基于格点网格与段尺度注意力机制的知识图谱构建  

Knowledge graph construction based on grid and segment attention mechanism

作　　者：王体春[1,2] 李昊[1,2] 王贤伟 WANG Tichun;LI Hao;WANG Xianwei(National Key Laboratory of Helicopter Aeromechanics,Nanjing University of Aeronautics and Astronautics,Nanjing 210000,China;College of Mechanical and Electrical Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210000,China)

机构地区：[1]南京航空航天大学直升机动力学全国重点实验室,江苏南京210000 [2]南京航空航天大学机电学院,江苏南京210000

出　　处：《计算机集成制造系统》2025年第4期1368-1382,共15页Computer Integrated Manufacturing Systems

基　　金：直升机动力学全国重点实验室基金资助项目(2023-HA-LB-067-07);国家自然科学基金(面上)资助项目(51775272);江苏省自然科学基金(面上)资助项目(BK20221481)。

摘　　要：为解决当前知识图谱构建模型过程中训练样本特征单一、关系抽取准确率低下的问题,建立一种基于格点网格与段尺度注意力机制的知识图谱自动构建模型(KG-GSAM)。针对实体识别任务,引入格点网格结构对双向门控循环神经网络进行改进;针对关系抽取任务,引入段尺度注意力机制,搭建关系抽取神经网络。在公开数据集和近三年自动导引车领域的专利文本构建的数据集上分别进行实验,结果表明所建立模型在Precision、Recall和F1-score三个指标上与其他知识图谱构建模型相比有一定的优越性。To address the issues of single training sample features and low accuracy in relation extraction during the current knowledge graph construction process,a Knowledge Graph based on Grid and Segments Attention Mechanism(KG-GSAM)was established.For entity recognition tasks,a lattice structure was introduced to improve the bidirectional gated recurrent neural network;for relation extraction tasks,a segment scale attention mechanism was introduced to build a relation extraction neural network.Experiments were conducted on publicly available datasets and datasets constructed from patent texts of automated guided vehicles in the past three years.The results showed that the proposed model had certain advantages over other knowledge graph construction models in terms of Precision,Recall,and F1 score.

关 键 词：知识图谱 格点网格 段尺度注意力机制 BERT模型 关系抽取神经网络 自动导引车 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]