-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello Mark,

Tuesday, June 10, 2003, 8:57:51 AM, you wrote:

MJ> I understand that spams that aren't caught should be learned has
MJ> --spam.  

MJ> But should false positives be considered --ham or can any e-mail
MJ> from my inbox be considered --ham?

I'm by no means an expert (not even allowed to create my own rules, since
I'm "just" an end user on the server, and if anyone feels what I'm doing
is wrong, please do let me know), but

1) Yes, spams that aren't caught should be learned as --spam.

2) I feed as --spam even those spams which are caught, but which
a) aren't auto-learned,
b) have a score just barely above my required threshold of 9 (I usually
limit this to those with scores between 9 and 10), and
c) have a bayes percentage below 70%.
I figure that if the bayes system doesn't yet have confidence in this
spam, I'll give it more confidence, so a later email, which might not
score as highly with more normal rules, can also get flagged as spam. 3)
Yes, any/all false positives are fed back into the system as --ham 4)
Yes, those ham email which come through my system with a significant
positive score are also fed back into the system as --ham. 

I have increased the "required" threshold to 9, and increased the bayes
score for 80% to 5 and 90% to 7 and 99% to 9 (can you tell I like
bayes?).

If ham reaches my system with a score above 4 or 5, then there's
something that sender is doing which matches various SA spam rules,
and I have two options: I can either modify the score of those rules, or
I can have bayes reverse that impact. At this time, bayes seems to me to
be the more productive method.

I don't bother doing this with ham that scores under 3 or 4, since it's
far enough away from my required threshold that I don't have to worry
about similar emails being falsely seen as spam.

Bob Menschel

-----BEGIN PGP SIGNATURE-----
Version: PGP 8.0

iQA/AwUBPuaNX5ebK8E4qh1HEQKRigCgwyApTBIlb8ylHck1TF+99qSKD9EAoMwy
MpGyFlQ4agIEwqaxl5fRgML+
=6+dg
-----END PGP SIGNATURE-----




-------------------------------------------------------
This SF.net email is sponsored by:  Etnus, makers of TotalView, The best
thread debugger on the planet. Designed with thread debugging features
you've never dreamed of, try TotalView 6 free at www.etnus.com.
_______________________________________________
Spamassassin-talk mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/spamassassin-talk

Reply via email to