Develops and tests a recommendation engine that extends collaborative filtering with demographic and item attributes and uses multi-armed bandit techniques to reveal consumer preferences to overcome the challenge of “cold-starts” with little or no prior data and problems of scale associated with multiple users and attributes.