Assume I use the Naive Bayes classification algorithm. My question is simple: can I rank the words according to their posterior probabilities? I want to have a measure of "importance" of the words in predicting my outcome.
The reason I am asking is that I think the individual probabilities are actually biased while, somehow, the product of the probabilities gives the right classification (this is why the naive bayes works so well). In other words, I am not sure looking at the individual word probabilities makes any sense.