Collaborative Filtering Matches Decision Templates: A Practical Approach to Estimate Predictions

• Guilherme Brandão Martins UFSCar
• João Paulo Papa UNESP

Resumo

Collaborative Filtering stands as an underlying strategy to reasonably deal with large-scale problems like scalability and high sparsity. In the classifier fusion context, one could benefit from adopting such a strategy to learn decision templates effectively for the sake of computation efficiency. This paper introduces a framework that explores collaborative filtering-based latent factors models for fast decision template generation, assuming it has a sparse matrix structure. Experiments conducted over five general-purpose public datasets and statistically assessed have demonstrated its feasibility for building decision templates under low sparsity conditions and datasets labeled with fewer classes. Under such conditions, the proposed framework showed competitive recognition rates, significantly reducing computational costs, particularly when distance-based classifiers are employed for ensemble learning purposes.

Palavras-chave: Training, Graphics, Computational modeling, Collaborative filtering, Scalability, Buildings, Distributed databases