Enhanced Social Recommendation Method Integrating Rating Bias Offsets

Current social recommendations based on Graph Neural Networks (GNNs) often neglect to extract rating bias from user and item statistics, leading misinterpreting real preferences. For example, a high with lenient standards average does not always indicate preference for the item. This situation highlights inherent flaws in existing recommendation algorithms that do adequately account ratings trends. To address this problem, paper proposes an enhanced method GNNs integrated offsets (SR-BS). Firstly, we obtain users items by subtracting their value historical each user/item. enhance model’s learning capability, transform biases into vector representations. Secondly, model learning, diverse meta-paths are predefined modeling interaction relations between graph nodes (e.g., user–item–user, user–user). The aggregation of semantic information these relational paths is achieved stacking multiple GNN layers, enabling fusion higher-order information. Finally, experimental results four datasets—Ciao, Epinions, Douban, FilmTrust—show our outperforms other state-of-the-art methods tasks, exhibiting stability personalization.