Deep Learning Based Collaborative Filtering Recommendation System

Recommendation systems are automated tools and techniques that facilitate and expedite decision-making by compiling opinions from many sources and directing them to appropriate users. Nearly everyone’s daily life is impacted by recommendation systems, which are widely employed in various fields like entertainment, social networking, e-commerce, books, etc. It plays a vital role in information retrieval by addressing scalability, sparsity, and cold start problems. This paper presents a deep learning-based collaborative filtering approach utilizing a Restricted Boltzmann Machine (RBM) for the dataset Movielens-1M. The methodology includes data preprocessing, partitioning into training and test sets, and applying the k-Nearest Neighbor (kNN) algorithm to calculate user similarity. The RBM was trained and evaluated using 5-fold cross-validation. Results show a significant improvement over traditional collaborative filtering methods, achieving Root Mean Squared Error (RMSE) of 0.8736 and Mean Absolute Error (MAE) of 0.6854. These findings highlight the potential of deep learning to improve the accuracy and efficiency of recommendation systems.