基于信任传递和项目相似的图神经网络推荐模型  

Graph neural networks recommendation model based on trust transfer and item similarity

作　　者：李婷 王怀彬 LI Ting;WANG Huaibin(School of Computer Science and Engineering,Tianjin University of Technology,Tianjin 300384,China)

机构地区：[1]天津理工大学计算机科学与工程学院,天津300384

出　　处：《天津理工大学学报》2025年第1期77-82,共6页Journal of Tianjin University of Technology

基　　金：国家自然科学基金(U1536122)。

摘　　要：为了解决现有的基于图神经网络的推荐研究在利用用户的信任网络时忽略了信任关系的传递性以及存在对图节点隐含关系利用不充分等问题,提出了基于信任传递和项目相似的图神经网络推荐模型。该模型包括节点潜在特征表示以及评级预测两个方面。用户节点的特征学习主要基于信任关系的传递特性,同时获取目标用户的一阶信任和二阶信任的评级信息并利用图神经网络对用户节点实现聚合。项目节点的特征学习首先挖掘用户项目评分矩阵中隐含的项目相似关系,再应用图神经网络对项目节点进行聚合操作。评级预测通过聚合用户和项目的节点潜在特征表示预测用户对未评分项目的评级。最后,通过在两个真实的数据集上进行实验证明了模型实现评级预测的有效性。To address the existing recommendation studies based on graph neural networks that ignore the transfer characteristics of trust relationships when using users′trust networks and the existence of inadequate utilization of the implicit relationships of graph nodes,a graph neural networks recommendation model based on trust transfer and item similarity is proposed.This model includes two aspects of node potential feature representation as well as rating prediction.The feature learning of user nodes is mainly based on the transfer characteristic of trust relationship,and the rating information of first-order trust and second-order trust of target users is obtained and aggregated using graph neural networks for user nodes.The feature learning of item nodes first mines the item similarity relationship implied in the user-item rating matrix,and then applies the graph neural networks to the item nodes for aggregation operation.Rating prediction predicts users′ratings of unrated items by aggregating node potential feature representations of users and items.Finally,the effectiveness of the model to achieve rating prediction is demonstrated by conducting experiments on the two real world datasets.

关 键 词：图神经网络 推荐 信任传递 项目相似 

分 类 号：TP305[自动化与计算机技术—计算机系统结构]