Community-Based Matrix Factorization (CBMF) Approach for Enhancing Quality of Recommendations

Area of Science:

• Data Science
• Network Analysis
• Recommender Systems

Background:

• Matrix factorization is a standard technique for extracting insights and recommendations from user rating networks.
• Large datasets present computational challenges for traditional matrix factorization methods.
• Community detection algorithms identify groups within complex networks.

Purpose of the Study:

• To introduce a novel framework, Community-Based Matrix Factorization (CBMF), that integrates community information to enhance matrix factorization for recommendation systems.
• To address the computational limitations of matrix factorization on large-scale rating networks.

Main Methods:

• Model user rating data as a bipartite network.
• Apply community detection (Louvain algorithm) to partition the network.
• Extract and process community-specific rating matrices in parallel using matrix factorization (MF) techniques (basic MF, SVD++, FANMF).
• Merge community predictions and evaluate performance using Root Mean Square Error (RMSE).

Main Results:

• CBMF significantly enhances recommendation quality across six diverse datasets.
• On the MovieLens 100K dataset, CBMF with SVD++ reduced RMSE from 1.26 to 0.21 by utilizing 25 communities.
• Similar RMSE reductions were observed for FilmTrust, Jester, Wikilens, Good Books, and Cell Phone datasets.

Conclusions:

• Community-Based Matrix Factorization offers a scalable and effective approach to improving recommendation systems.
• Integrating community structure into matrix factorization overcomes computational bottlenecks and boosts prediction accuracy.
• The CBMF framework demonstrates broad applicability and performance gains across various real-world datasets.