On Wed, May 31, 2006 at 11:23:40AM +0200, Yves Goergen wrote:
[...]
> A future change to my server should be to enable spam scanning for every
> incoming message but only deny messages depending on whether the user
> has wished so. Maybe even with separate required spam scores or so. This
> will require an independent handling of a message for each recipient anyway.
this ought to be practical -- you can run SpamAssassin at
SMTP time on all messages, but not discard based on score
then, and then at delivery time, when you know the
recipient and therefore what rules to use, you can pick
the list of matched tests out of the X-Spam-Status:
header, add up a user-specific score based on that user's
rules, and then decide whether to deliver or reject/drop
the message. I think the last part of this would require
some coding (SA doesn't seem to have a `recalculate score
based on existing test results' mode) but once implemented
it should be much quicker than scanning the message anew.
What I can't see how to do neatly in this system is
statistical (e.g. naive Bayes) classification, since for
this to be effective it has to have a per-user dictionary,
which probably means scanning at delivery-time is
necessary. There are much faster programs for this than
SpamAssassin, though.
--
``There has to be a balance between telling the truth and reassurance.''
(David Blunkett, on terrorist alerts)
