基于标签校正的端到端实体关系联合抽取  

End-to-end Entity Relationship Joint Extraction Based on Label Correction

作　　者：陈鹏之 张瑾[1] 刘悦[1] 程学旗[1] CHEN Pengzhi;ZHANG Jin;LIU Yue;CHENG Xueqi(CAS Key Lab of Network Data Science and Technology,Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院计算技术研究所中国科学院网络数据科学与技术重点实验室,北京100190 [2]中国科学院大学,北京100049

出　　处：《重庆理工大学学报（自然科学）》2021年第10期163-171,共9页Journal of Chongqing University of Technology：Natural Science

摘　　要：随着深度学习的快速发展与应用,联合式抽取被广泛应用于实体抽取和实体间的关系预测。虽然端到端的联合抽取方法在该领域得到了较大关注,但这类方法目前未考虑multi-token实体;同时,抽取过程中忽略了关系预测与实体抽取之间的相互影响。针对以上问题,结合Encoder-Decoder框架的特点,引入标签校正机制,提出了一种基于标签校正的端到端实体关系联合抽取方法CopyLC。实验结果证明:在更严格的评价方式下,所提出的方法与当前主流方法相比,在NYT和WebNLG数据集上均能获得更好的抽取效果。Entity relationship extraction is one of the important research topics in information extraction.In recent years,with the rapid development and application of deep learning,joint extraction has been widely used in entity extraction and relationship prediction between entities.Although the end-to-end joint extraction method has received considerableattention in this domain,the multi-token entities is seldom considered,simultaneously,the interaction between relationship prediction and entity extraction is ignoredduring the extraction process.Therefore,this paper combines the characteristics of the Encoder-Decoder framework and introduces a label correction mechanism,then proposes an end-to-end entity relationship joint extraction method CopyLC based on label correction.Experimental results prove thatthe proposed method can obtain better extraction results on NYT and WebNLG datasets,by comparing with the current mainstream methods under the strict evaluation method.

关 键 词：实体关系抽取 端到端 标签校正 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]