Najib Abi Fadel wrote: > Hi all, > > since it is hard for a person to teach the spamassassin which mails > are spam and which are not for all users, i was thinking about doing > the following: > > Create a mail account let's say : [EMAIL PROTECTED] > Tell trusted users to forward the mails they identify as spam to > "[EMAIL PROTECTED]" . > Run a cron job that launch "sa-learn" on the mailbox > "[EMAIL PROTECTED]", making the spamassassin identify them as spam. > > Since i don't really know how Bayesian filter works internally; i have > some concerns: In general, forwards are pretty useless for bayes training..
Try forwarding a multipart/alternative spam message with dissimilar text/plain and text/html sections and look at what really gets forwarded. All the headers are new, the originals are discarded. That's bad. Some of the body has been discarded. The rest of the body has been modified, reformatted and re-encoded differently. The two messages aren't the same. Sometimes you can do a "redirect" "bounce" or "forward as attachment" and end up with something reasonable, but plain forward nearly always destroys the message.
