摘要： 提出了一种“知网”支持下的跨语句汉语事件信息抽取(IE)方法。该方法的基本思想是:从经过标注的描述某类事件的一个事件实例集合中,学习出事件的各个角色的特征义原集合和角色所属的语义类集合。然后,对于一个新的该类事件的实例句群,使用每个角色的特征义原集合来定位角色所在的单句,并通过该角色所属的语义类集合从定位后的单句中进行该角色的识别和抽取。实验表明,该方法能够完成从不同领域中对不同类别汉语事件的IE任务,并得到较高的抽全率和抽准率。

关键词: 汉语事件信息抽取, 知网, 语义类集合

Abstract: This paper provides a method to do Chinese event IE from a multiple sentences event narration with the support of hownet. The main idea behind this method is like this: firstly, learning the featuring sememe set for each event role and the semantic class set that each event role belongs to from a labeled event case set in which every event case belongs to a same event class. Then, to a new event case of this event class, using each event roles featuring sememe set to locate the single sentence that may contains the role and using its semantic class set to identify and extract this role from the located single sentence. The experiment done indicates that this method can fulfill various event IE tasks from different domains and get a high recall and precision.

中图分类号: 

• TP391.1