Can recommenders compensate for low QoS?

Content recommendation systems, also known as recommenders, are pervasive and impact a significant portion of users demands over the Internet. Although recommenders have been primarily devised to account for users interests with respect to the content catalog, mobile users are typically served by a network that is unreliable and subject to losses and low QoS. Can content recommenders compensate for low QoS? To answer this question, we conducted experiments over the Internet, and report our findings on (i) the characterization of QoS and (ii) the compensation for low QoS. Our measurements suggest that content that is far from the trends tends to be far from the user. We quantify the extent at which unpopular content tends to be served with lower QoS and establish a methodology to determine the relationship between contents’ popularity and its physical proximity to the users. Then, we verify that making requests a bit trendier can hit much closer content. In particular, our results suggest conditions under which a recommender can compensate for low QoS, at zero costs for operators.