Given the parameters you site automatic learning and automatic Bayes may
be what you need, although I do tend to recommend automatic learning be
turned off. My situation is not your situation. {^_-} I would widen the
thresholds for automatically learned ham and spam. And this begs for the
problems involved with automatically learning false scores. With 4000
users it's a little hard to cycle through individual IMAP ham and spam
folders and learn from them. Although if you scheduled the task every
2 seconds or so you'd get through 4320 accounts per day. Keep an index
into the accounts file (passwd or other) and swing around learning ham
and spam and purging the learned messages to avoid "toxic buildup".
The main thrust of my comments would be to spend some time up front at
the SpamAssassin Rules Emporium, http://www.rulesemporium.com/. Select
the rules tab. Read the description of each rule set, determine how
fussy you want to be. Then follow the RulesDuJour link and get that to
keep rules updated: http://www.exit0.us/index.php?pagename=RulesDuJour.
This will take maximum advantage of SpamAssassin's dual nature as a
Bayes tool and a rule based tool. But do select carefully depending
on the interest shape of your client list. A company that sells goods
would not want to be as strict about "generic marketing" spam as other
sites might.
And do use the blacklists in a scored mode. The default scores for SA
work pretty well. Some BLs are more aggressive than others. And they thus
tend to make more mistakes. So they get scored lower. But sometimes the
slight kick of a SpamCop rule may be enough to properly mark some spam.
{^_^} Joanne
----- Original Message -----
From: "Ryan Kather" <[EMAIL PROTECTED]>
To: <users@spamassassin.apache.org>
Sent: Thursday, March 30, 2006 14:25
Subject: Best Practices: SpamAssassin
I am about to evaluate SpamAssassin as a replacement in my environment for our present
spam solution (Symantec Mail Security for SMTP without the BrightMail add-on).
I wish to compare SpamAssassin's performance directly with DSPAM, Brightmail, and a
Barracuda Spam Filtering Appliance. I also intend to publish my findings and test
configurations to help other people make a decision.
So I'm writing to ask if anyone would like to provide some insight into the best practices
for making SpamAssassin as effective as possible.
