Training Naive-Bayesian Classifier using 4J classifier java library

I need to categorize input queries on mobile platform sent by users as SMS into ten p开发者_如何学编程re-defined categories such as sports, finance, health etc.

How do I begin and then go about training Bayesian Classifier, i.e. how do I prepare a training document, how extensive should it be, are there any assured directives that can be followed or do I just keep on modifying the training document till good results start showing up?