jdow at earthlink.net
Tue Jun 27 01:09:51 UTC 2006
From: "Aaron Konstam" <akonstam at sbcglobal.net>
> On Mon, 2006-06-26 at 12:11 +0100, Paul Howarth wrote:
>> Aaron Konstam wrote:
>> > On Mon, 2006-06-26 at 10:29 +0100, Samatason Ltd wrote:
>> >> Hi
>> >> What's the best approach to filtering out spam at server mailbox level? I am
>> >> running FC5 and SendMail...
>> >> Best Regards
>> >> Chris
>> > There seems to be a sisagreement on this but I would try spamasasassin.
>> > The program is isntalled but you can find installation configuration
>> > instructions on the web.
>> > Since sendmail runs procmail on each messgae it receives you can use
>> > a .procmailrc script to run spamassassin and distribute your mail
>> > (including the spam found) in to appropriate mail boxes.
>> Running spamassassin using a milter has the advantage that you can
>> reject the spam during the SMTP transaction. So in the rare case where a
>> legitimate message is mis-identified as spam, the sender gets to know
>> that it wasn't delivered. You can't safely do that using procmail.
> Huh! I have being doing that for years byt having .procmailrc direct
> identified spam to a special file. Doesn't that do it or do I not
> understand you point?
Heh heh - I can think of three interesting ways right off the bat.
But then, I have become more familiar with procmail than I ever wanted
to. (And I still have not plumbed its scoring system.)
FWIW I deliver spam with the subject header rewritten to include an
indication that it scored as spam and the actual score. I filter spam
into a distinct <cough> OE </cough> folder. I can then sort the folder
for scores, laugh at the particularly high ones, train on low ones that
show low Bayes scores, and vet the low ones for ham or spam. It'd all
takes only a few seconds every time I do the sort if I didn't get
distracted by some of the fascinatingly silly things spammers will
do. (One subject recently for an all .gif spam was "(Spam?)". It scored
a healthy 21+ points thanks to the rules and the multiple BL hits.)
More information about the users