spam classification breaker

John Graham-Cumming google at jgc.org
Sat Feb 7 17:17:58 CET 2004


"Delaney, Timothy C (Timothy)" <tdelaney at avaya.com> wrote in message news:<mailman.1274.1076023169.12720.python-list at python.org>...
> Of course, once I've started training emails with such web bugs (which 
> contain specific URLs, etc) as spam, pretty soon none of those emails 
> are going to be successfully classified as ham.

Right, to make it work you'd have to make the web bugs unique over the
message (perhaps use a zombie army of hacked machines so that each web
bug goes to a different IP that redirects to you).
 
> John didn't state in his article (IIRC - I read it a few days ago when 
> it appeared on Slashdot ;) but I don't think he was training his corpus 
> on those emails - just letting them be classified.

No, and that's an interesting question that I'd like to explore. 
Suppose I train on each mail as it gets through do I still find the
ham.

John.



More information about the Python-list mailing list