Fast and accurate evaluation of collaborative filtering recommendation algorithms

Collaborative filtering are recommender systems algorithms that provide personalized recommendations to users in various online environments such as movies, music, books, jokes and others. There are many such recommendation algorithms and, regarding experimental evaluations to find which algorithm per forms better a lengthy process needs to take place and the time required depends on the size of the dataset and the evaluation metrics used. In this paper we present a novel method that is based on a series of steps that include random subset se lections, ensemble learning and the use of well-known evaluation metrics Mean Absolute Error and Precision to identify, in a fast and accurate way, which algorithm performs the best for a given dataset. The proposed method has been experimentally evaluated using two publicly available datasets with the experimental results showing that the time required for the evaluation is significantly reduced, while the results are accurate when compared to a full evaluation cycle.