mattg at assembly.state.ny.us
Mon Jan 12 14:23:08 UTC 2009
In addition to what David and Pavel have mentioned, I'll
add that if you train on backscatter as spam, you might
also want to train on some legitimate bounces as ham,
otherwise your scoring might be skewed against DSNs of
all sorts. This might not be a problem for you... just
something to consider.
Personally I find that, overall, bounces (both legit and
backscatter) tend to defy Bayesian training, so I have
some bizarre custom MIMEdefang/SpamAssassin rules in place
to handle bounces. Of course, the bogofilter score weighs
heavily in these rules.
More information about the Bogofilter