基于自监督的多视角图协同过滤推荐方法OA北大核心CSTPCD

Existing graph collaborative filtering algorithms suffer from data sparsity in real-world scenarios and make feature learning more susceptible to interaction noise when aggregating adjacent information.To address these issues,a recommendation method based on Self-supervised Multi-view Graph Collaborative Filtering(SMGCF)is proposed.The SMGCF learns the embedded representations of the user and item nodes by using Graph Neural Network(GNN).In the process of learning the embedding representation of nodes,self-supervised learning is introduced to assist the graph collaborative filtering algorithm in mining relationships from multiple views,considering the influence of the interaction relationships between individual nodes and the clustering relationships between clustered nodes on the recommendation results.For the node-interaction-level relationship view,multiple user-item interaction bipartite graphs are obtained by data augmentation,and a contrastive learning method for node-interaction-level relationships is proposed.For the node-clustering-level relationship view,a contrastive learning method for node-clustering-level relationships is proposed.The node-embedding effect is enhanced by fusing the two types of contrast learning methods through a multi-view integration strategy.Experiments are conducted using four public datasets.The experimental results demonstrate the feasibility and effectiveness of the SMGCF.Compared with the best-performing baseline method,NCL and SMGCF achieved the highest improvements of 2.1％in Recall@10 metric and 4.3％in NDCG@10 metric.