A Graph ATtention Networks Model for Session-Based Recommender Systems

In recent years, session-based recommender systems (SBRSs) have emerged as a new paradigm of recommender systems to help users in their different decision-making processes. The goal of SBRSs is to capture dynamic and short-term user preferences within sessions to provide more timely and accurate next-item recommendations that are sensitive to be adapted in different contexts. In the literature, the proposed approaches for SBRS development are limited to some models that lack more precision and efficacy, which are the primary purposes for this kind of recommender system. This paper aims to formally present a graph neural networks model for session-based recommender systems based on Graph ATtention networks (GAT). GAT can capture complex transitions between items through a session modelled as graph-structured data. The effectiveness of the pro-posed GAT-based model is extensively evaluated on three public real-world datasets. Experimental results show the superiority of our model against some used baselines in the SBRS field.