Leveraging Leiden communities for enhanced collaborative filtering with matrix factorization techniques

Personalized recommendation systems are essential for providing customized content to users, significantly enhancing their experiences across various platforms. Their main objective is to offer precise recommendations aligned with user preferences. Collaborative filtering techniques, driven by matrix decomposition methods, have demonstrated their utility in practical applications. However, as datasets become increasingly voluminous and intricate, challenges related to scalability and effective recommendation arise. To tackle these obstacles, this study proposes a novel strategy combining matrix factorization with Leiden community detection to enhance the scalability and quality of recommendations. The proposed methodology comprises as outlined below: (1) representing the rating dataset as a bipartite graph, (2) identifying communities within the graph, (3) isolating community-specific rating matrices and concurrently applying matrix factorization methods, and (4) consolidating the predicted rating matrices and assessing performance using performance indicators. The investigation employs various matrix decomposition approaches, including MF, SVD++, and FANMF, in conjunction with the Leiden algorithm for communities. Trials were conducted on diverse datasets to validate the effectiveness of the proposed approach, yielding the significant improvements in effective recommendation with notably reduced error values in terms of RMSE, MSE, and MAE. This proposed methodology not only improves the effectiveness of recommendations but also boosts computational efficiency, making it a robust solution for scaling recommendation systems in practical scenarios.