Collaborative Filtering for Personalized Recommendations: A Comparative Analysis of NMF and CoClustering

Abstract

Collaborative filtering is central to personalized recommendation systems, especially in environments with limited data. In this research, we perform a comparative analysis of two matrix-based collaborative filtering methods: Non-negative Matrix Factorization (NMF) and Co-Clustering. For assessing performance, both algorithms underwent testing using a standard 75/25 train-test division, followed by 5-fold cross-validation to ensure robustness. NMF reported a marginally reduced RMSE  and MAE  compared to CoClustering  when evaluated on the test set.

The effectiveness of Top-N recommendations was evaluated through the use of Precision@5 and Recall@5 metrics. NMF demonstrated better precision and recall than CoClustering , suggesting that its top recommendations are more relevant.

Our results indicate that although both models are computationally efficient and suitable for small-scale recommendation tasks, NMF reliably surpasses CoClustering in predictive accuracy and top-N relevance.  Future research might involve integrating hybrid frameworks or real-world implicit feedback datasets to investigate wider generalizability.