Based on the problem that a large amount of existing event extraction methods extract event elements from a single sentence,while the event elements are often distributed in pairs of sentences,this paper raises a joint extraction method of sudden hot events,which is on the foundation of graph attention network.The method is divided into three stages:event sentence extraction,chapter-level event joint extraction,and event completion.Once the main event of the news is extracted,the whole news will be extracted,then at last the main event of the news is complemented by the event vector similarity and the event similarity between the candidate events and the main events.The experimental results indicate that the method achieves 83.2%and 59.1%of the F1-measure concerning trigger word extraction and argument role extraction tasks on the Chinese emergency corpus.The F1-measure reaches 82.7%and 58.7%respectively when extracting the tasks mentioned above,which fully verifies the rationality and effectiveness of the model.

关 键 词：事件抽取联合方法 事件触发词抽取 论元角色抽取 图注意力网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]