I think these are good questions. But I also think there are good answers.

One solution is for my client to sentiment-analyze as a filter after the api-call and downrank that way (much like reddit/HN downrank controversials).

Maybe a more generic solution is to allow parties to "tag" twitter accounts. Clients could then make intelligent deductions based on tags (e.g. "political," "controversial," "satirical," "pg13").

I'm sure many other people have many other smart ideas.