摘要
针对现有流水线式事件抽取方法依靠大量训练数据、在低资源情况下难以快速迁移运用等问题,利用提示调优技术,提出适用于低资源场景下的流水线式事件抽取方法(low-resource event extraction method using the multi-information fusion with prompt tuning,IFPT)。该方法通过构造语义映射和提示模板充分利用事件类型描述、实体类型等多种信息,能够高效使用有限训练数据,流水线式地完成事件检测和论元抽取。实验结果表明,在低资源情况下,IFPT方法论元抽取性能超过了所有基准模型,采取流水线方式能够达到与SOTA模型相媲美的性能。
Existing pipeline-based methods rely on a large amount of training data while not being able to use external information effectively,making it difficult to transfer and apply quickly in low-resource situations.This paper used prompt-tuning to fuse multiple information and proposed a low-resource event extraction method using the multi-information fusion with prompt tuning(IFPT).By constructing semantic verbalizers and prompt templates,this method was able to complete event detection and argument extraction with pipeline-based paradigm,making full use of various information such as event type description and entity type.Experimental results show that IFPT outperforms all baseline models in argument extraction and achieves comparable performance to the SOTA model in event extraction with pipeline-based paradigm.
作者
苏杭
胡亚豪
潘志松
Su Hang;Hu Yahao;Pan Zhisong(College of Command&Control Engineering,Army Engineering University of PLA,Nanjing 210007,China)
出处
《计算机应用研究》
CSCD
北大核心
2024年第2期381-387,400,共8页
Application Research of Computers
基金
国家自然科学基金资助项目。
关键词
事件抽取
低资源
提示调优
预训练语言模型
论元抽取
event extraction(EE)
low-resource
prompt tuning
pretrained language model
argument extraction
