Searchable Mail Archives

[skip at pobox.com]

    Stephen> What I'd rather have, if you would, is advice on integrating
    Stephen> SpamBayes into the Mailman pipeline.  

I can find out how this is set up on mail.python.org.  That was what
SpamBayes was originally designed for.  I imagine it's a fairly normal
insertion of SpamBayes into the chain of email aliases known to the MTA.

    Stephen> Would it make sense to call SpamBayes from procmail?

Sure.  That's exactly how I do it.

    Stephen> How does one train SpamBayes?

There are various ways, but basically you have a mailbox of known good mail
and another mailbox of known bad mail and run the appropriate training
program to generate the database it uses at runtime.  When you add new mails
to either mailbox you retrain.  This requires physical access to the machine
on which SpamBayes runs.  If that's not possible you can always have
xemacs-beta forward to some machine to which you do have access, run
SpamBayes from procmail there, then at the end of .procmailrc forware to
xemacs-beta-internal at xemacs.org where normal processing picks up again.  You
then configure Mailman to hold any mail for moderator approval which
SpamBayes marks as "unsure" or "spam".

Skip