Recommender systems make use of different sources of information for providing users with recommendations of items. Such systems are often based on collaborative filtering methods which make automatic predictions about the interests of a user by collecting taste information from many users. As an alternative approach, we propose to use the concept of resolving set that allows to determine the preferences of the users with a very limited number of ratings. We also show how to make recommendations when user ratings are imprecise or inconsistent, and we indicate how to take into account situations where users possibly don't care about the attribute values of some items. All recommendations are obtained in a few seconds by solving integer programs.
Published February 2019 , 15 pages