New Graph AI method boosts recommendation accuracy with multi-level contrastive learning

A new research paper by Zhifei Hu and Feng Xia introduces a multi-level graph attention network contrastive learning framework for knowledge-aware recommendation systems. The work, published on arXiv, tackles key limitations in current graph-based recommenders: sparse user-item interaction labels, insufficient graph structure learning, and noisy entities in knowledge graphs. The proposed method enhances user representations through multi-view knowledge graph distillation, allowing more accurate modeling of preferences over entities and relations. It aggregates neighborhood entity information to build informative item representations, and introduces a multi-level self-supervised contrastive learning module that operates across three perspectives: Inter-Level (comparing different samples), Intra-Level (comparing within same class), and Interaction-Level (comparing user-item interactions).

This design improves the model's ability to generalize across intra-class samples while increasing discrimination between inter-class samples, enabling more effective multi-dimensional feature modeling. The authors conducted extensive experiments on three public datasets using both baseline and ablation settings, demonstrating that the proposed framework consistently outperforms existing state-of-the-art methods. Ablation studies further verified the effectiveness of each module. The code is expected to be released via the paper's arXiv page, making this a promising advance for production recommendation systems seeking better accuracy with noisy knowledge graphs.