Re: automating sa-learn via cyrus mailbox?

[Christoph Wegscheider <wegi@despammed.com>]

Will Trillich <will@serensoft.com> wrote:
> so why not create a "user.spam" cyrus mailbox, BOUNCE any spams
> there and have cron do some sort of automated "sa-learn --spam"
> on the results, and then delete them?
My approach is to save all spam identified by spamassassin in a SPAM
folder on a per user basis. The spam folder is scanned every night with
--spam and all other folders of that user with --ham. If there is a
false-positiv and the user manually moves it to one of the non-SPAM
folders the next night sa-learn identifies the message as already
processed spam and reclassifies it as ham. I use this for ~8 month now
and have had ~3 false-positives in that time (in the beginning only). Of
course a more conservative setup is also possible where all mails
arrives in the inbox and the SPAM folder is fillied manually by the
user. 

The big advantage of a per user bayes filter is that it is trained on
the specific user rather than the average of all users, reducing the
danger of false-positives. 



	Christoph