A method for recommending items includes forming one or more choice sets of recommendable items. At least one choice set is a personalized choice set of recommendable items formed for a first user based in part on data associated with that user. Recommendable items can be divided into classification groups. Forming a choice set of recommendable items can be based on a user's classification group preferences. The method also includes forming a recommendation set of recommendable items from the choice sets of recommendable items. Forming the recommendation set is based in part on additional data associated with the user. The additional data is not used in forming the personalized choice set.