A User-Interaction Parallel Networks Structure for Cold-Start Recommendation

Abstract

AbstractThe goal of the recommendation system is to recommend products to users who may like it. The collaborative filtering recommendation algorithm commonly used in recommendation systems needs to collect explicit/implicit feedback data, and new users do not leave behavioral data on the product, which leads to cold-start problem. This paper proposes a parallel network structure based on user interaction, which extracts features from user interaction information, social media information, and comment information and forms a matrix. The graph neural network is introduced to extract high-level embedded correlation features and the role of parallelism is to reduce computing cost further. Experiments based on standard data sets prove that this method has a certain degree of improvement in NDCG and HR indicators compared to the baseline.