Robust Recommender Systems

by Robin Burke (DePaul University)

Recommender systems bring thousands and sometimes millions of users together to share opinions, automating the large-scale sharing of expertise. However, some of these users are not genuine enthusiasts seeking to share their experiences, but rather marketers, spammers and vandals, all of whom seek to bias a system’s results. It is obvious that a sufficiently large population of such malicious users will render any collaborative system useless. What options do system implementers and maintainers have? The purpose of this tutorial is to explore some of the recent research in the area of recommender system robustness. In particular, we will look at recent work in attack modeling, the quantification of attack impact, alternate recommendation algorithms, trust maintenance, and attack detection.