融合知识图谱和图注意力网络的旅游推荐算法  

Tourism recommendation algorithm based on knowledge graph and graph attention network

作　　者：徐春[1] 王萌萌[1] 孙彬[1] XU Chun;WANG Meng-meng;SUN Bin(School of Information Management,Xinjiang University of Finance and Economics,Urumqi 830012,China)

机构地区：[1]新疆财经大学信息管理学院,新疆乌鲁木齐830012

出　　处：《计算机工程与设计》2024年第5期1420-1427,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(62266041);新疆财经大学科研基金项目(2022XGC073)。

摘　　要：为缓解旅游推荐模型面临的数据稀疏和冷启动的问题,提出一种融合知识图谱和图注意力网络的旅游推荐算法KRGAT(knowledge ripple graph attention network)。借助水波网络从用户的历史旅游行为和知识图谱中挖掘用户偏好增强用户特征表示,针对当前旅游项目特征学习的方法难以提取节点深层特征的问题,利用图注意力网络聚合相关度更高的邻居节点信息,增强旅游项目特征表示。实验在自建立的旅游数据集上与5个基线方法进行对比,其结果表明,KRGAT的精确率(P)、召回率(R)和AUC值分别提升了5.73%、4.42%和1.42%。To alleviate the problems of data sparsity and cold start faced by tourism recommendations model,a tourism recommendation algorithm combining knowledge graphs and graph attention networks was proposed.RippleNet was used to mine user preferences from user historical tourism behaviors and knowledge graphs to enhance user feature representation.In response to the problem of difficulty in extracting deep node features using current tourism project feature learning methods,graph attention networks were used to aggregate neighbor node information with higher correlation to enhance tourism projects feature representation.The experiment was compared with five baseline methods on a self-established tourism dataset.The results show that the accuracy,recall,and AUC values of KRGAT are improved by 5.73%,4.42%,and 1.42%,respectively.

关 键 词：旅游推荐算法 图注意力网络 知识图谱 水波网络 注意力机制 大语言模型 知识表示学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]