基于潜层关系增强的实体和关系联合抽取  

Joint extraction of entities and relationships based on subtext relationship enhancement

作　　者：王鹏 刘小明[1,2] 杨关[1,2] 刘杰 刘阳[5] WANG Peng;LIU Xiao-ming;YANG Guan;LIU Jie;LIU Yang(School of Computer Science,Zhongyuan University of Technology,Zhengzhou 450007,China;Henan Key Laboratory on Public Opinion Intelligent Analysis,Zhongyuan University of Technology,Zhengzhou 450007,China;School of Information Science,North China University of Technology,Beijing 100144,China;Research Center for Language Intelligence of China,State Language Commission,Beijing 102206,China;School of Telecommunications Engineering,Xidian University,Xi’an 710071,China)

机构地区：[1]中原工学院计算机学院,河南郑州450007 [2]中原工学院河南省网络舆情监测与智能分析重点实验室,河南郑州450007 [3]北方工业大学信息学院,北京100144 [4]国家语委中国语言智能研究中心,北京102206 [5]西安电子科技大学通信工程学院,陕西西安710071

出　　处：《计算机工程与设计》2024年第6期1780-1788,共9页Computer Engineering and Design

基　　金：国家科技创新-2030重大基金项目(2020AAA0109700);国家自然科学基金项目(62076167);东北师范大学应用统计教育部重点实验室基金项目(135131007);国家自然科学基金青年基金项目(61906141)。

摘　　要：为充分发掘文本序列中潜层语义关系信息,提出一种实体和关系联合抽取的潜层关系增强模型SREM(text subtext relationship enhancement model)。在潜层关系表示层利用结构化对齐的方式,获取并保持文本序列中的语义信息结构。在融合注意力机制的关系网络层中对数据进行建模,提高模型对文本词汇间关系信息的捕获能力。结合注意力机制获取细粒度语义信息,对上下文信息进行选择过滤。实验结果表明,在数据集NYT和WebNLG上取得的F1值分别为92.40%和92.52%,验证了模型的有效性。To fully explore the latent text semantic relations in text sequences,a latent relationship enhanced joint extraction model SREM(text subtext relationship enhancement model)based on the relationship attention mechanism was proposed.The structured alignment was used at subtext relationship presentation layer to maintain semantic information structure in the text sequence.In the relationship network layer where the attention mechanism was integrated,the relationship network was used to model the data to improve the model’s ability to capture the relationship information between text and vocabulary.The relationship attention mechanism was combined to obtain fine-grained semantic information,and the context information was selected and filtered to reduce the impact of useless information.Experimental results of the model on two public datasets NYT and WebNLG,achieves F1 values of 92.40%and 92.52%,which verify that the model is effective.

关 键 词：联合抽取 语义关系 结构化知识 潜层表示 注意力机制 关系网路 信息过滤 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]