VAEMRS: Variational Autoencoder Based Movie Recommender System

Nowadays, deep learning is an emerging technique used in many research domains.The recommender system, specifically collaborative filtering, has significantly improved its performance by deploying this technique.Neural collaborative networks and their related neural network models are the bench-mark models in this domain.However, these models do not exhibit to create a continuous, robust, and structured latent space like autoencoder.On the other hand, autoencoder does not perform well in sparse data like Movielens.This article proposes a variational autoencoder based movie recommendation system (VAEMRS) to handle the above issues.In our proposed model, we consider implicit data like click vectors, normalize the interaction matrix, and pass them to the dropout layer to learn the VAE.Further, our approach applies variational concepts in neural networks.Also, use multinomial likelihood and Bayesian inference for parameter estimation.The proposed model has been tested using different quality measures on open-source datasets such as Movielens and compared with baselines.The performance results of the proposed work show the superiority over the baselines.