基于图注意神经网络的知识图谱推荐系统  

Knowledge Graph-based Recommendation System with Graph Attention Neural Networks

作　　者：袁翔 宋媛媛 杨熙鑫 YUAN Xiang;SONG Yuanyuan;YANG Xixin(College of Computer Science&Technology,Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学计算机科学技术学院,山东青岛266071

出　　处：《青岛大学学报(工程技术版)》2025年第1期17-22,48,共7页Journal of Qingdao University(Engineering & Technology Edition)

基　　金：山东省自然科学基金(ZR2024ME210);山东省重点研发计划(科技型中小企业创新能力提升工程)(2024TSGC0268);青岛市自然科学基金(23-2-1-216-zyyd-jch;23-2-1-125-zyyd-jch)。

摘　　要：针对当前推荐系统对邻阶信息利用不高的问题,将知识图谱应用到推荐系统,增加注意嵌入传播层,学习知识图谱中与实体相邻的节点信息,通过挖掘实体之间的相关属性,能够有效的学习实体间的关系,大幅度提高信息的利用效率以及推荐系统的性能。为验证该方法的有效性和正确性,在两个知识图谱数据集上进行实验,并与相关基线模型进行对比。实验结果表明,本方法能够有效利用邻阶信息,提高推荐系统正确率。Addressing challenges of inadequate utilization of higher-order neighbor information in conventional recommendation systems,this study introduces the integration of knowledge graphs into the recommendation framework.By incorporating an attention-based embedding propagation layer,it effectively harnesses neighboring node information from the knowledge graph,enabling the extraction of relevant attributes between entities,facilitates the modeling of inter-entity relationships,thereby significantly enhancing the efficiency of information propagation and improving the overall performance of the recommendation system.To assess the efficacy and validity of the proposed method,experiments are conducted using two knowledge graph datasets,and the results are compared against several state-of-the-art baseline models.The experimental findings indicate that the proposed method can effectively leverage higher-order neighbor information,leading to a significant improvement in recommendation accuracy.

关 键 词：知识图谱 注意力机制 推荐系统 图注意神经网络 

分 类 号：TP24[自动化与计算机技术—检测技术与自动化装置]