基于特征加权的事件要素识别  

Feature Weighting Based Event Argument Identification

作　　者：付剑锋[1] 刘宗田[1] 刘炜[1] 单建芳[1] 

机构地区：[1]上海大学计算机工程与科学学院,上海200027

出　　处：《计算机科学》2010年第3期239-241,共3页Computer Science

基　　金：国家自然科学基金(60575035);上海市重点学科建设项目(J50103)资助

摘　　要：事件抽取是自动内容抽取(Automatic Content Extraction,ACE)会议评测的任务之一,事件要素识别是事件抽取的一个子任务。分析了事件抽取和事件要素识别的研究现状,提出了一种基于特征加权的事件要素识别算法(Feature Weighting Based Event Argument Identification,FWEAI)。该算法首先对分类算法中的ReliefF特征选择算法进行改进,将其应用于聚类算法中。改进的ReliefF算法(FWA)根据各个特征对聚类的不同贡献分配不同的权值,然后采用KMeans算法对事件要素进行聚类。实验结果表明,FWEAI算法可以提高事件要素识别的准确率。Vent extraction is a task of Automatic Content Extraction （ACE） program. Event Argument Identification is sub-task of Event extraction. The state-of-the-art of event extraction and event argument identification was given. An algorithm named Feature Weighting Based Event Argument Identification （FWEAI） was proposed, which improved ReliefF, a feature selection algorithm in classification algorithm, and employed it in clustering algorithm. The improved ReliefF （FWA）can assign different weights to different features according to their contributions to lustering,then Event arguments were clustered by using KMcans clustering algorithm. Experimental results demonstrate that FWEAI algorithm improves the precision on Event Argument Identification.

关 键 词：特征加权 RELIEFF算法 事件要素识别 事件抽取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]