DF-AT: A Neural Influence Diffusion for Social Recommendation Based on Attention Mechanism

Conventional recommender systems that based on Collaborative Filtering face difficulties like sparse data and the cold start issue that limit their performance and effectiveness to some extent. To address these issues, social recommendation has emerged as a research hotspot. Wu et al. proposed a neural influence diffusion network (DiffNet) model for social recommendation, which has achieved impressive results by simulating the latent embedding of users that changes with the diffusion process of social influence. By incorporating the attention mechanism into the model, this paper proposes a novel model DF-AT assigns corresponding weights to users based on their social influence and the importance of their interests, which adjusts their influence on users' interests during the information dissemination process, leads to more accurate preference predictions. Finally, experiments are carried out on two data sets, and the result shows the validity of the enhanced method.