[extropy-chat] Webpanto

Robert J. Bradbury bradbury at aeiveos.com
Fri Jan 23 01:24:49 UTC 2004


On Thu, 22 Jan 2004, scerir wrote about junk mail including:

> > it calumny furthermore handy chivalry paucity impermissible
> > ineluctable castillo ionospheric conch empathy cayenne cassock
> > crankcase ancestry imperial admix emil infancy carryover bone
> > inferential dim

You can read all about this in an article by Paul Graham:
  So Far So Good
  August 2003
  http://www.paulgraham.com/sofar.html

He discusses why this approach to getting around Bayesian
filtering will probably not work.

As far as combining rule based filtering and Bayesian filtering
my stats so far this week (5 days) are:

Messages     Lines    Words    Chars
    1358    130296   524989  5953093  %SPAM
     360     23638    83258  1008384  %BLOCKED
      63      4988    23744   236576  %SPAMBAYESIAN

The first two files are generated by a rule based filter
(SpamBouncer), the last file is produced by a trained
Bayesian filter (SpamProbe).

SpamProbe only gets to look at what SpamBouncer doesn't catch.

So far I'm averaging between 1 and 2 SPAMs a day getting
through this defense.  For those who don't want to do the
math it works out to about 1 SPAM message every 4-5 minutes.

Robert





More information about the extropy-chat mailing list