Enhancing Session-Based Recommendation with Lossless Gated Graph Neural Networks
Boudjemaa Boudaa, Ilias Sid Ahmed Makboul, Mohamed Ali Gafour, Sara Melissa Ait Arab · 2024
The last decade has witnessed the emergence of Session-Based Recommendation (SBR) as a novel type of recommendation system (RS) that captures dynamic and shortterm user preferences within a session, aiming to offer more relevant and precise next-item recommendations. Recently, there has been a growing emphasis on utilising graph neural networks (GNN) to enhance SBR development. However, most GNN-based approaches in SBR neglect the relevant order graph of user-item interactions when constructing their graphs, resulting in the loss of crucial session-specific information. To address this limitation and incorporate this sequence order information into GNN, the present paper introduces a novel model named L-GGNN-ATT. It leverages Lossless Gated Graph Neural Networks to learn item embeddings that capture intricate ordered item transitions, and a soft-attention mechanism as Readout to aggregate global user preferences. To validate the effectiveness of the proposed method, extensive experiments were conducted on two real datasets. The results demonstrate the superiority of L-GGNN-ATT over several state-of-the-art approaches in the context of session-based recommendation.