Comparison of collaborative deep learning and nonnegative matrix factorization for recommender systems
Mine Oggretir, Ali Taylan Cemgil · 2017
Collaborative filtering and content-based methods are two main approaches for recommender systems, and hybrid models use advantages of both. In this paper, we made a comparison of a hybrid model, which uses Bayesian Staked Denoising Autoencoders for content learning, and a collaborative filtering method, Bayesian Nonnegative Matrix Factorisation. It is shown that the tightly coupled hybrid model, Collaborative Deep Learning, gave more successful results comparing to collaborative filtering methods.