topic classifier, take 3

here's my third attempt at a newspaper article topic classifier (19
categories plus uncertain and out):

:try-to-classify: (:article:)
{
    {
	isolate (:stats:)
	classify <osbf microgroom> [:article:] \
		(bus.cfc cli.cfc con.cfc cul.cfc dom.cfc \
		eco.cfc edu.cfc ene.cfc ent.cfc env.cfc \
		for.cfc hea.cfc loc.cfc pol.cfc sci.cfc \
		spo.cfc tec.cfc tra.cfc wor.cfc | unc.cfc out.cfc) (:stats:)
	match (:: :class-conc:) [:stats:] /(...)\.cfc/
	{ #classify succeeded, checking pR:
	    match (:: :pR:) [:stats:] /Best match.*: (.*)  \nTotal/
	    output /pR::*:pR:. /
	    eval /:@::*:pR: > 30:/ #play around with this number
	    output /Classify succeeded as ':*:class-conc:'\n/
	    return
	} #pR wasn't high enough; asking the user for input:
	isolate (:conc-file:) /:*:class-conc:.cfc/
	isolate (:interaction:) /:*:class-conc::*:article::*:stats:/
	call /:interact:/ [:*:interaction:] #asking the user for the right category
	isolate (:user-feedback-conc-file:) /:*:user-feedback:.cfc/
	isolate (:count:) /12/
	{ # entering the training loop
	    learn <osbf microgroom> (:*:user-feedback-conc-file:) [:article:]
	    output /Learned as :*:user-feedback-conc-file:\n/
	    #reclassifying to check improvement:
	    isolate (:stats2:)
	    classify <osbf microgroom> [:article:]	\
		    (bus.cfc cli.cfc con.cfc cul.cfc dom.cfc \
		    eco.cfc edu.cfc ene.cfc ent.cfc env.cfc \
		    for.cfc hea.cfc loc.cfc pol.cfc sci.cfc \
		    spo.cfc tec.cfc tra.cfc wor.cfc | unc.cfc out.cfc) (:stats2:)
	    isolate (:pR2:)
	    match (:: :pR2:) [:stats2:] /Best match.*: (.*)  \nTotal/
	    isolate (:improvement:)
	    eval (:improvement:) /:@:(:*:pR2: - :*:pR:):/
	    output /imp::*:improvement:\n/
	    {
		eval /:@::*:improvement: > 3:/
		eval /:@::*:pR2: > 10:/ #play around with this number too
		output /Improvement high enough, and new threshold met; NO further
action taken\n/
		return
	    } # imp wasn't high enough, and new threshold wasn't met
	    isolate (:refute-args:) /:*:user-feedback-conc-file::*:article:/
	    #call /:refute:/ [:*:refute-args:] #this refutes the article from
ALL the other categories
	    eval (:count:) /:@::*:count: - 1:/
	    output /decremented::*:count:\n/
	    eval /:@::*:count: > 0:/
	    liaf
	    output /Improvement never increased enough\n/
	} #exiting the loop because pR never improved enough
    } # try-to-classify failed (category is unc or out)
    # my ad-hoc learning function (simply looks at the url to find keywords):
    call /:try-to-learn:/ [:*:article:]
    { # did try-to-learn fail?:
	match [:learn-conc:] /unc/
	output /Could not confidently classify the piece\n/
	return
    } # try-to-learn succeeded
    output /Never succeeded in confidently classifying the piece as
':*:user-feedback:', but Try-To-Learn succeeded as ':*:learn-conc:'
after Try-To-Classify FAILED as ':*:class-conc:'\n/ #this means that
the program trusts the newspapers' urls more than the user input. hm.
    return
}

apart from this, i am seeing, even after having run try-to-learn for
quite a while, that try-to-classify 'succeeds' very confidently in
classifying everything as technology, even though i can clearly see
that it belongs to sports.

------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com