面向社会舆情的中文事件抽取及其可信度计算的研究

项目名称： 面向社会舆情的中文事件抽取及其可信度计算的研究

项目编号： No.61472265

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 计算机科学学科

项目作者： 李培峰

作者单位： 苏州大学

项目金额： 82万元

中文摘要： 互联网为社会公众提供了前所未有的舆情表达新手段，如何从海量舆情信息中获取有价值内容成为目前急待解决的一大挑战。事件作为表述社会舆情语义的基本要素，从海量文本中抽取事件并计算其可信度是快速、及时、高效地分析社会舆情的基础。本项目将在话题结构理论和语义一致性理论的指导下，根据中文所特有的语言特点，研究面向社会舆情的中文事件抽取及其可信度计算方法，重点解决其信息缺失问题、全局优化问题和可信度计算问题。主要特色如下：1）基于话题结构理论，提出了新颖的跨事件、跨实体和跨角色的事件推理方法，解决中文事件缺省问题；2）基于语义一致性理论，提出了根据事件间内在关系进行事件抽取的联合学习模型，解决全局优化问题；3）基于篇章结构理论和事件间的关联性，提出了一个利用贝叶斯网络进行推理的事件可信度计算模型。本项目对于探索人类语言理解的认知机理，推动面向社会舆情分析的语义知识获取方法的发展，具有重要的科学意义。

中文关键词： 中文事件抽取；可信度计算；社会舆情；联合模型；事件推理

英文摘要： Currently, the Web provides an unprecedented way to express social opinions and how to extract valuable contents from such mass information about the social opinions remains a challenge. As event is the basic semantic element of the social opinion, extracting events from large scale texts and then computing their confidences is found mental to analyze the social opinion timely and effectively. Following the topic structure theory, the semantic consistency theory and the nature of Chinese language, this project will focus on Chinese event extraction and its confidence computing for social opinion analysis to solve three problems: ellipsis, global optimization and confidence computing. The contributions of this project are as follows: 1) based on the topic structure theory, it proposes a novel cross-event, cross-entity and cross-role event inference mechanism to solve the ellipsis problem; 2) based on the semantic consistency theory,it provide a joint learning model, employing the intrinic relationship among events in a topic, to slove the global optimization problem; 3) based on the discourse structure theory and the event relevance, it puts forward an Bayesian Network-based event confidence computing model. The project is helpful and meaningful to explore the cognitive mechanism of human language understanding and to promote development of the semantic knowledge acquisition for the social opinion analysis.

英文关键词： Chinese Event Extraction;Confidence Computing;Social Opinion;Joint Modeling;Event Inference