spam filters

[Jason Stephenson]

Kevin D. Clark wrote:
[snip]
> 
> However, I recall that I encountered two problems when I attempted to
> use SpamAssassin's spamd/spamc combination:
> 
>    o  my databases for spam/ham wordlists seemed to get misplaced, or,
>       at least, not consulted when spamc was processing incoming
>       mail. 

Spamassassin looks in the home directory of the user it is running as 
for the bayes database. It is really designed to be called from a 
procmail filter running as the user whose mail is being scanned. 
However, most setups configure it to run as user nobody or user mail. If 
the user that spamd is running as doesn't have a home directory and the 
bayes database isn't there, then bayesian filtering is more or less 
deactivated and you're filtering on the default rules.

I got around this on both my work and home installations by creating a 
home directory for user nobody and putting the bayes files there. I made 
a symlink from them to root's home directory so that root can add 
mis-scored mails to either the ham or spam database.

I have to say that when a specific spamassassin installation is trained 
on a single user's email, and it is given sufficient examples of both 
spam and ham, it does an amazing job. In a month, I've had only 3 
messages score less than 5 that were spam and no false positives on the 
installation that just handles my personal mail.

If you can set it up so that your users call spamassassin through 
procmail, you'll have a great deal more accuracy. However, this is 
difficult if not impossible for most real world network environments.



> 
>    o  my use of Vipul's Razor seemed to get munged somehow (identity
>       problems, I think)
> 

This, I can't help you with. I'm not using Vipul's Razor at work, and at 
home I installed from ports on FreeBSD and Vipul's Razor just plain 
works for me.