ValuePick: Towards a Value-Oriented Dual-Goal Recommender System

Given a user in a social network, which new friends should we recommend, the dual goal being to achieve user satisfaction and good network connectivity? Similarly, which new products are better to recommend to satisfy customers´ taste/needs as well as increase vendor profit? Typical recommender systems use merely past purchases, product ratings, demographic meta-data, and network `proximity´ to make recommendations. This traditional approach, however, does not take into account the profitability of products to vendors in a customer-product network, or the efficacy of new links in a social network. We argue that it is more appropriate to view the problem of generating recommendations as an optimization problem. In this paper, (a) we propose Value Pick, a framework which incorporates the `value´ of recommendations into the system while still providing accurate recommendations that retain user trust; (b) our method is parsimonious (requires only a single parameter τ), flexible (τ is used to flexibly adjust the level of balance between `user satisfaction´ and `gain´), and general (can be used with any `value´ metric); and finally (c) we examine the problem in the social networks setting, simulate comprehensive experiments to compare our method to several basic heuristics, and show that Value Pick yields higher `gain´ while keeping user satisfaction high.