面向半监督中文事件抽取的事件推理方法  

Event inference for semi-supervised Chinese event extraction

作　　者：徐霞[1] 李培峰[1] 郑新[1] 朱巧明[1] 

机构地区：[1]苏州大学计算机科学与技术学院,江苏苏州215006

出　　处：《山东大学学报（理学版）》2014年第12期12-17,共6页Journal of Shandong University(Natural Science)

基　　金：国家自然科学基金资助项目(61272260);江苏省自然基金资助项目(BK2011282);江苏省高校自然科学重大基础研究项目(11KIJ520003)

摘　　要：半监督中文事件抽取系统的性能依赖于种子模板,但自动获取的种子模板的表达方式与覆盖范围有限,导致某些语言现象下的事件实例很难被识别。为解决这一难题,基于篇章内的事件一致性理论提出基于同指事件与相关事件的推理方法,根据已抽取的事件实例来推理可能有同指关系与关联性的其它事件,从而进一步提高半监督中文事件抽取系统的性能。在ACE 2005中文语料上的测试表明,该方法可有效地提高半监督中文信息事件抽取系统的性能。The performance semi-supervised Chinese event extraction depends on the quality of seed patterns. However, the expression styles and coverage of those seed patterns, which are extracted automatically, is limited and that leads to lots of event mentions cannot be identified for their contexts. To solve this issue, an event inference mechanism based on co-reference events and relevant events was proposed, which follows the theory of event consistency in a topic. This mechanism can infer those event mentions which have the co-reference or relevance relations with the extracted event men- tions in the same document, and then the performance of semi-supervised Chinese event extraction was further improved. The experimental results on the ACE 2005 Chinese corpus show that our approach outperforms the baseline significantly.

关 键 词：半监督 事件抽取 事件推理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]