Using hypothesis margin to boost centroid text classifier

Centroid Classifier is a simple and yet efficient method for text categorization. However it often suffers from the inductive bias or model misfit incurred by its assumption. In order to address this issue, training-set errors as well as training-set margins are regarded as training criterions. Based on these two criterions, an overall (or global) objective function over all training examples is constructed, and optimized to produce a refined Centroid classification model. The empirical assessment conducted on four benchmark collections evidence that proposed method performs comparably to state-of-the-art SVM classifier in classifying performance, as well as beats it in running time.