On Sun, Jun 22, 2003 at 10:08:07AM -0400, Matt Kettler wrote:
> At 08:30 PM 6/21/03 -0400, Gordon Cormack wrote:
> >Auto-learn and auto-whitelist use different scoring criteria from those
> >used in spamassassin's spam filtering.
> 
> The bayes auto-learning does not use "it's own" scoring mechanism, it uses 
> scoreset 0. This is the score the email would get by the main SA engine if 
> bayes and network checks were off.
> 
> Certianly you do NOT want the bayes scores to feed back into bayes 
> learning,  If you can't see why, think "feedback amplifier with positive 
> gain".

In supervised mode, positive feedback is exactly what you want.

For the reasons that I've mentioned before, the lack of feedback in the
current setup causes the system to 'learn' progressively less accurate
information.

The proof would, of course, be in a controlled experiment.  I may do this
some day by re-classifying my last 5000 messages with and without my
modification, but I can assure you that I haven't heard any screeching
noises eminating from my mailbox.  What I have observed is < 0.2% false
positives and < 1.0% false negatives.

-- 
Gordon V. Cormack     CS Dept, University of Waterloo, Canada N2L 3G1
[EMAIL PROTECTED]            http://cormack.uwaterloo.ca/cormack


-------------------------------------------------------
This SF.Net email is sponsored by: INetU
Attention Web Developers & Consultants: Become An INetU Hosting Partner.
Refer Dedicated Servers. We Manage Them. You Get 10% Monthly Commission!
INetU Dedicated Managed Hosting http://www.inetu.net/partner/index.php
_______________________________________________
Spamassassin-talk mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/spamassassin-talk

Reply via email to