A novel image recommendation model based on user preferences and social relationships

Abstract

Recent work on personalized recommendation advocates that exploring user visual preferences from their interactions with images has become a future endeavor for image recommendation. However, there are inevitable discrepancies between the images favored and the visual preference representations reflected by users, therefore we strive to reveal user visual preferences from their behaviors and social relationships. In this work, we propose a collaborative graph model of user preferences for image recommendation, which establishes a higher-order collaborative relationship between user visual preferences and image semantic information. In user visual preferences, the underlying visual preferences hidden in social relationship is explored by combining the idea of contrast learning. In the higher-order collaboration module, a recursive interest propagation strategy is used to aggregate user preferences and image visual information. The experimental results clearly show that the model has an improvement of about 8.82% and 7.25% on Recall@20 and NDCG@20 compared to the latest methods.

Keywords

• Graph neural network
• Social networks
• User preference understanding
• Image recommendation