Mike McCarty wrote:
[about the penny-stock image spams]
Yes, I get several a day myself. The actual "text" of the message is
often actually an image, while the body of the message is randomly
selected sentences or words from a collection which would make a
Bayesian filter delete most of my e-mail.
I delete them by hand, to prevent messing up my other filter which
is working reasonably well ...(snip)
I use a statistical filter (aka "bayesian") that seems to do a
reasonable job. I block most of the spam I get directly by other means,
but I submit some of the stock-peddling image spams I get through
mailing-lists to the filter. Since I began running the filter I've
received 30453 non-spam messages, caught 882 spams and had 36 false
positives, and it has been ages since the last false positive. This is
while submitting 202 mails for training puposes., that were missed by
the filter. Like I said, the spams I want to catch is just from a couple
of mailinglists that have been particuliarly badly hit, hence the low
spam ratio.
The filter I use is known as dspam <http://dspam.nuclearelephant.com/>.
I believe you might be pleasantly surprised by the other implementations
aswell. The filter has more than the message text to work with.
Everything from the MIME headers to the message headers contribute in
characterizing the mail, so even though the text is random dspam will
catch the spam.
--
Håkon Alstadheim
