MFGCL: Light graph contrast learning with multi-feature information for recommendation

Collaborative filtering(CF) uses user history data to predict user preferences and recommend items that users may be interested in. Although there are many ways to use graph convolutional networks to extend traditional CF, effective information cannot be used to fully express user preferences. At the same time, the sparse interaction data between users and items in the scenario is also an important factors that affect the recommendation effect. To address the above issues, we propose a light multi-feature information-enhanced graph contrastive learning model for recommendation scenarios (MFGCL). Based on LightGCN, when propagating neighborhood information, it not only propagates the interaction information between users and items but also learns the relationship between the user and item features. This aggregation method more comprehensively represents user preferences. Secondly, to solve the problem of data sparsity, a three-channel self-supervised learning framework is constructed to generate comparative views by adding uniform noise and extracting effective information from sparse data to the greatest extent, which can improve the expressiveness and scalability of the model. Experimental results on three public datasets demonstrate the effectiveness of the proposed method, which promotes the precision and accuracy of recommendations greatly. The code can be available at https://github.com/mintZYJ/MFGCL