Didactic Evaluation of Some Useful Predictive Methods in Neighborhood-Based Recommendation Systems

Abstract

This paper presents a didactic-experimental study of three ratings prediction techniques that have been widely used in neighborhood-based Recommender Systems. The process of calculating the predictions and estimating the error for academic purposes is exemplified. The results of experimentation based on a MovieLens Dataset are presented. The objective was to measure the quality of the recommendations generated by the prediction techniques studied, with respect to the density of the data. As a measure of similarity, to generate the neighborhood, the Pearson correlation coefficient was used. To measure the quality of the recommendations, the metrics were used: mean absolute error and mean square error. The results show that the best predictive technique is what we have called F3, and that as the density of the data decreases, the error in the predictions increases.