For the problem that AdaBoost works poorly with Naive Bayesian (NB) categorization, an improved re-weighting
rule for training examples is proposed. It is not only considering the classification result of the current iteration, but also
considering how much those previous classifiers disagree on their decision-making for each training sample. Moreover, the confidence of every base classifier is determined not only by the error rate, but also by the diversity among base classifiers. Thus, the accuracy and instability of the base NB classifiers are increased. Experimental results show that BoostVE-NB is effectively to improve the performance of NB text categorization.