Incorporating Global Context into Multi-task Learning for Session-Based Recommendation

Abstract

Session-based recommendation aims to predict the potential items that user may interact with next time from given anonymous sessions. However, existing session-based recommendation models mainly utilize the current given session without considering the global context information. The models that take collaborative neighbor information into account are vulnerable to noise, and their performance is not efficient and stable enough. To provide better prediction, in this paper, we propose a novel model, namely GCM-SR, which incorporates Global Context into Multi-task learning for Session-based Recommendation. Rather than directly integrating the global context information with the session-level item transition information, GCM-SR regards the global context as implicit type information and integrates it in the form of an auxiliary task to enhance recommendation performance. Then, local learning task and global learning task are joined through adjustable weights to complete prediction and recommendation. Experiments on three public datasets demonstrate the superiority of GCM-SR over the state-of-the-art models. The results may give suggestion for better performance in session-based recommendation systems.