[Spambayes] Spam probabilitytest
Tim Peters
tim.one at comcast.net
Fri Oct 31 13:07:20 EST 2003
[Jens Rantil]
> Can someone explain the attached screenshot? How come that the
> probability for spam isn't 100%?
You mean because the word appeared in 6 training spam and no training ham?
Estimates of probabilities based on counting alone are never taken at face
value. Gary Robinson explains the Bayesian adjustment we use in detail
here:
http://www.linuxjournal.com/article.php?sid=6467
Your intuition matches what he calls p(w) in that article. We use an
adjusted value, called f(w) in the article.
More information about the Spambayes
mailing list