融合限定关系和交互信息的实体关系联合抽取模型  

Joint Entity and Relation Extraction Based on Constrained Relation and Interactive Information

作　　者：唐瑞雪[1,2,3] 秦永彬 陈艳平[2,3] TANG Ruixue;QIN Yongbin;CHEN Yanping(School of Information,Guizhou University of Finance and Economics,Guiyang,Guizhou 550025,China;Text Computing&Cognitive Intelligence Engineering Research Center of National Education Ministry,Guizhou University,Guiyang,Guizhou 550025,China;College of Computer Science and Technology,Guizhou University,Guiyang,Guizhou 550025,China)

机构地区：[1]贵州财经大学信息学院,贵州贵阳550025 [2]贵州大学文本计算与认知智能工程教育部工程研究中心,贵州贵阳550025 [3]贵州大学计算机科学与技术学院,贵州贵阳550025

出　　处：《中文信息学报》2024年第10期106-116,共11页Journal of Chinese Information Processing

基　　金：贵州省省级科技计划项目(黔科合基础ZK[2022]一般027)。

摘　　要：实体关系抽取作为信息抽取领域的核心任务,旨在从非结构化文本中自动抽取所有的关系三元组。现有研究较难处理句子中关系重叠的情况,存在识别冗余和语义依赖不足的问题。鉴于此,该文提出一种融合限定关系和交互信息的实体关系联合抽取模型。该模型首先对句子进行关系预测,构成限定关系集。其次,利用限定关系分别地预测可能存在关系的头实体和尾实体,解决关系重叠问题,同时缓解冗余识别。为了加强句子中实体与关系的交互,利用注意力机制强化句子中关系有关信息,通过双仿射和卷积操作来构建评分矩阵。最后,通过评分矩阵对候选三元组进行校正,确定最终的关系三元组。实验结果表明,该模型在NYT和WebNLG数据集上F1值分别达到92.0%和88.7%,相比于所对比的基线模型F1值分别提高了2.8%和1.0%,验证了模型的有效性。Entity and relation extraction,as a core task in the field of information extraction,aims to automatically extract all relation triples from unstructured text.Current researches are difficult to solve the problem of overlapping relation in the sentences owing to the redundant identification and insufficient semantic dependencies.This paper proposes a joint entity and relation extraction model with both constrained relations and interactive information.The model first predicts the relations in the sentence and constructs a set of constrained relations.Then,the model predicts the possible head and tail entities by the constrained relation,alleviating overlapping relations and the redundant identification.To strengthen the interaction between entities and relations in the sentence,the attention mechanism is used to enhance the relation-related information in the sentence,and a scoring matrix is constructed through bi-affine and convolution operation.Finally,the output relational triples are decided by the scoring matrix.Experimental results on NYT and WebNLG datasets show that the model achieves F1 scores of 92.0%and 88.7%o,respectively,outperforming the baseline by 2.8%and 1.0%.

关 键 词：实体关系抽取 联合抽取 重叠关系 限定关系 交互信息 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]