I've been trying an experiment where I'm using another bayesian filter (spamprobe) and feeding it only tokens which are the rule names and nothing more. The idea was that bayes could basically self score.

Well - it's not working - and I haven't figured out why yet.

But - I still think that self tuning scores is a good idea. It seems to me that the system can keep track of the rules it gets wrong - the ham/spam scores on each rules - and eventually self adjust the scores to match reality.

Perhaps we can still use the scores we calculate with mass checks as a "base" score and then after enough messages are processed start gradually changing the base score some as the totals are accumulated. And perhaps report back to SA rules that are inconsistent with reality.

--
Marc Perkel - [EMAIL PROTECTED]

Spam Filter: http://www.junkemailfilter.com
   My Blog: http://marc.perkel.com
My Religion: http://www.churchofreality.org
~ "If it's real - we believe in it!" ~




Reply via email to