A Modular k-Nearest Neighbor Classification Method for Massively Parallel Text Categorization

This paper presents a Min-Max modular k-nearest neighbor (M-k-NN) classification method for massively parallel text categorization. The basic idea behind the method is to decompose a large-scale text categorization problem into a number of smaller two-class subproblems and combine all of the individual modular k-NN classifiers trained on the smaller two-class subproblems into an M-k-NN classifier. Our experiments in text categorization demonstrate that M-k-NN is much faster than conventional k-NN, and meanwhile the classification accuracy of M-k-NN is slightly better than that of the conventional k-NN. In practical, M-k-NN has intimate relationship with high order k-NN algorithm; therefore, in theoretical sense, the reliability of M-k-NN has been supported to some extend.