Enhanced Session-Based Recommendation Using Multi-Channel Hypergraph Network

In today’s digital landscape, the drive for supremely personalized content made Session-Based Recommendation (SBR) systems crucial. Traditional methods often overlook the complex high-order relationships among items during the recommendation process. Our paper introduces a Multi-Channel Hypergraph Convolutional Network to enhance SBR systems by addressing these gaps. Our framework first transforms session data into a hypergraph, facilitating many-to-many relationship modeling between items, beyond linear and pairwise associations. It employs hypergraph convolutional layers and a line graph channel to process these intricate relationships and capture inter-session connections. Additionally, it utilizes multi-channel convolution operations with both parallel and serial channels, varying kernel sizes, and filters designed to thoroughly process the encoded multidimensional data to extract essential features. To combat data sparsity, we apply self-supervised learning techniques to maximize mutual information between the learned session representations. This approach significantly improves the system’s accuracy and context-aware recommendations, as demonstrated by our model’s performance improvements on various datasets, including a precision at 10 (P@10) of 40.22% and a mean reciprocal rank at 20 (M@20) of 19.45% on the Diginetica dataset, reflecting substantial advancements over existing models.