Scholarlink

<p>To improve content delivery efficiency, next-generation data centers employ joint caching and recommendation mechanisms to strategically place popular content at the network edge. Despite its effectiveness, this approach has two significant limitations: (1) increased computational overhead and (2) reduced accuracy. To mitigate this challenge, we adopt a hybrid approach. On one hand, we leverage Probabilistic Matrix Factorization (PMF) to capture collaborative patterns in user–content interactions, and integrate Bidirectional Encoder Representations from Transformers (BERT) to extract rich semantic information from contextual metadata. On the other hand, we apply Principal Component Analysis (PCA) to enhance interpretability and preserve key data patterns, followed by Fuzzy C-Means (FCM) clustering to segment the user–item interaction space into coherent and meaningful subgroups We conducted experiments on two widely used MovieLens datasets to assess the effectiveness of the proposed model in caching optimization. The proposed model significantly outperforms the PMF baseline, achieving 32.00% and 36.06% improvements in HR@10, as well as 32.19% and 36.27% improvements in NDCG@10 on the MovieLens 1&#xa0;M and MovieLens 100&#xa0;K datasets, respectively. By integrating PCA with FCM, the model achieves efficient user segmentation, resulting in a nearly 90% reduction in execution time compared to the baseline hybrid Model.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimizing data caching and placement in next-generation data centers using transformer and probabilistic matrix factorization

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • Mbarek Marwan,
  • Yassine Afoudi,
  • Safae Hmaidi,
  • Abdelkarim Ait Temghart,
  • Mohamed Lazaar

摘要

To improve content delivery efficiency, next-generation data centers employ joint caching and recommendation mechanisms to strategically place popular content at the network edge. Despite its effectiveness, this approach has two significant limitations: (1) increased computational overhead and (2) reduced accuracy. To mitigate this challenge, we adopt a hybrid approach. On one hand, we leverage Probabilistic Matrix Factorization (PMF) to capture collaborative patterns in user–content interactions, and integrate Bidirectional Encoder Representations from Transformers (BERT) to extract rich semantic information from contextual metadata. On the other hand, we apply Principal Component Analysis (PCA) to enhance interpretability and preserve key data patterns, followed by Fuzzy C-Means (FCM) clustering to segment the user–item interaction space into coherent and meaningful subgroups We conducted experiments on two widely used MovieLens datasets to assess the effectiveness of the proposed model in caching optimization. The proposed model significantly outperforms the PMF baseline, achieving 32.00% and 36.06% improvements in HR@10, as well as 32.19% and 36.27% improvements in NDCG@10 on the MovieLens 1 M and MovieLens 100 K datasets, respectively. By integrating PCA with FCM, the model achieves efficient user segmentation, resulting in a nearly 90% reduction in execution time compared to the baseline hybrid Model.