Varying Naive Bayes Models with Applications to Classification of Chinese Text Documents