Comparative Study of Collaborative Filtering Models - Matrix Factorization and Factorization Machine

Recommendation systems help business organizations in suggesting to their customers the most relevant and preferred items to purchase. For designing these recommendation systems industry analysts have to choose among various Collaborative Filtering (CF) models. Choosing the right CF model for a specific recommender system design based on historical data remains a challenge. In this study among the different models available, we have compared Matrix Factorization (MF) and Factorization Machine (FM) models of CF using two real-world datasets and experimentally evaluated based on optimized model parameters with various metrics. When using MF we also studied the effect of mean, item bias, and user bias terms. The quality of recommendation is measured using RMSE, Precision, recall, and F1 score as evaluation metrics. The results are plotted graphically providing insight into the quality of the two models. The comparative study will offer researchers and industry analysts an in-depth understanding of the two models, allowing them to make an informed decision for their specific application of their recommender system.