语义引导的全局-局部图神经网络的关系抽取  

Relation extraction of semanitc-guided global-local graph neural networks

作　　者：任楚岚 刘长胜 邹绍强 井立志 REN Chu-lan;LIU Chang-sheng;ZOU Shao-qiang;JING Li-zhi(Department of Computer Science and Technology,Shenyang University of Chemical Technology,Shenyang 110142,China;Key Laboratory of Industrial Intelligence Technology on Chemical Process,Shenyang University of Chemical Technology,Shenyang 110142,China)

机构地区：[1]沈阳化工大学计算机科学与技术学院,辽宁沈阳110142 [2]沈阳化工大学辽宁省化工过程工业智能化技术重点实验室,辽宁沈阳110142

出　　处：《计算机工程与设计》2025年第3期705-711,共7页Computer Engineering and Design

基　　金：辽宁省教育厅科学研究基金项目(LJKZ0449);辽宁省教育厅科学研究基金项目(LJKZ0434)。

摘　　要：为解决实体间距离过长导致关系抽取性能不佳的问题,提出一种基于上下文语义引导的全局-局部图神经网络的关系抽取方法。通过注意力增强神经网络集中不同时间步的单词的重要性和相关性,获取上下文语义引导的信息;构建全局-局部图神经网络增强全局结构和局部实体之间的交互,通过改进的APPNP(approximate personalized propagation of neural predications)算法增强全局依赖关系;融合两个模块进行关系抽取。在NYT数据集上的实验结果表明,F1达到83.7%,较目前主流方法更具优势,验证了模型的有效性。To solve the poor performance of relationship extraction due to the long distance between entities,a relationship extraction method based on contextual semantic-guided global-local graph neural network was proposed.The importance and relevance of words at different time steps were focused through an attention-enhanced neural network to obtain contextual semantic guidance information.A global-local graph neural network was constructed to enhance the interaction between global structures and local entities,and the improved APPNP(approximate personalized propagation of neural predications) algorithm was proposed to enhance global dependencies.The two modules were fused to perform relationship extraction.Experiments on the NYT dataset show that the F1 reaches 83.7%,which is more advantageous than that of the current mainstream methods.The effectiveness of the model is verified.

关 键 词：关系抽取 上下文语义 注意力增强神经网络 图神经网络 全局结构 局部实体 长距离 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]