Recommender Systems for Movies Using a Class of Neural Networks

Abstract: In this project a recommendation system for suggesting movies is implemented, in the field of Collaborative Filtering (CF). The system is created with a Restricted Boltzmann Machine (RBM), which is a two-layer neural network. The main tool used for programming the RBM is the TensorFlow library, imported to Python. The performance of the system is evaluated with Root Mean Squared Error (RMSE) where the error between the observed movie ratings and the predicted ratings is computed. This study shows how different parameters of the RBM, e.g. number of hidden units, mini-batch size, epochs and learning rate, affect the prediction error. The results show that parameter values within a specific range can generate good recommendation with low prediction error. The lowest RMSE, with optimal values for RBM’s parameters, is documented at 0.80, while the aim of this project is to reach a prediction error lower than 1.1. How to improve the accuracy of the model is discussed and the result is compared to previously done studies in the area of CF.