On Thu, Aug 14, 2008 at 8:24 PM, David C. Rankin
John Andersen wrote:
On Thu, Aug 14, 2008 at 6:04 PM, Patrick Shanahan
wrote: * David C. Rankin
[08-14-08 18:21]: Spamassassin has decided that several of my list serve messages are spam. How do I tell spamassassin to unlearn the messages as spam so that they are not-spam?
man sa-learn and/or whitelist in ~/.spamassassin/user_prefs
--
First, I don' think this IS off topic....
Second, David, this will be a reoccurring situation over time. All your users will find false negatives and false positives and you might as well figure out some solution that is easily implemented across all users.
We set up two mailboxes on each Imap account (this was one of the issues that got us to Imap in the first place), labled "Missed-Spam" and "Not-Spam" (or what ever).
We also have a mailbox called just plain "Spam". Our system wide sieve scripts puts spam there.
All our users have to do is move / copy messages that spamassassin falsely tags into the Missed or Not mail boxes.
A script on the server then runs sa-learn against those mailboxes every so often. Sa-learn will also do razor-report so you help the community if you do automate this.
The script also purges these Missed and Not boxes for each user so that they don't have to manage them.
Its worked quite well for us, keeping our SA bayes database up to date, unlearning falsely accused spam (and razor revoke as well).
Thanks John,
That's the practical howto I was looking for. We run a similar setup, but have never implemented an unlearn function. I like the default:
~/Mail .. spam spam-learn spam-probably ..
setup with suspected spam going to spam-probably, user copy to spam-learn, then an hourly cron job running to have spamassassin learn the messages in spam-learn and then delete them. I think I'll add "spam-unlearn" to the end of the list and add the unlearn features to the cron job.
-- David C. Rankin, J.D., P.E.
Yes, I can see that we are on the same track there. With opensuse's default setup for Imap (postfix -->amavisd (vscan, spam) -->posfix-again -->cyrus there is only ONE systemwide bayes database. And you have to run every body's imap folder against that database to Learn, unlearn, etc. If you take amavisd out of the picture each user has their own bays database. This is less than Ideal, but I spent a good 8 months training personal databases with spam/ham before I had that forehead smacking "DOH!" moment when I realized amavis only uses a system wide bayes. If you are using Imap I can share my cron script with you for comparison if you wish. Note that I have no "probably" category. I have my spamassassin dev-nul level set around 10, and anything above that just gets dumped by Amavisd. Anything below that gets flagged as spam unless its below my spam threshhold, in which case it passes through. So my Spam category is your Probably category. And your Spam category is my Dev nul. I can afford to do this in my line of work, YMMV. -- ----------JSA--------- Someone stole my tag line, so now I have this rental. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org