关键词: 关键词: 知识感知推荐, 异构拓扑协同, 层次对比学习, 动态偏好建模, 知识图谱

Abstract: Abstract: To address knowledge-aware recommendation limitations, including inadequate high-order semantic modeling, shallow item-correlation mining, and weak exploitation of knowledge-graph structure, a knowledge-aware Recommendation performance enhancement method based on Hierarchical Multi-topology collaborative Contrastive Learning(HMCLR) was proposed. First, three heterogeneous topology graphs were constructed: a collaborative filtering-based item-item graph(CFIG) was designed to capture collaborative filtering signals; a knowledge graph-based item-entity graph(KGIG) was established to mine semantic associations; and a user preference graph(UPG) was developed to model multi-hop interest evolution via dynamic attention mechanisms. Then, a hierarchical contrastive learning strategy was engineered to form a complementary mechanism with heterogeneous topological views. In the Local Layer, semantic fidelity was ensured through direct interaction anchoring. In the Dynamic Layer, attention-weighted aggregation was utilized to enhance multi-hop relation propagation and scenario adaptability. In the Global Layer, multi-view information was fused to guarantee semantic consistency. Finally, regularization constraints were adopted to suppress overfitting, combined with an integrated loss function for method optimization. Experimental results demonstrate that HMCLR achieves the highest AUC (Area Under Curve) and F1 (F1-score), outperforming the best baseline MCCLK by 2.23 and 2.19 percentage points on Book-Crossing, 1.24 and 1.78 percentage points on MovieLens-1M, and 1.86 and 1.97 percentage points on Last.FM. Ablation studies confirm the critical contributions of multi-view collaboration and hierarchical contrastive learning to performance enhancement.

Key words: Keywords: knowledge-aware recommendation, multi-topology collaboration, hierarchical contrastive learning, dynamic preference modeling, knowledge graph