事件因果关系抽取研究  

Research of Event Causality Extraction

作　　者：陈洁 张琨[1] 朱浩华 陈智源 方自正 CHEN Jie;ZHANG Kun;ZHU Haohua;CHEN Zhiyuan;FANG Zizheng(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094)

机构地区：[1]南京理工大学计算机科学与工程学院,南京210094

出　　处：《计算机与数字工程》2024年第10期3026-3030,3058,共6页Computer & Digital Engineering

基　　金：江苏省高等教育教改研究课题“新时代荣誉学院拔尖创新人才培养的探索与实践研究”(编号:2021JSJG247);江苏省研究生科研与实践创新计划项目(编号:SJCX19_0053)资助。

摘　　要：事件因果关系抽取是信息抽取任务的重要组成之一,也是当前自然语言处理的研究热点和难点问题。事件因果关系抽取研究文本中事件之间的潜在关系,有利于深度剖析事件发展的原因和趋势,在多个领域得到了广泛的应用。根据事件因果关系抽取方法的不同,可以分为基于模板匹配、基于机器学习和基于深度学习三大类。论文对事件因果关系抽取任务进行了简单的介绍,回顾了事件因果关系抽取的发展历程。然后,介绍事件因果关系抽取的三大类方法和相关预训练语言模型,并总结展望了未来的发展趋势。Event causality extraction is one of the important components of information extraction tasks,and it is also a hot and difficult issue in current natural language processing research.Event causality extraction studies the potential relationship be-tween events in the text,which is conducive to in-depth analysis of the causes and trends of the development of events,and has been widely used in many fields.According to the different methods of event causality extraction,it can be divided into three catego-ries:based on template matching,based on machine learning and based on deep learning.This paper introduces the task of event causality extraction,and reviews the development of event causality extraction.Then,three categories of methods for event causality extraction and related pre-trained language models are introduced,and the future development trends are summarized and prospect-ed.

关 键 词：自然语言处理 因果关系抽取 机器学习 深度学习 神经网络 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]