Discrete deep learning based collaborative filtering approach for cold start problem

Abstract

The recommendation system plays a major role in e-commerce services and online applications such as social-networking, product recommendation, etc. The recommendation system uses the most popular techniques namely Collaborative Filtering (CF) and Deep Learning Neural Network (DLNN) approach. The Cold-Start (CS) problem and the recommendation efficiency are considered as crucial challenges and affected the efficiency of the most popular techniques. In this paper, a Discrete Deep Learning (DDL) based on hashing framework is implemented to achieve Hamming space for mapping Users and Items (UIs). Hamming distance used here to calculate the user's preference for an item, and the efficiency of online recommendation significantly improved by this computation technique. The performance of DDL framework evaluated by conducting various experiments on the large Netflix rating dataset. The experimental outcomes showed that online recommendation efficiency and CS recommendation accuracy of DDL outperformed the existing methods.