A Feature-Weighted Attention Diffusion Model for Social Recommendation

Social recommendation leverages social relationships among users to enhance personalized recommendations. However, existing methods face challenges in handling noisy or irrelevant social ties, which can reduce recommendation accuracy and affect the modeling of user-item interactions. To address these issues, we propose a Feature-Weighted Attention Diffusion Model for social recommendation. The model integrates attention mechanisms and diffusion processes to selectively enhance relevant user-item interactions while reducing the influence of irrelevant social connections. A multi-step diffusion process in the embedding space is employed to refine user representations, improving their robustness against noise. The attention mechanism assigns appropriate weights to informative features during the diffusion process, supporting the modeling of user preferences and interactions in social environments. Experiments on the Ciao and Yelp recommendation datasets demonstrate that the proposed model excels in recommendation accuracy, robustness, and training efficiency.