Would it be a decent idea to extend the recently added Naive Bayes Classifier to support the various document types that this package offers?