OT: spam filtering idea

Paul Rubin phr-n2002b at NOSPAMnightsong.com
Mon Jan 13 22:13:51 EST 2003


Tim Peters <tim.one at comcast.net> writes:
> Cursed to do so, yes <wink>.  Trying to train one of these classifiers to
> serve a diverse group of users at once is demonstrably and quantifiably much
> less effective.  

Yes, the hope is you get some of the effectiveness back by giving
extra weight to words found in recently received spam.  The
observation is individual pieces of spam tend to circulate for fairly
short periods, so if you spot words from them during that period, that
tells you something even if the messages mutate (all the similar
Nigerian spams).

By the way, here's a hysterically funny variation ("urgent
counter-proposal") on the Nigerian spam:

  http://www.nightsong.com/phr/urgent.txt

It's from comp.dcom.telecom and I saved it.




More information about the Python-list mailing list