I am only half kidding. "Profiles of specialized Twitter readers" would be an excellent dataset if it could somehow be filtered down to that.