GrayHat wrote:
> 
> Hmm... that sounds like an idea which was brought on some 
> time ago (John was still the dev for ASSP at the time); that 
> is, set up some kind of TTL parameter for corpus files so 
> that the spamdb rebuild should check the file date/time and 
> if over the TTL (say "n" days) it should then delete the file.

My thought is that the "TTL" would only be in effect for the purpose of
keeping BlockReporting working (for however many days or weeks you wish
the emails to be guaranteed resendable).  After that time, the TTL is
null and the files are game for replacement.  I thought it a simple idea
for working around the BlockReporting problem Thomas mentioned.
 
> While at first it may sound like a "cool idea", it has some 
> drawbacks, especially when it comes to low and high traffic 
> boxes; in the first case the spam/notspam folder would 
> quickly "age" and get almost empty; in the second one is that 
> on a high traffic box it would then be easy to corrupt the 
> corpus by sending in a bunch of identical messages :P

I suppose for a high-traffic box that is able to overwrite its corpus in
a day or few this might not be the best idea.  But, Thomas has said that
we shouldn't rely on Bayesian as our first line of defense even though I
find it catches a lot of things that the other checks don't.

On a low-to-medium traffic box, though, this would not be a problem.  We
already deal with bunches of identical messages from time-to-time
(nothing new).
 
> Bottom line; the bayes filter should work by /learning/ this 
> means that it should NOT discard the previous data, but 
> rather REFINE them from further data coming in; so maybe the 
> whole bayes approach used inside ASSP should be revised NOT 
> to deal just with the latest data but to learn/improve during time

Just an idea, but how do you "NOT" discard data while keeping rebuild
times low and maintaining free hard drive space (realistically)?

Kind Regards,
Brett


------------------------------------------------------------------------------
Come build with us! The BlackBerry® Developer Conference in SF, CA
is the only developer event you need to attend this year. Jumpstart your
developing skills, take BlackBerry mobile applications to market and stay 
ahead of the curve. Join us from November 9-12, 2009. Register now!
http://p.sf.net/sfu/devconf
_______________________________________________
Assp-test mailing list
Assp-test@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/assp-test

Reply via email to