基于依存分析的事件识别  

Dependency Parsing Based Event Recognition

作　　者：付剑锋[1] 刘宗田[1] 付雪峰[2] 周文[1] 仲兆满[1] 

机构地区：[1]上海大学计算机工程与科学学院,上海200027 [2]南昌工程学院计算机科学与技术系,南昌330099

出　　处：《计算机科学》2009年第11期217-219,共3页Computer Science

基　　金：国家自然科学基金(60575035);上海高校选拔培养优秀青年教师科研专项基金(shu-07027);上海市重点学科建设项目(J50103)资助

摘　　要：事件抽取是信息抽取的重要组成部分,事件识别是事件抽取的基础,事件识别的效果直接影响了事件抽取的结果。基于机器学习的方法识别事件需要从词汇中发掘更多的特征。针对当前事件识别方法中存在的不足,提出了一种基于依存分析的事件识别方法。用依存分析发掘触发词与其它词之间的句法关系,以此为特征在SVM分类器上对事件进行分类,最终实现事件识别。实验表明,基于依存分析的事件识别优于传统的事件识别方法,而融合多特征的事件识别F值可提高到69.3%。Event Extraction is an important part of information extraction. As the basis of Event Extraction,Event Recognition directly affects the results of Event Extraction. Machine learning based Event Recognition needs to find more features in words. For the deficiency of present Event Recognition method, this paper presented a novel method of Depen-dency Parsing based Event Recognition （DPER）. Dependency parsing was used to find the syntactic relation among triggers and other words. As one of features, this relation was used to event classification on SVM and then to event recognition. The experiments show DPER has better performance than traditional method, and Event Recognition integrating multi-features improves F-measure to 69. 3%.

关 键 词：事件识别 依存分析 支持向量机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]