基于正负加权的中文事件识别研究

CHINESE EVENT RECOGNITION BASED ON POSITIVE AND NEGATIVE WEIGHTING

事件识别是事件抽取的子内容,其主要任务是找出文本中的事件触发词.针对现有的事件识别方法对单一特征的利用还不够充分的问题,通过构建包含正负特征的触发词表,提出一种基于正负加权的事件识别方法.定义一种关联词特征,该特征对事件识别有较好的效果;根据单一特征所属的词是否为触发词将特征分为正特征或负特征,并将正负特征结合起来进行触发词识别,提升单一特征在事件识别时的作用.在此基础上,将正负关联词特征、正负词性特征以及正负依存关系特征结合起来进行触发词识别,进一步提升事件识别效果.实验结果表明,基于正负加权的事件识别方法得到了比较理想的效果.

Event recognition is a sub-content of event extraction.Its main task is to find event triggers in text.Aiming at the insufficient utilization of single feature in current event recognition methods,we propose an event recognition method based on positive and negative weighting by constructing triggers table with positive and negative features.A related word feature was defined,which had a good effect on event recognition.According to whether the word of single feature was a trigger word,the feature was divided into positive or negative features,and the positive and negative features were combined to recognize the trigger word,so as to improve the function of a single feature in event recognition.On this basis,we combined the positive and negative related word features,positive and negative part of speech and positive and negative dependency relationship to recognize trigger words,and further improved the effect of event recognition.The experimental results show that the event recognition method based on the positive and negative weighting has a better effect.