[vox-tech] SpamAssassin training

[Ken Bloom]

SpamAssassin includes a naive bayesian classifier that can be used to 
recognize spam based on keywords (in a probabilistically trained way). 
The results of classification using the bayesian classifier are boiled 
down into one of several rules: BAYES_00, BAYES_05, BAYES_20, ..., 
BAYES_95, BAYES_99. These rules have statically assigned scores. 
Combined with a whole pelathora of other more complex rules (for things 
like header bugs, DNSBLs, body formatting, etc...) the scores for any 
rules a message triggers are added up and used to determine whether a 
message is actually spam.

The scores for these rules can be customized manually in 
~/.spamassain/user_prefs or systemwide in files in /etc/spamassassin.

Is there any utility for spamassassin that could be used to train the 
scores for all of its rules automatically, in a bayesian or 
support-vector-machine kind of way? Note that I'm not talking about 
training the bayesian filter, as I just explained, I'm curious about 
automatically training the step that comes after the bayesian filter.

--Ken Bloom

-- 
Ken Bloom. PhD candidate. Linguistic Cognition Laboratory.
Department of Computer Science. Illinois Institute of Technology.
http://www.iit.edu/~kbloom1/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.lugod.org/pipermail/vox-tech/attachments/20070101/c1d795cf/attachment.pgp