An enhanced probabilistic neural network approach applied to text classification

Abstract

Text classification is still a quite difficult problem to be dealt with both by the academia and by the industrial areas. On the top of that, the importance of aggregating a set of related amount of text documents is steadily growing in importance these days. The presence of multi-labeled texts and great quantity of classes turn this problem even more challenging. In this article we present an enhanced version of Probabilistic Neural Network using centroids to tackle the multi-label classification problem. We carried out some experiments comparing our proposed classifier against the other well known classifiers in the literature which were specially designed to treat this type of problem. By the achieved results, we observed that our novel approach were superior to the other classifiers and faster than the Probabilistic Neural Network without the use of centroids. © 2009 Springer-Verlag Berlin Heidelberg.