A Switching Hybrid CNN-based Group Recommendation Framework for Edge Caching in Data Centers Using ConvSVD++ and Graph Convolutional Networks

Read full text Summarise with VIRAT
Authors: Mbarek Marwan, Hind Ait Temghart, Youssef Ben Amar, Abdelkarim Ait Temghart, Mohamed Lazaar
Cited by: 0
References: 23

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Since edge data centers typically serve a large number of heterogeneous IoT devices and users, we suggest the use of a group-based recommender system for popularity-driven content caching through a Virtualized Internet of Everything (VIoE) method.We leverage CNNs to obtain rich content representations, GNNs to encode high-order structural dependencies among users and items, and SVD++ to uncover latent interactions between users and items.Furthermore, we introduce a weight-based user clustering strategy that combines Fuzzy C-Means (FCM) with the Pearson Correlation Coefficient (PCC) to construct coherent preference groups, enabling robust and scalable recommendation at the edge node.This approach improves both accuracy and fairness by combining soft membership from FCM with the flexibility of enhanced similarity computations using PCC during group modeling of users or IoT devices.Experimental evaluations conducted on the MovieLens-1M and Amazon Review datasets demonstrate that the Group GCN model achieves superior performance in ranking-oriented tasks, while Group ConvSVD++ is more effective for rating prediction.Notably, Group GCN achieves the highest performance across top-K evaluation metrics, with Precision@10 of 0.1283, F1-score@10 of 0.1405, and NDCG@10 of 0.1583.Conversely, Group ConvSVD++ yields the lowest prediction errors, achieving an MAE of 0.6011 and RMSE of 0.8923.These complementary findings strongly justify the proposed hybrid switching mechanism, in which graph-based models are preferred for rankingoriented objectives, whereas matrix-factorization-based models achieve superior accuracy in rating prediction.The proposed framework demonstrates superior performance over all baseline methods, including KNN, SVD, PMF, LSTM, CNN, NCF, and LightGCN, as confirmed by k-fold cross-validation and statistical validation using paired ttests.

Fields of Study & Subjects

Recommender Systems and Techniques Caching and Content Delivery IoT and Edge/Fog Computing

Publication details

Journal / source
International journal of intelligent engineering and systems
Volume
19
Issue
4
Pages
861–879
Publication date
23 March 2026
Language
EN