An Adaptive Session Selection Method for Recommendation System

Predicting user's next action based on anonymous session is a challenging problem in session-based recommendation. Recent advances utilize attention mechanism and graph neural network to achieve excellent performance in session-based recommendation. However, most of these studies ignore the characteristics of different types of sessions, which results in user's preference from each session not being captured by the most suitable model. In this paper, we propose an adaptive session selection method for the session-based recommendation called ASSM to address these issues. In ASSM, sessions are adaptively distinguished into frequent sessions and infrequent sessions based on whether they contain high frequency items. Then a graph structure is constructed for frequent session to learn complex item transition and obtain user's local preference via graph neural network. At the same time, an improved attention network is applied to capture user's global preference from infrequent session. We conduct extensive experiments on two real-world datasets and the results demonstrate the effectiveness of our method.