Ivan82,

Deleting entries in the database has no effect on the bayesian filter. However, if a good email was incorrectly blocked by one of the filters, forcing its delivery will resubmit the email to the bayesian filter, tagging it as "good", and this will help train the bayesian filter in recognizing "good" tokens within that email, so as to lessen the chances of mistakes in the future.

Oh, and the bayesian filter just isn't kicking in..not even after 60,000+ mail attempts discarded, 8000+ mail forwarded, and 17000+ spam blocked...

Ivan,

Please note that the bayesian filter "learns" emails as they are quarantined. If you delete them from the quarantine, this has no effect on the bayesian learning process as they have already been processed. If however you force the delivery of those quarantined emails, you will be telling the bayesian filter "this spam that you blocked is not really spam, reprocess the emails and re-learn that the content is actually good". The bayesian filter will thus re-eaxmine the email, try to learn about the email contents so next time similar emails will be passed, and will then deliver the current email to the intended recipient.

RE: your statement " so I've removed the maps & subrl quarantine", we're not sure we understood. The more filters you have active, the more chances are that SpamFilter will block spam. Both the MAPS an SURBL filters are very effective, and will thus greatly help the bayesian filter in the learning process.

As for the Bayesian filter not kicking in, please note that the Bayesian filter is used as a last resort to check for spam, after all the other filters have had a chance to do so. Only if they all fail is the Bayesian filter used. As such, it will have mostly pre-screened emails to check, and will only tag a very small percentage of them.

As an example we provided a snapshot of our filter stats for 3 days worth of emails on the forum as follows:

94,828 IP found in MAPS search
74,161 IP address is from a blacklisted country
10,810 Invalid sender domain MX record
7,896 SPF Sender Policy Framework match
3,044 Keywords found in content
763 Exceeded maximum number of RCPT TO
526 Mail From and Mail To domains are equal
345 Statistical filter match
27 Mail From and Mail To are equal

According to the above, the Bayesian statistical filter on our own server only blocked 0.2% of the spam found by the other filters. However that is still 354 spam emails that were successfully blocked.


To answer your last posting, "if mail is being sent by unknownusername@mydomain.com to knownusername@mydomain.com", as you mentioned the SPF filter will take care of that. There is also another option that rejects emails if the "FROM" domain is the same as the "TO" domain. That will also cause a reject if the user is within your domain in the scenario you described.

Ivan82,

Yes, SpamFilter will still pass on the emails to the Bayesian learning engine, even if you disable the quarantining of emails for some filters.