Deep Transfer Recommendation Model with Spatio-Temporal Enhancement and Adaptive Gating Fusion

Existing recommendation models often suffer from low initial accuracy, poor cold-start performance, and weak trans fer adaptability, making it difficult to precisely capture user prefe rences. To address these limitations, this paper proposes a Deep <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{T}$</tex> ransfer Recommendation Model with Spatiotemporal Enhanceme nt and Adaptive Gating Fusion (STAGF-DTRM). The model leve rages a spatiotemporal collaborative encoding module to capture sequential dependencies and spatial correlations in user behavior. By integrating transfer learning strategies, it effectively mitigates cold-start issues and data sparsity. Additionally, a graph attention network is employed to dynamically adjust feature weights, whi le a dynamic gating mechanism enhances the model's ability to ad aptively select critical features. Experimental results demonstrate that STAGF-DTRM achieves the highest AUC scores across mult iple datasets (Yoochoose1/64, Criteo, and Amazon) while significa ntly reducing Logloss, highlighting its superior recommendation accuracy and robustness. This framework offers a novel technical perspective and a practical solution for designing cross-domain recommendation systems.