On Fri, Aug 16, 2002 at 08:51:24PM -0700, Kenneth Porter wrote:
> On Fri, 2002-08-16 at 13:28, Timothy H. Keitt wrote:
> > The essential ingredient is to have a "mark as spam" button readily
> > available so that one can train a classifier (employing whatever method
> > one chooses). I guess it would help to have a "mark as false positive"
> > button as well, or perhaps make it a toggle.
> 
> I don't see it happening in evo, although I'd like to get at this kind
> of feature.
> 
> It wouldn't be hard to add if evo let one apply a single filter to a
> selection of messages. Filters can invoke an external script, so you
> could define two scripts to add the messages to the selected
> spam/nonspam corpus.
> 
> AFAIK, evo only allows one to apply the entire filter list to the
> message list selection.
Wouldn't it be possible to separate the index creation form the  filter and
the classification.  Run the index creation ones every 15min and use it as
an input for some of the bow programs from cmu for classification in a rule.
I guess that will do the trick, I haven't used an external rule pgm but if
that doesn't work writing a rule extension using libbow shouldn't be that 
difficult either. 
see:
http://www-2.cs.cmu.edu/~mccallum/bow/

Ronald
> 
> 
> _______________________________________________
> evolution maillist  -  [EMAIL PROTECTED]
> http://lists.ximian.com/mailman/listinfo/evolution

_______________________________________________
evolution maillist  -  [EMAIL PROTECTED]
http://lists.ximian.com/mailman/listinfo/evolution

Reply via email to