Creative Commons

A recommender system is a powerful tool that improves customer’s experience through personalized recommendations. In order to provide recommendations, it analyzes the user behavior, such as the ratings available and browsing history. A well-designed recommender system is able to significantly increase the revenue of e-commerce web sites and applications. There are many different models of recommender systems implemented. A successful recommender system must be accurate in its predictions and fast at the same time. A satisfied customer is very likely to be loyal to the web site or application. For that reason, the different models must be evaluated in terms of accuracy and performance. The goal of this dissertation is to implement and parallelize three popular collaborative filtering methods. The methods implemented are the user-based collaborative filtering with the Pearson correlation, the item-based collaborative filtering with the adjusted cosine correlation and the model-based alternating least squares method. These methods are also going to be evaluated for their accuracy and performance. For that reason, a set of different experimental metrics is used to evaluate their computing performance and their ability to provide accurate predictions