-----Original Message-----
From:
[EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Kami Razvan
Sent: Friday, February 20, 2004 7:14 AM
To: [EMAIL PROTECTED]
Subject: RE: [Declude.JunkMail]
Zombie Test
Matt:
What I like about Scott's new
feature (not his own but that of Declude :) ) - is it makes triggering the CPU
intensive filters (like our URL in body, etc.) less likely to be used.
A lot of the ones that are deleted
in our system hit our word filters and URL in body filters. If this test
can add enough weight so we skip those filters we will be in good shape..
Scott: Any chance for adding
a skip test if the weight is below a certain negative number. We have
negative weights and also with John's AutoWhite software our emails get weights
of -200+ at times depending how many times we have emailed the person.
Naturally these emails go through all filters since they do not exceed the
skipifweight setting..
Regards,
Kami
From:
[EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Matt
Sent: Friday,
February 20, 2004 9:22 AM
To: [EMAIL PROTECTED]
Subject: Re: [Declude.JunkMail]
Zombie Test
Kami,
I turned SKIPIFWEIGHT off for this version of the test and found that it scored
over 40% of my spam. With SKIPIFWEIGHT on, it scores around 3% to of the
spam (stuff that would have been held or hadn't yet reached a hold
weight). Because this hits only combinations of tests, you are much more
likely to be hitting things that have already scored at least close to your
hold range. I expect that this will have only a little impact on the spam
that's getting through my system, though there's not much opportunity there,
but it will most definitely help to clear out about my hold file.
I have another version that I may share here later on that increases the hit
rate to well over 50% of spam without SKIPIFWEIGHT (5% of spam with) and
targets about 10% of a hold file, and has a much better chance of finding stuff
that might have scored lower than Hold. The new test adds points to 80%
of the DUL hits that land in my Hold file, and from about 2,500 DUL hits a day
on my system, I'll probably only end up with about 10 that are left within 250%
of my Hold weight. The newer/unreleased version is also much more
conservative with points, only adding 5 per combination, though most such
E-mail ends up failing multiple combinations.
I consider my system to be very tight, only holding less than 5% of spam from
10 to 24 points, so many might see an even better effect.
Can you tell that I like numbers :)
Matt
Kami Razvan wrote:
"Please return the favor and
share with me any FP's that you see on this test so that I can make adjustments
for the benefit of myself as well as others."
Matt:
Interesting test.. I added your
filters to our system and just checked the log files. I simply added the test
with zero weight to watch how it performs.
In the 3 hours it has been added
(with 0 weight) it has had 100% success rate with about 100 or so
catches. All emails had already exceeded the delete weight. So it
seems like so far it has not had any false positives since every single hit is
already in the delete range.
As you stated I think it is a good
30%-50% delete weight test to just help the ones that are not deleted to go
over the mark.
--
=====================================================
MailPure custom filters for Declude JunkMail Pro.
http://www.mailpure.com/software/
=====================================================