Spam filters routinely filter out emails based on a common set of words that are used without understanding the subject matter. For example, if an email contains the words viagara and store, the probability of that email being spam go up tremendously. Paul's "A plan for Spam" essay explains a lot more. And, from what I understand, Bayesian filtering is at the root of most spam blockers out there.
I'd be willing to wager that Bayesian filtering would be pretty powerful in filtering out trolls on social sites, as well. For example, if a post contains the words "asshole" and "fucktard" the probability of that post coming from a troll goes up exponentially.
If you trained a Bayesian filter, or spam algorithm well enough, it should be able to flag trollish posts fairly easily.