[Spambayes] Is this a sign of future problems ?
tim.one at comcast.net
Tue Dec 16 16:53:57 EST 2003
[David L. Kindred]
> What if the goal is not to try and trick the Bayesian filter into
> treating spam as ham, but the inverse?
I already know what the goal is, and already explained it <wink>.
> Could this be an attempt at a kind of "denial-of-service" attack by
> trying to get the filter to start treating everything as spam? Would
> that idea work?
No. The empirical proof is in the pudding: is *anyone* here seeing an
increase in false positives due to training on these
collection-of-random-word msgs as spam? If the spammers were well-organized
and picked on sets of *common* words in a coordinated way, just maybe, but
these appear to be picked out of a dictionary at random. It doesn't matter
one whit to me, e.g., whether "bedimmed" gets treated as hammy or spammy in
my database, because I'll never see it in real life. The same is true of
most words picked at random out of a dictionary (the day-to-day working
vocabulary of most adults is a small percentage of all the words in even a
poor dictionary -- and a tiny percentage of the words in, e.g., the OED).
More information about the Spambayes