[Spambayes] Dealing with devious spam

[Stephan Hoyer]

Alright, I've attached the latest one of these spam messages I have received
along with the SpamBayes report which shows the amount of spam/ham I have
(881/2828), the full message including headers and the values for individual
tokens.

Stephan

-----Original Message-----
From: Skip Montanaro [mailto:skip at pobox.com] 
Sent: Saturday, June 18, 2005 4:02 PM
To: Stephan Hoyer
Cc: spambayes at python.org
Subject: RE: [Spambayes] Dealing with devious spam


(Best to keep the mailing list cc'd for such discussions.  You'll get a lot
more brain power applied to your problem.)

    Stephan> The only thing is, I've already trained it on a dozen of these
    Stephan> messages. I think it's because it has too much other junk stuff
    Stephan> that looks like a real message. SpamBayes isn't picking them
    Stephan> up, and that's worrying me.

Let's do a little diagnosis.  Do you have the individual ham/spam scores
from one of these messages?  Have you saved one of these messages?  If so,
can you send it to the list?  (Save it to a file, zip the file, then attach
it to a message.  SB ignores binary attachments, so should be little chance
that it will wind up in anyone's spambox.)  What are the properties of your
current training database?  How many ham?  How many spam?

Once we have that sort of information, we can begin to make an educated
guess at why SB isn't scoring these messages properly for you.

Skip
-------------- next part --------------
A non-text attachment was scrubbed...
Name: spam clues.zip
Type: application/x-zip-compressed
Size: 6520 bytes
Desc: not available
Url : http://mail.python.org/pipermail/spambayes/attachments/20050619/455df34b/spamclues.bin