Efficient Edge Caching in Data Centers Using Latent Dirichlet Allocation and Hybrid Matrix Factorization With SVD++ and ALS
Mbarek Marwan, Hind Ait Temghart, Safae Hmaidi, Mohamed Lazaar · IEEE Access · 2026
Intelligent edge caching has emerged as a promising solution for supporting next-generation data centers by improving content delivery. However, the increasing complexity of deep neural network architectures often results in significant computational overhead and increased service latency. To overcome these limitations, this study introduces a lightweight hybrid matrix factorization framework that combines Singular Value Decomposition (SVD++) and Alternating Least Squares (ALS) based on the Bates–Granger (BG) weighting scheme. Furthermore, Latent Dirichlet Allocation (LDA) is employed for topic-based content clustering, enabling a better trade-off between recommendation accuracy and computational efficiency. The simulation results, statistically validated using the Friedman test followed by the Nemenyi post-hoc analysis, demonstrate that the proposed model consistently outperforms competing approaches. Specifically, the hybrid model achieves accuracy improvements of 20–30% on the dense Amazon Reviews dataset and 4–10% on the sparse Book-Crossing dataset compared with both classical and neural baseline models. In addition, clustering improves computational efficiency by reducing training time by 10–42% compared with models such as Long Short-Term Memory (LSTM) networks, Convolutional Neural Networks (CNNs), and Graph Neural Networks (GNNs). Overall, the results indicate that the proposed proactive caching technique achieves a better balance between accuracy and efficiency, while cross-validation confirms its robustness and strong generalization capability across different data partitions.