A Framework for Recommending Collections

To date, the vast majority of recommender systems research has addressed the problem of recommending individual items that the user will like. Recommending collections of items rather than individual items is an important open space of research in the recommender systems community. In this paper, we present a comprehensive framework for describing and evaluating collections of items. This framework is designed to be domain independent and applicable to any collection recommendation problem. Our framework includes a categorization scheme for describing collections and a list of features upon which a collection can be evaluated. We present a number of examples that showed how these different attribute and evaluation techniques can be combined and applied in a given domain. We then discuss issues relevant to the building of these systems. This includes challenges in obtaining data about users’ preferences for collections. We propose methods that include obtaining and analyzing existing collections from websites and developing multi-player online games to generate data about replacements and preferences. In addition, we look at how collection recommenders could be used to assist users either by creating collections from scratch or by assisting users in their own collection creation tasks. We believe this framing of an important problem will lead to new research in the development and evaluation of algorithms for recommending collections in interesting applications and with cross-domain applicability.