Re: training bayes and newsletters

I think you are missing that a particular newsletter is not
intrinsically ham or spam.  It is ham if the user has subscribed, and
spam if they have not affirmatively subscribed.

I have seen the very same content arrive at my mailserver for 2 users.
For one it is ham and the other it is spam.

Therefore the determination of ham/spam has to be done based upon
per-user data, and from that user's classification (or records of
subscription, in theory, but that's unrealistic).