摘要
篇章事件抽取任务从给定的文本中识别其事件类型和事件论元。该任务包括事件检测和论元识别两个子任务。目前篇章事件普遍存在数据稀疏和多值论元耦合的问题。基于此,该文提出了一种基于框架语义映射和类型感知的篇章事件抽取方法。该文分析发现事件描述与框架语义网有着相似的结构,因此将汉语框架网(CFN)与中文篇章事件建立映射,并在文本输入层引入触发词释义信息和滑窗机制充分感知上下文信息,改善了事件检测的数据稀疏问题;使用基于类型感知标签的多事件分离策略缓解了论元耦合问题。为了提升模型的鲁棒性,进一步引入对抗训练。在DuEE-fin和CCKS2021数据集上的实验结果显示,该文模型较当前主流模型测试结果有较大提升,验证了方法的有效性。
Document-Level event extraction,consisting of two subtasks of event detection and argument identification,identifies the event type and event arguments from a given text.This paper proposes a document-level event extraction method based on frame semantic mapping and type-awareness.To leverage the similar structure in frame semantic nets,a mapping is established between Chinese FrameNet(CFN)and Chinese text events.The trigger word interpretation information and the sliding window mechanism are introduced to fully perceive the context information in the text input layer.A multi-event separation strategy based on type-aware labels alleviates the problem of argument-coupling.In order to improve the robustness of the model,adversarial training is further introduced.Experimental results on DuEE-fin and CCKS2021 datasets show that the proposed method has a significant improved performance compared with the current mainstream models.
作者
卢江
苏雪峰
李茹
闫智超
陈加兴
LU Jiang;SU Xuefeng;LI Ru;YAN Zhichao;CHEN Jiaxing(School of Computer and Information Technology,Shanxi University,Taiyuan,Shanxi 030006,China;MOE Key Laboratory of Computational Intelligence and Chinese Information Processing,Shanxi University,Taiyuan,Shanxi 030006,China;School of Modern Logistics,Shanxi Vocational University of Engineering Science and Technology,Jinzhong,Shanxi 030609,China)
出处
《中文信息学报》
CSCD
北大核心
2024年第5期53-64,共12页
Journal of Chinese Information Processing
基金
国家自然科学基金(61936012)
山西省重点研发计划项目(202102020101008)
山西省基础研究计划(202203021211286)。
关键词
汉语框架网
框架语义映射
类型感知
事件抽取
Chinese FrameNet
frame semantic mapping
type awareness
event extraction
