Secure Position-Aware Graph Neural Networks for Session-Based Recommendation

Session-based recommendation, a specific type of recommendation system, leverages users’ interaction sequences to provide recommendations. Unfortunately, these approaches tend to overlook user privacy protection and are susceptible to session sequence leakage. By introducing BGV homomorphic encryption and position information into session-based recommendation, a secure position-aware session-based recommendation is proposed. We propose the secure session-based recommendation (SSBR) and position-aware graph neural network (PA-GNN). By leveraging item embedding learning and session embedding learning with graph neural network in both local and global contexts, based on BGV fully homomorphic encryption, we provide efficient and secure session recommendations for users. While ensuring user privacy preservation, our proposed model demonstrates superior performance over state-of-the-art baselines in session-based recommendations (SBRs), as indicated by the experimental results on two benchmark datasets.