Presenting Challenging Recommendations : Making Diverse News Acceptable

Recommender systems find relevant content for us online, including the personalized news we increasingly receive on Twitter and Facebook. As a consequence of personalization, we increasingly see content that agrees with our views, we cease to be exposed to views contrary to our own. Both algorithms and the users themselves filter content, and this creates more polarized points of view, so called “filter bubbles” or “echo chambers”. This paper presents a vision of a diversity aware recommendation model, for the selection and presentation of a diverse selection of news to users. This diversity aware recommendation model considers that: a) users have different requirements on diversity (e.g., challenge-averse or diversity seeking), and that b) items will satisfy these requirements to different extents (e.g., liberal or conservative news). By considering both item and user diversity this model aims to maximize the amount of diverse content that users are exposed to, without damaging system reputation.