多注意力机制的文本强化关系抽取方法  

Text-enhanced relationship extraction method with multiple attention mechanisms

作　　者：丁翠 王占刚[1] DING Cui;WANG Zhan-gang(School of Information and Communication Engineering,Beijing Information Science and Technology University,Beijing 100101,China)

机构地区：[1]北京信息科技大学信息与通信工程学院,北京100101

出　　处：《计算机工程与设计》2025年第4期1103-1110,共8页Computer Engineering and Design

基　　金：国家重点研发计划课题基金项目(2018YFC1800203);北京市科技创新服务能力建设-基本科研业务费(市级)(科研类)基金项目(PXM2019_014224_000026)。

摘　　要：关系抽取作为知识图谱的关键技术,存在关系抽取干扰、模型适用性低等问题亟待解决。研究优化实体注意力机制,结合位置特征向量联合约束;引入句子注意力机制,增强关系抽取效果;自适应更新机制,充分利用知识图谱的语义信息,提升关系抽取效果。在多个数据集进行测试,结果表明,相比其它算法,该算法拥有更好的抽取效果,有效解决了句子间的隐含关系问题。Within the knowledge graph domain,relationship extraction was identified as a crucial technology,facing challenges such as interference in relationship extraction and low model applicability that urgently need solutions.focusing on optimizing entity attention mechanisms,positional feature vectors were incorporated with joint constraints.A sentence attention mechanism was introduced to enhance the effectiveness of relationship extraction.The adaptive update mechanism was used to make full use of the semantic information in the knowledge graph to enhance the effectiveness of relation extraction.The proposed algorithms were tested on multiple datasets.The results demonstrate superior extraction performances compared to that of other algorithms,effectively addressing implicit relationship issues between sentences.

关 键 词：知识图谱 关系抽取 注意力机制 句子注意力 远程监督 抗干扰实验 信息管理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]