A comparison of several ensemble methods for text categorization

Text categorization (TC), as an important domain of machine learning, has many unique traits, such as huge number of features, serious redundant features, dataset imbalance, etc. In this paper the various ensemble methods of naive Bayes classifiers and SVM classifiers are experimentally compared on the TC tasks. Besides, a new type of classifiers, moderated asymmetric naive Bayes classifiers, is proposed. Its advantages over the conventional naive Bayes classifiers in performance and computational efficiency are demonstrated.