基于迁移学习的小样本事件抽取技术及军事应用展望  

Small Sample Event Extraction Technique Based on Transfer Learning and the Prospect of Military Application

作　　者：刘涛 蒋国权[1] 刘姗姗 周泽云 陈涛 LIU Tao;JIANG Guoquan;LIU Shanshan;ZHOU Zeyun;CHEN Tao(The Sixty-third Research Institute,National University of Defense Technology,Nanjing 210007,China;School of Computer Science,School of Software,School of Cyberspace,Nanjing University of Information Science and Technology,Nanjing 210044,China;The Information Center of Equipment Development Department,Beijing 100034,China;A Bureau of the Equipment Development Department,Beijing 100034,China)

机构地区：[1]国防科技大学第六十三研究所,南京210007 [2]南京信息工程大学计算机学院软件学院网络安全空间学院,南京210044 [3]装备发展部信息中心,北京100034 [4]装备发展部某局,北京100034

出　　处：《火力与指挥控制》2023年第10期9-17,共9页Fire Control & Command Control

基　　金：国家自然科学基金;国家高技术研究发展计划(863计划)资助课题(2008AA000000)。

摘　　要：事件抽取作为信息抽取的任务之一,旨在从非结构化文本中抽取出结构化事件信息,从而更好地应用在相关应用领域和下游任务上。基于迁移学习的事件抽取技术是当前低资源场景下的事件抽取研究的主流方法,通过设计更鲁棒的模型来迁移知识,解决事件抽取面临的训练数据缺乏问题,从而提升小样本事件抽取的效果。对基于迁移学习的小样本事件抽取技术及其军事应用展望作了全面的阐述,回顾事件抽取的起源与发展,描述迁移学习事件抽取的研究背景,并重点总结目前的技术方法及其军事研究现状,提出其军事上应用展望,最后对其面临的问题挑战和未来研究热点作了分析。As one task of information extraction,event extraction aims to extract structured event information from unstructured text,so as to be better applied to the related application fields and downstream tasks.Event extraction based on transfer learning is the mainstream method of event extraction research in current low resource scenario.A more robust model is designed to transfer knowledge and solve the training data scarcity problems faced by event extraction so that the effects of small sample event extraction can be improved.The technology of small sample event extraction based on transfer learning and the prospect of military application are expounded comprehensively.Firstly,the origin and development of event extraction are reviewed,the research background of transfer learning event extraction is described,and the present technological methods and military research status are summarized emphatically,and then the prospect of military application in future is presented.Finally,the problems and challenges it faces as well as research hotspots in the future are analyzed.

关 键 词：事件抽取 迁移学习 小样本 低资源场景 军事应用展望 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]