Clusters of opinion may be accidental, e.g. many lemmings follow Eliezer Yudkowsky who is correct on three topics and wrong on two. Or some other pundit. I think such accidental correlations will drown out whatever useful signal you were hoping to uncover by factor analysis. It's a fishy endeavor anyway, smells like determining truth by popular vote spiced up with nifty math. What if all smart people start using your algorithm? You could get some nasty herd effects...