I think this is a cool project, but it "needs work."

I tried George Bush, figuring that it would work perfectly in this SUPER well know case.

The results were close to bogus. Too bad.

Have you tried Bayes Nets to make more plausible inferences? In a case like this there has to be enough web data to give a good answer.

Good luck with a nice project.

Which project are you referring to?