Bogofilter w/ SpamAssassin

[Devin Nate]

Hi Folks;

I'm a recent bogofilter user, inspired to give it a try after reading a 
good article comparing bogofilter with other systems. Prior to that, I 
had started work integrating another bayesian system (CRM114) with 
SpamAssassin. Since I liked bogofilter so much, I added a similar 
interface for Bogofilter. Since I see that the FAQ has a section on 
training bogofilter from SpamAssassin, I thought this experimental 
feature may be of interest to some people here since it adds this 
capability right into SA, plus giving points against messages that 
bogofilter thinks are spams. The URL with patches and documentation is 
available here:

http://bugzilla.spamassassin.org/show_bug.cgi?id=2301

As a basic run down, SA uses a near identical 'autolearning' feature as 
already built into it. The autolearning system will pass emails into 
bogofilter that it considers to be ham or spam based on an autolearning 
set of thresholds. The manual training via sa-learn is not done yet. SA 
keeps track of how many hams/spams learned, and individual message id's 
seen so that it won't relearn a message twice. SA also will perform 
train-on-error automatically by default (but can be turned off), once 
Bogofilter has seen enough to start classifying on its own. In addition 
to all of this, I've run a number of statistical tests on a corpus of 
about 2700 hams and 1400 spams; bogofilter did excellent. Excellent job!

The patch includes CRM114 stuff too- both CRM114 and Bogofilter can be 
dealt with separately, so if you only want to try bogofilter that's 
fine. It is a first revision of the patch, highly experimental, etc., so 
it's probably best that people familiar with bogofilter and SA use it 
and not on a mission critical server. Feed back is welcome!

Thanks,
Devin Nate

-- 

____________________________________________________________

Devin Nate
Chief Consultant & General Manager
BridgeComm Corporation
http://www.bridgecomm.net/
mailto:devin.nate at bridgecomm.net
____________________________________________________________