融合实体语义的实体关系抽取联合解码  

Joint decoding for entity relation extraction with integrated entity semantics

作　　者：张鑫 张思佳 ZHANG Xin;ZHANG Sijia(College of Information Engineering,Liaoning Provincial Key Laboratory of Marine Information Technology,Dalian Ocean University,Dalian 116023,China;MOE Key Laboratory of Environment Controlled Aquaculture,Dalian Ocean University,Dalian 116023,China;Dalian Key Laboratory of Smart Fisheries,Dalian 116023,China)

机构地区：[1]大连海洋大学信息工程学院辽宁省海洋信息技术重点实验室,辽宁大连116023 [2]大连海洋大学设施渔业教育部重点试验室,辽宁大连116023 [3]大连市智慧渔业重点实验室,辽宁大连116023

出　　处：《现代电子技术》2024年第14期41-45,共5页Modern Electronics Technique

基　　金：辽宁省教育厅高等学校基本科研项目面上项目(LJKNZ20221095);辽宁省教育科学“十四五”规划课题(JG21DB076)。

摘　　要：针对复杂语境中存在多义词或上下文联系不强的实体,导致模型难以正确识别其关系的问题,提出一种基于BERT和联合解码的实体关系抽取模型。该模型首先采用BERT对实体进行语义编码,提取出实体的上下文信息;然后,利用自注意力机制标记出头实体,并对尾实体进行预测;最后,设计联合解码机制,结合实体语义信息和关系抽取任务进行联合解码。实验结果表明,与基准模型相比,所提模型在纽约时报(NYT)数据集和WebNLG数据集上的准确率和F1值均有所提高,能够有效地提高实体关系提取的准确性。In allusion to the problem of polysemous words or entities with weak contextual connections in complex contexts,which makes it difficult for the model to recognize their relationships correctly,an entity relationship extraction model based on BERT and joint decoding is proposed.In this model,the BERT(bidirectional encoder representations from transformers)is used to semantically encode entities and extract their contextual information.Then,the self attention mechanism is used to label the head entity and predict the tail entity.A joint decoding mechanism is designed to combine entity semantic information and relationship extraction tasks for joint decoding.The experimental results show that,in comparison with the benchmark model,the proposed model can improve the accuracy and F1 value on the New York times(NYT)dataset and WebNLG dataset,effectively improving the accuracy of entity relationship extraction.

关 键 词：实体关系抽取 实体语义 BERT 联合编码 自注意力机制 知识图谱 

分 类 号：TN919.5-34[电子电信—通信与信息系统] TP391.1[电子电信—信息与通信工程]