Event extraction is an essential task which aims to extract structural event information from unstructured text.Event extraction consists of two sub-tasks,event type detection and event argument extraction.Recent prior work intro-duced pre-trained language model to get semantic representation of the text.Some works formulate EE into a sequence labeling task(BIO),which is prone to label ambiguity.While others integrate feature knowledge into event text representa-tion to avoid ambiguity.However,the existing methods of integration ignore the dependencies between each sub-task of event extraction.This paper proposes a joint learning method for the event extraction task.The event type feature is inte-grated into the text representation as new knowledge through gate fusion mechanism.Experimental results show that this method performs better than the baselines on event argument extraction.
关键词
文本事件抽取/预训练语言模型/门融合
Key words
text event extraction/pre-trained language model/gate fusion
维普
万方数据