基于要素图的舆情事件线索化方法研究  

Threading Events of Public Opinion Based on Argument Graph

作　　者：何绯娟[1] 杨宽[2] 缪相林[1] 高建忠[3] 

机构地区：[1]西安交通大学城市学院,西安710018 [2]西安交通大学电子与信息工程学院,西安710049 [3]西安交通大学图书馆,西安710049

出　　处：《情报杂志》2017年第7期101-104,共4页Journal of Intelligence

基　　金：教育部人文社会科学研究青年基金项目"新闻事件的线索化方法研究"(编号:15YJCZH057);陕西省社会科学基金项目"开放环境下知识的刻面组织方法研究"(编号:2016N005);西安交通大学城市学院校级科研项目"大规模在线学习中的隐性知识挖掘方法研究"(编号:2016KZ01)

摘　　要：[目的/意义]将事件按照内在关联组织成有序结构的过程称为事件线索化,事件线索化能够展现事件间的因果联系与演化过程,有助于理解舆情事件的起因与背景,预测事件的后续发展,为舆情事件的处理提供依据。现有事件线索化方法主要依据事件要素的共现特性,然而部分存在关联的事件之间共现特性并不明显,导致现有方法存在局限性。[方法/过程]利用百度百科中的条目及其中的关键词来扩充事件要素,建立了由关键词与超链接构成的要素图模型,并提出了基于该模型的舆情事件线索化方法。所提方法通过要素图差异性计算与密度聚类两个阶段将事件集划分为由相互关联事件组成的簇,从而实现事件线索化。[结果/结论]在舆情事件数据集上的实验表明,所提方法在共现特性不明显情况下能有效地实现事件线索化。[ Purpose/Significance ] Event threading is defined as the process of organizing the events of public opinion into a structured form according to the associations among them. Event threading can show the causality and evolution of events, which is helpful to under- stand the cause and background of the events, predict the development of events and provide the basis for the handling of public opinions. The existing event threading methods are mainly based on the co-occttn^nce of the event arguments; however, the co-occurrence phe- nomenon in some associated events is not significant, which limits the existing methods. [ Method/Process] Therefore, this paper utilizes the articles and keywords of Baidupedia to expand the event arguments, and establishes an argument graph model consisted of keywords and hyperlinks. Based on this model, an event threading method is proposed. The proposed method applies a two-stage strategy, i. eargu- merit graph similarity calculation and density clustering, on an event set to divide it into dusters composed of associated events. [ Result/ Conclusion ] Experiments on the event data set demonstrate that the proposed method can effectively thread the events with insignificant co -occurrence.

关 键 词：舆情 事件线索化 事件关联 要素图 百度百科 

分 类 号：G353.12[文化科学—情报学]