Introduction:

Recommender systems are focused on helping users to find the information that best fits their preferences and needs, in a search space overloaded with possible options. Most of the research works in this area are focused on proposing new recommendation approaches that improve the accuracy of previous works.

Methods:

Based on this idea, the objective of this research is the development of new data preprocessing methods for removing natural noise in collaborative filtering recommender systems, in order to improve the accuracy of the generated recommendations. Several of the approaches proposed in this research are based on fuzzy logic for managing the uncertainty associated with the user’s preferences as a relevant part of recommender systems. As a result, four new approaches for data preprocessing methods in collaborative recommender systems were obtained, both for individual and group recommendation.

Results:

The methods proposed were evaluated using international databases created for this purpose, thus verifying their accuracy. Furthermore, these methods were applied in a real recommendation situation associated with Programming Online Judges in Cuba, evidencing an improvement in the recommendation accuracy in such context. These results were published in 8 Group I papers, in Web of Science Journals such as Knowledge-Based Systems and Decision Support Systems. Conclusions-The methods proposed then can be integrated into any collaborative filtering recommendation system, in order to improve its accuracy. Such methods have been clearly applied to situations such as programming online judges.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )