Deep Collaborative Filtering based Recommendation System

Abstract

A recommendation system seeks to make relevant suggestions to users based on their individual preferences. It primarily analyses existing data and identifies latent patterns within the same. Based upon these patterns user behaviour is predicted and thereby, suggestions are offered. In this paper, a collaborative filtering approach has been applied to a deep learning model to develop a recommendation system. An autoencoder was exercised as the fundamental building block in the construction of the model architecture. To achieve working functionality first, the model was trained on two datasets from movielens. Next, this model was compared against a Restricted Boltzmann Machine run model. The results were then compared and analysed.