基于跨语言信息投影的泰语新闻事件抽取

Tai-news event extraction based on cross-lingual information projection

事件抽取是信息抽取研究领域中的一个关键问题。针对采用传统的自我训练方法进行泰语新闻事件抽取,具有错误从上向下传播导致事件抽取系统性能不高的问题,针对这一问题,文中提出一种新的引导框架来进行泰语新闻事件抽取,即跨语言信息投影。该方法将从中文事件抽取系统中获取的信息映射为泰语,并采用结合单语与跨语言半协同训练的方法进行泰语新闻事件抽取。最后通过实验验证了本文所提方法是可行的,并且显著地提高了泰语新闻事件抽取的性能。

Event extraction is a key issue in the field of information extraction research.In view of the error spreads from top to down when the traditional self-training methods for Tai-news event extraction is used,which resulting that event extraction system performance is not high.This paper presents a new guidance framework for the Thai-news events extraction,that is,cross-language information projection.The method maps the information obtained from the Chinese event extraction system to the Thai,and it uses the method of combining monolingual and cross-lingual semi-co-training to extract Thai news events.Finally,the experiments show that the method proposed is feasible,and significantly improves the performance of the Thai-news event extraction system.