More than the sum of its members: challenges for group recommender systems