On Mon, 21 Apr 1997, James L. McGill wrote:
With all this talk about spam filters, I'm surprised nobody
has come up with a general purpose filter that uses AI principles
to detect a sales pitch or other offensiveness in the body of the
message.
Actually, I did something like that. I had a file of fairly
predictable phrases in spam. Each phrase was put in regex format and
the body was egreped with those regexs. It worked pretty good, too.
But I abandoned it to make more room after I just started matching for
things like mail not to me or my mailing lists, mail from domains not
named in the Received, and other more profitable clues. I actually
_subscribed_ to subscribe(_at_)cyberpromo(_dot_)com (lets see how many filters
trip on that phrase in this body) and haven't had anything squeak
through for some time now. I get about 10 spams a day. A lot of that
is because I have several usa.net addresses.
And I have concluded that sending auto-responses is a waste of time
and bandwidth. I don't even set exitcodes anymore, because every time
my address bounces because I set an exitcode, the bounce notice winds
up in my sysops mailbox. I just put them in the /tmp dir and log the
From: and To: so I have a chance to fish back out anything I want to
look at. /tmp gets emptied every few days so I don't have to concern
myself with disk quota.
David