A Behavior Relation-Aware Graph Contrastive Networks for Multi-Behavior Recommendation

Multi-behavior recommendation is an effective solution to alleviate the data sparsity problem faced by traditional multi-behavior models. It uses various types of user-interactions (e.g., browsing, adding to cart and purchasing) to learn user preferences. However, the interactions between users and items are often complex and highly dependent. To address this challenge, we propose a multi-behavior recommendation system based on cascaded residual and graph contrastive networks, which aims to leverage the interaction behavior information and contrastive enhancement information to improve the accuracy of recommendation. The model has two notable advantages: (1) It models each behavior individually, constructing user and item representations from different behavior graphs, and then refines each user-item representation and aggregates them through a cascaded residual graph network, greatly alleviating the data sparsity problem. (2) It adds a contrastive enhancement module to capture the information of high-order heterogeneous paths and reduce information loss. Finally, we use a multi-task learning approach to jointly optimize our model according to different behaviors. Experimental outcomes on two authentic datasets reveal that, in comparison with the optimal benchmark model, the proposed system (BRGCNet for short) attains over a 2% enhancement in Recall@50 and above a 3% increase in NDCG@50, substantiating the efficacy of the devised approach.