Mailscanner plus SpamAssassin and CRM114

[Kai Schaetzl]

Rick Cooper wrote on         Mon, 29 Mar 2004 10:09:09 -0500:

> It's not a bayes engine ...

Thanks for the correction/information, yes, I put that tool in the wrong
drawer. The problem I see with all these tools that "learn" and base their
analysis on that is that you have to teach them before they start being
useful. As I see with many of our clients they get so few mail (be it ham or
spam) that it would take a lot of time, months, for this. On the other hand
many clients get email which looks very spammy but isn't spam for them, so you
can't use a site-wide database, at least not with weighting measures and
white-listing. As it seems CRM114 is single-user, anyway, as you explained.
Maybe it could be a good replacement for the Bayes engine in SA but not for SA
in general. I'm quite sure that the accuracy will diminish if you have
thousands of users with diverse interests unless they all keep feeding and
feeding the learner. With SA I have nearly 100% accuracy right from the start.
If the Bayes engine fails then the rules will hit and vice versa. Actually,
with our own domain I see that almost all spam is flagged with Bayes_99 and
the rules just make up for reassurance and auto-learning the messages as spam.
Good against false positives.



Kai

--

Kai Schätzl, Berlin, Germany
Get your web at Conactive Internet Services: http://www.conactive.com
IE-Center: http://ie5.de & http://msie.winware.org