Thalita Firmo Drumond and Fernando J. Von Zuben
intelligent data analysis, collaborative filtering, ensembles, random projections
Information technologies are each day more pervasive in society, with data collection reaching unprecedented levels. Developing new methods for efficiently analyzing large datasets is becoming mandatory. This scenario motivates the development of recommender systems, engines capable of recommending to an interested user or consumer the services or products he/she is most likely to enjoy. Matrix factorization models have been successfully used for collaborative filtering, but Boolean matrix factorization (BMF) has not been thoroughly explored in this context. In this paper, we further explore a previous proposal of BMF-based collaborative filtering, and propose an ensemble framework using multiple random projections in order to obtain more robust and accurate recommendation. Results show that this approach produces better recommendation lists while being less sensitive to some parameters.