Structured Feedback for Preference Elicitation in Complex Domains