Multi-view session-based social recommendation with self-supervised learning

Session-based social recommendation (SSR) is an important branch of session-based recommendation that aims to generate more accurate recommendations by combining users’ historical sessions with their social information. Existing SSR models are usually based on graph neural networks. Although they have shown better performance compared to traditional SSR models, they ignored users’ temporal behavior patterns and data sparsity issues. Therefore, we propose a multi-view SSR model MVSSR that effectively constructs multiple views to model the temporal and global user preferences in sparse SSR scenarios. On one hand, we construct user data into multiple views to thoroughly capture users’ temporal preferences and global preferences. On the other hand, we constructed a view-aware self-supervised learning method. Through self-supervised learning within and between views, we enhance the user’s session representation to alleviate the data sparsity is-sues in SSR. We performed ex-tensive experiments on three real-world datasets to confirm the effectiveness of MVSSR. The results indicate that our model surpasses the current state-of-the-art models in terms of hit rate and mean reciprocal rank.