Deep Non-Negative Matrix Factorization for Recommender Systems

Abstract

A recommender system is a specific form of information filtering that helps users find items from enormous catalogs to present them based on their preferences and behavior. Items can be books, films, news, and music. Selecting the right approach has gotten difficult due to the abundance of available methods. When a recommender system performs better than other approaches, it can be compared and found to be a valuable tool. This work uses deep learning approach to implement collaborative recommendation algorithms for online resources (MFDL Model). This model is based on the matrix factorization. The Matrix Factorization with Deep Learning contributes to deep learning collaborative recommender systems by effectively bridging the gap between traditional matrix factorization (MF) and the expressive power of deep neural networks (DNNs). To compare the proposed model, we have implemented, evaluated, and compared two machine learning approaches: Nearest Neighbors (KNN) and Non-Negative Matrix Factorization (NMF). The metrics evaluation allowed us to deduce that the MFDL model present good results for the dataset used.