基于图神经网络的社会感知顺序推荐模型

SOCIAL PERCEPTION SEQUENTIAL RECOMMENDER SYSTEMS MODEL BASED ON GRAPH NEURAL NETWORK

由于用户的偏好是动态多变的,且受社交关系的影响,传统的推荐方法常常考虑不周全。针对此问题,提出一种基于图神经网络的社会感知顺序推荐模型(GASR)设计动态兴趣提取层来捕捉用户的动态偏好。同时设计社会感知层,利用图神经网络构建用户的社交关系图,使用注意力聚合方法来权衡不同朋友对用户偏好的影响。在两个实际数据集上的实验结果表明,该模型优于包括现有最新的社交推荐模型以及几个具有竞争性的基线模型。

Since user preferences are dynamic and changeable and are affected by social relationships,traditional recommendation methods are often incomplete.To solve this problem,a social perception sequential recommendation model based on graph neural network(GASR)is proposed.The dynamic interest extraction layer was designed to capture the user s dynamic preferences.The social perception layer was designed and the graph neural network was used to construct the user s social relationship graph.The attention aggregation method was used to weigh the influence of different friends on user preferences.Experimental results on two actual data sets show that the model is superior to the latest social recommendation models and several competitive baseline models.