On 1/29/10 12:37 PM, Michael Thomas wrote:
no, no, no, you misunderstand me. I'm not saying that prefiltering is
bad. far from it.
this started out with the claim that end users with their TiS buttons
are between useless
and dangerous. Users may not get all bad things (duh), but they also
know a heck of a
lot more about what they don't want than some system-wide classifier
that is BY DESIGN
allergic to false positives.
And I'll take issue with "only absolute classification helps solve
Rich's problem." In fact,
users on a day 0 exploit are your only line of defence so you better
damn well hope that
some percentage of them push the panic button, and you'd be foolish to
not
build systems that take those early warnings into account.
The point i'm making is that the user MUST be a part of the larger
problem of managing
their own firehose. The absolutist
"spam/ham"-must-be-part-of-priesthood is but one way
to achieve a level of filtering with relatively few false positives,
but it is not the whole
picture. The whole picture is "don't show me what I don't want to see".
Mike,
It really depends upon how feedback is applied. When applied globally
and not for the individual offering the feedback, then this _will_ cause
false positive detections. You are right, this type of information does
provide early indications that content filtering is failing to detect
some new spam variant, but then this requires human intervention to repair.
What I don't understand is your desire to muddle with spam definitions
calling the feedback "This is Spam" and not "This is Junk". Calling it
junk accurately describes how it is used, and what it ultimately means.
You don't need to be part of priesthood to understand what might cause
false positive blocking of otherwise legitimate messages.
Recently, our group made this mistake of mis-applying this type of
feedback, where the confusion appeared due to the sloppy labeling of
feedback as "spam" when it really meant, "not wanted". Not wanted is
usually the output of any content filter as well, the result of the
filter's training.
-Doug
_______________________________________________
Asrg mailing list
Asrg(_at_)irtf(_dot_)org
http://www.irtf.org/mailman/listinfo/asrg