基于知识感知提示与对比调优的事件元素抽取方法

Event Argument Extraction Method Based on Knowledge-Aware Prompting and Contrast Tuning

提示学习在事件元素抽取领域的应用越来越广泛。由于缺乏对实体信息的考虑,现有的事件元素抽取提示学习方法并没有达到满意的效果。现有的大部分基于提示学习的模型对于事件元素的表示也不够充分,分类效果欠佳。提出一种基于知识感知提示与对比调优的事件元素抽取方法,该方法基于预训练模型,构建知识感知模板,将实体知识注入预训练模型中,通过中心对比学习充分区分元素表示,在预测阶段使用CRF-Viterbi解码算法提升解码效果。实验结果表明,在ACE2005数据集上,该方法相较于基线模型取得了更优越的效果。

Recently,prompt learning has been applied more and more widely in the field of event argument extraction.However,due to the lack of consideration of entity information,the existing event argument extraction prompt learning methods haven't got satisfactory effects.In addition,most of the existing models based on prompt learning are not sufficient for the representation of arguments,so that the classification effect isn't good.Therefore,an event argument extraction method based on knowledge-aware prompting and contrast tuning is proposed.The method is based on a pre-training model,the knowledge aware template is built,the entity knowledge is injected into the pre-training model,and the argument representation is fully distinguished by central contrast learning.Finally,CRF-Viterbi decoding algorithm is used to improve the decoding effects in the prediction stage.The experimental results show that the proposed method has better results than the baseline model on ACE2005 dataset.