An Efficient Deep Learning Approach for Collaborative Filtering Recommender System

Abstract

Abstract Owing to the enormous growth in information over the past few decades, the world has become a global village. The recommendation system remains the most widely used type of commercial websites. The personalized recommender system is of paramount importance in modeling user’s preference on items based on their past interactions (e.g., ratings and clicks), known as collaborative filtering (CF) technique. Although CF is very important among the algorithms used in recommendation systems, it suffers some setbacks such as the sparsity of matrix ratings, scalability, and integrals nature of data. Several research studies have shown that the above-mentioned obstacle could be tackled with the help of matrix factorization (MF) techniques. In spite of the fact that the technique is likely to suffer from lack of some meaningful signals by using a low ranked approximation as well as lack of sparsity in times of denser singular vectors. Recently, deep learning techniques have proven to learn good representation in natural language processing, image classification, and so on. In this work, we propose a deep learning method of collaborative recommender systems (DLCRS). We have made a comparative study of the proposed method and existing methods. Experimental results demonstrate that our approach gives improved results compared to already existing methods. We empirically evaluate DLCRS on two famous datasets: 100K and 1M Movielens.