Toward a decision-theoretic framework for affect recognition and user assistance