Le 18/11/2010 21:06, Stan Hoeppner a écrit :
Victor Duchovni put forth on 11/18/2010 12:52 PM:
This filter is too fragile IMHO. My best advice is to find filters that
detect spam.
I think you've missed some of my previous posts regarding my spam
filtering setup. I kill about 99.5+% of it at SMTP time, without
resorting to body filters or any external content filters such as
SpamAssassin. I'm stalking the last half percent Victor. Probably a
fruitless effort. Not long ago someone stated on another list that the
goal in fighting spam isn't necessarily to kill it all, but to simply
make the problem manageable. I'm probably crossing that line with this
effort.
If you give me a budget, I can certainly demonstrate mathematically that
you can't reduce your FN rate under a certain level without increasing
your FP rate.
But the fact that we don't receive legit mail composed in
non-English languages does make it spam.
FPs are those rare messages that don't match your history. remember,
we've blocked each other!
RFC 2047 is not about non-Enlish. It enables encoding of non-ASCII
characters, these may crop up in English text from time to time.
It is naïve to assume otherwise :-)
Yes, I fully understand this possibility. My analysis and action is
based on mail flow seen here. And I'm not looking to filter all mail
with this, simply the suspect stuff I've identified.
see above. you can't reach 0 FN without increasing the FP rate.
Does my motivation here make more sense now?
Viktor suggestion is very wise (as usual!). focus on blocking a large
ratio of spam.
I can understand your reaction when you see a spam that slips. but calm
down. count how much spam is getting in. if it's not much, accept it.
think about the law of "diminishing returns". it may make you feel
strong to think that you can win by investing in more weapons, but look
around (did I say Afghansitan?) and you'll see that this approach
doesn't work. you don't win by going inside the battle field decied by
the opponent (ask a chess player, as it'll be hard to ask Napoleon...).
