ClusAnnRS: A Novel Dual-Layered Collaborative Filtering ApproachBased on Clustering and Neural Networks to Tackle Data Sparsity andEnhance Recommendations

Abstract

Introduction: Collaborative Filtering (CF) is a cornerstone technique in Recommender Systems (RSs), widely adopted due to the increasing availability of user–item interaction data and advances in machine learning. However, data sparsity remains a persistent challenge, limiting the system’s ability to accurately capture user preferences and reducing recommendation performance. Methods: This study proposes ClusAnnRS, a hybrid model that combines clustering and artificial neural networks to address the data sparsity problem in CF. The model first segments users and items using clustering techniques, followed by training a neural network to predict ratings based on cluster-specific interactions. The proposed approach is evaluated on two benchmark datasets from the movie and book domains Results: ClusAnnRS was evaluated across eight experimental scenarios using evaluation metrics such as precision, recall, accuracy, F1-score, and ROC curves. The results consistently demonstrate that ClusAnnRS outperforms conventional CF methods in prediction accuracy and recommendation quality under sparse data conditions. Discussion: The experimental analysis highlights the effectiveness of integrating clustering with neural learning to capture latent patterns in user behavior, even with limited interaction data. This dual-layered approach enhances the model’s capacity to generalize and personalize recommendations. Conclusion: ClusAnnRS presents a robust solution to the data sparsity issue in CF recommender systems. By combining neural networks and clustering, the model improves both prediction accuracy and recommendation relevance, offering a valuable contribution to the advancement of intelligent, user-centric recommendation systems.