Recommender Systems for Movie Recommendations

Recommender systems are becoming a large and important market, with commerce moving to the internet and the ability to keep a larger stock of products, one of the biggest hurdles is to organize and show the right product to the right customer. Recommender systems aim at tailoring their products based on their customer need, by predicting how much a user would like a particular product. The recommender systems implemented in this project are within Collaborative filtering (CF) and Content-based filtering (CBF), with a final hybrid system based on combining the systems of CF and CBF. The aim is to evaluate how features such as number of latent factors, regularization factor and learning rate affect prediction accuracy for CF using Matrix factorization and compare the Root-mean square error (RMSE) for the three different systems.Collaborative filtering using matrix factorization resulted in lower RMSE than CBF and the largest factor in lowering error was learning rate. The results did indicate that CBF might perform better than CF when the user-base is small, while also having possibility of somewhat different functionality by recommending products which themselves are similar. The Hybrid recommender system had the lowest RMSE but with insignificant improvements from that of the CF method