Enhancing User Experience with Session-Based Recommendation Systems Using Gated Graph Convolutional Networks

Abstract

A session-based recommender system (SBRS) focuses on user’s interests depending on their browsing habits to make appropriate recommendations in a working session. Most of the existing work in growing research makes only use of the recently observed interactions on user- items. Session-based recommendation is the type where the recommender tries to suggest actions to users based on their previous anonymous session. In past works, a session is treated as a single time-point moving user models and not items models to recommend for. But they are insufficient to retrieve precise user vectors in sessions and intricate interactions of products. To achieve accurate item embedding and intricate transitions of items, we propose a novel method, called Session-Based Recommendation with Gated Graph Convolutional Neural Networks (SB-GGCNN). In this proposed method, global representations of temporal session sequences are modelled as graph-structured sequences through Gated Recurrent Units (GRUs) and GNNs to account for considerable item transitions. Furthermore, a Graph Convolutional Network (GCN) as a message passing network was applied to learn better representation of vectors in the global space. In this way, sequences are obtained: global preference of that user, current interest of that session, which is done by applying attention networks. The performance is verified utilizing extensive experiments on the SB-GGCNN model focused on e commerce datasets. Results indicated that the proposed model outperforms the best.