Re: Best way to classify...

On Tue, Apr 07, 2009 at 09:42:54AM -0700, Tracy Reed wrote:
> 
> Or is it a binary decision sort of thing where I have to build a
> binary decision tree and consider spam vs ham and of the ham consider
> ham about cats vs ham about dogs?

check Thomas' article sorter (recent thread) for a multiclass example,
have a look at the gatling sorter in the example scripts in source for
another way.
 
> Is one more accurate than the other?

depends on classifier type and 'glue logic' around it, ie your training
algo/strategy, beside nature of stuff to classify itself, of course.
We know 2 classes - binary - works fine, hence an n-classes, n>2, 
binary-tree classifier should work fine. We've limited (reported) 
experience for multiclass cases, and more precisely we're lacking methods 
to measure the performance of such things (we'd need to extend quality 
assessment procedure/figures as were employed for antispam filter 
'competition' - see spamconference stuff, links on our website).

I suggest you try to implement both way with elementary scripts, and
different classifiers (ATM if you want to use different classifiers
you need to go the binary-tree way), and see what you get.


HTH
-- 
paolo


------------------------------------------------------------------------------
This SF.net email is sponsored by:
High Quality Requirements in a Collaborative Environment.
Download a free trial of Rational Requirements Composer Now!
http://p.sf.net/sfu/www-ibm-com