网络舆情信息中事件篇章关系检测方法的研究

项目名称： 网络舆情信息中事件篇章关系检测方法的研究

项目编号： No.61272259

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 姚建民

作者单位： 苏州大学

项目金额： 80万元

中文摘要： 事件篇章关系检测是信息抽取和舆情分析交叉领域的重要研究课题，对于以事件为主体元素的自然语言逻辑关系抽取，以及借助关联事件挖掘舆情信息的衍生线索和发展脉络，都具有很高的实用价值。目前，事件关系检测的相关研究较少，尤其借助篇章分析从语义层面深入解释和描述事件关系的研究尚属空白。本课题将重点研究刻画事件关系的语言学规律，并基于篇章分析，探索事件语义关系的机器学习和自动检测方法。主要研究内容包含如下四个方面：基于跨实体推理的事件抽取、基于动态话题模型的跨篇章关联事件识别、基于平行理论的事件篇章关系检测、事件关系层次作用域的自动构建。特别是研究借助宏观话题对事件关系的约束，识别浅层事件关系的方法；以及借助事件语义平行性识别，利用平行事件参与篇章关系形成过程的语言学共性，检测事件逻辑关系的数学建模方法。目标是实现针对舆情信息中各类事件逻辑关系的自动识别与检测，借以辅助事件衍生与发展的预测与预报。

中文关键词： 事件抽取；关联事件识别；双语语料挖掘；事件篇章关系检测；事件关系层次作用域

英文摘要： As an important research task in the cross-field of information extraction and public opinion analysis, Event Discourse Relation Detection (abbr., EDRD) has high practical value in extracting logical relations for nature languages that specially represent events, and mining occurrence cues and development veins of public opinions by using relevant events. However, there are still few researches focusing on the event relation detection, and especially it is still a blank field to use discourse analysis to thoroughly explain and describe the event relation at the semantic level. For this, the project focuses on researching the linguistic regularities of event relation, and based on the discourse analysis, exploring the methods of machine learning and automatically detecting event discourse relation. The research content can be divided into the following four parts: the first is cross-entity inference based event extraction, the second is dynamic topic modeling for cross-discourse relevant event identification, the third is parallel theory based event discourse relation detection, and the last is hierarchical scope establishing for event discourse relation. Especially the project will research the method of identifying shallow event relation under the restriction of global topics, and the mathematical modeling meth

英文关键词： Event extraction；event relation detection；bilingual corpus mining；event discourse relation detection；event relation hierarchical scope