[Spambayes] Spambayes not working as well as it used to

[Kenny Pitt]

Nothing has changed in the latest version in terms of how SpamBayes decides
what is spam and what isn't.  However, if you just retrained then you have
changed the information that SpamBayes has available to base its decisions
on.
 
Training on "several thousand recent spam and good" messages is not
recommended.  Some detailed discussions of training strategies are available
on the SpamBayes wiki:
 
http://entrian.com/sbwiki/TrainingIdeas
 
One problem that we often see when people retrain SpamBayes on a large
number of existing messages is that they have far more spam messages than
they do good messages.  This produces an imbalance that can, particularly in
extreme cases, have a significant negative effect on accuracy.
 
If you could post the clues from a spam message that was left in your Inbox
(get the clues before training the message, please), that would be extremely
helpful in diagnosing your problem.
 
-- 
Kenny Pitt
 

  _____  

From: spambayes-bounces at python.org [mailto:spambayes-bounces at python.org] On
Behalf Of Brett Brewer
Sent: Thursday, December 16, 2004 12:19 PM
To: spambayes at python.org
Subject: [Spambayes] Spambayes not working as well as it used to


I just upgraded spambayes to the latest version (1.0.1) and rebuilt my spam
database using several thousand recent spam and good messages. I've
continued to train spambayes daily for several days, but I'm still getting
tons of spam in my inbox and more "spam suspects" than I used to. I think
something got screwed up in the latest version because it is not working as
well for me. I used to have to sort only 5-10 messages by hand each day, now
I have dozens and dozens that get through. Are there any known issues with
the latest build?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.python.org/pipermail/spambayes/attachments/20041216/dcd5a707/attachment.htm