----- Original Message ----- 
From: "Brian Downey" <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Wednesday, January 28, 2004 11:56 AM
Subject: Re: [gentoo-user] SpamAssassin not as good as before :(

> Now four or five are slipping through a day.  The only thing I can suggest
> (and what I have been doing) is to feed the misses back through SA's
> bayesian engine.  It's a little time consuming but it helps SA "learn"
> what the spammers are up to.

I'm running courier-imap on my box and have spamassassin move any spam into
a spam folder (so I can make sure to check for false-positives). I have a
cron job set to run sa-learn every night that trains ham from my legit mail
folders and spam from the spam folder. This way, all I have to do if I get a
false-negative is just move it to the spam folder and sa-learn will catch it
that night. Using this technique, I've been able to keep my false-negative
rate very low (I get about 5000 spam a month and only see about 10 spam in
my inbox in a month, usually when I get more than 1 copy of a particular
spam on the same day).

I usually keep about 5000 messages in my spam folder and clear out the
oldest. That way it's always fresh spam that's being considered. I could
probably get away with fewer messages.

I will note that I'm also running some optional modules like razor, pyzor,
and dcc.

Andrew "frugal" Dacey
[EMAIL PROTECTED]
http://www.tildefrugal.net/


--
[EMAIL PROTECTED] mailing list

Reply via email to