Dirichlet Matrix Factorization: A Reliable Classification-Based Recommender System

Traditionally, recommender systems have been approached as regression models aiming to predict the score that a user would give particular item. In this work, we propose system tackles problem classification task instead of regression. The new model, Dirichlet Matrix Factorization (DirMF), provides not only prediction but also its reliability, hence achieving better balance between quality and quantity predictions (i.e., reducing error by limiting model’s coverage). experimental results conducted show proposed model outperforms other due ability discard unreliable predictions. Compared our previous which uses same approach, DirMF shows similar efficiency, outperforming former on some datasets included in setup.