Comparative Performance of Collaborative Filtering Recommendations Methods for Explaining Recommendations

Abstract

In this paper a comparative performance of some collaborative filtering methods for recommender systems is presented. The literature review focuses on identifying some models for explaining of recommendations. Methods discussed here are those based on probabilistic models and based on biclustering techniques. In addition, we analyze a memory-based method and three recommendation probabilistic models, in order to know theirs performance and like these work. Experiments carried out using Netflix dataset presented good results for model-based methods compared to memory-based method, achieving the best performance using the Normalized Discounted Cumulative Gain (nDCG) quality measure and also in the recommendation and prediction accuracy.