Session-Based Recommendation via Hierarchical Graph Learning

Session-based recommendation (SBR) is a pivotal approach aimed at forecasting users' subsequent interactions with items based on their behavioral patterns. Given the anonymity of users in this context, it becomes imperative to grasp their intentions during project transformation. While prior endeavors have predominantly concentrated on modeling user preferences within the current session, the inherent uncertainty surrounding user behavior can introduce noise into the preference signal within the user's project inter-action sequence, thereby rendering these methods insufficient for a comprehensive modeling of user-interacted items. In response to this challenge, we introduce the Dual-View Hierarchical Graph Learning Model for SBR, SRHGL, which is designed to effectively capture high-quality interactions between users and projects, consequently predicting the items that users may engage with in subsequent sessions. Our model constructs a hierarchical graph wherein the session-based graph is expanded into two distinct views, depicting the intra-session connections at the lower level and inter-session connections at the upper level. By scrutinizing these two views within the proposed hierarchical graph and leveraging diverse connectivity information recursively, we generate authentic samples. Empirical evaluations conducted on various benchmark datasets underscore the efficacy of hierarchical graph learning in augmenting session-based recommendation systems. The results demonstrate significant enhancements in performance, thereby establishing hierarchical graph learning as a promising avenue for advancing session-based recommendation methodologies. Through this approach, we achieve state-of-the-art performance levels, thus affirming the potential of our pro-posed SRHGL model in addressing the inherent challenges of session-based recommendation.