Collaborative Filtering using Deep Learning

Abstract

A recommender system is a subclass of information or data filtering system that tries to predict the “rating” or “preference” that a visitor or client might want to provide for a particular thing. Recommender systems have turned out to be progressively famous now a days, and are used in a various areas including motion pictures(films), music, news, books, articles, search queries, social tags, and items in general. Recommender system assumes an indispensable part in web innovation for information assembling and rating up an information. Step by step guidelines to take full benefit of these omnipresent information is turning into the fundamental part of a recommender framework. The well-known methodology utilized for obtaining suggestions or recommendations is collaborative filtering. Recommender systems based on collaborative filtering, predicts user's tastes for the items or web-services by learning from the past user-item relationships from a group of clients who share the same taste. Although this method is very popular, it has certain limitations and there is a scope of reducing prediction error. In this dissertation, we have attempted to address this limitation with the help of a deep learning technique. We have implemented user-based collaborative filtering using Restricted Boltzmann Machine - a deep learning technique. The results obtained are encouraging when compared with standard user-based collaborative filtering. All the experiments have been carried out on a well-known publically available dataset MovieLens.