Modeling User Preferences Using Relative Feedback for Personalized Recommendations