Using the Bayes Utilities in the released V2, I start checking some of the msgs in Quarantine db because it seemed to me that a lot of stuff was getting through that should have been tagged as spam.

If I checked a message that was caught by a Keyword, it showed 100% or close to it every time.

If I checked any messages cause by MAPS, it ALWAYS showed 0%, every time.

This puzzled me because when I was building up my corpus (5000 good/bad) I was using the MAPS rejects to build up my corpus db (known spam) , or so I thought.

I checked about 30 more random MAPS rejected messages in the quarantine db and ALL showed 0%. Is this correct?? It appears that ANY messages caught using MAPS are not entered into the corpus the same way as those using a keyword or other blacklist are. Can anyone else confirm this behaviour on their system?

Thanks!
Andy

As an addon to my initial email.

I did further testing this mornring chooing random MAP's quarantined emails and did find a few that were 100%, but only a small number. I guess I need to concentrate on good keywords to help get the majority fo these obviously spam emails caught.

Andy

Hi,

The only way i see a lot of email getting caught by the bayesian filter, is too leave my keyword rules on, and send the result to quarantine, then reject them manually (time consumming) and then the bayesian filter is learning from what is rejected. I have almost 20000/20000 spam/good emails in my corpus database. And it start to reject more and more based on the bayesian filter, but i need to keep him learning from keywords, and keep on the quarantine cleaning.

From what i see, if you don't quarantine the emails, as soon has it sees a keyword, it drops the connection and the bayesian doesn't seem to learn from these dropped emails ? Maybe i am wrong on this one ?