Recommender systems and their effects
Jakub Marecek, IBM Research Ireland
3-4pm 20th Mar 2019
Recommender systems are widely used in settings, where actions impact the recommendations, which in turn have impact on the actions. For an example of such a closed-loop setting, consider navigation systems, which use information about travel times to recommend a route. If the particular navigation system is used widely enough, the recommendation may impact the future traffic state, possibly rendering the recommendation suboptimal from both the point of view of the driver and the society as a whole. Similar effects can be illustrated on the recommendations of restaurants. If a small bistro without a table reservation system becomes top ranked, many customers may arrive at its door and get turned down, leading to poor reviews. Further, there can be issues related to priming, for example when the reviews suggest the place is not touristy. Several problems arise, including recovery of unbiased user models in the presence of recommenders and developing recommenders that allow for some guarantees on the closed-loop behaviour of the system. The talk will cover both the optimisation and control aspects.
Together with some fabulous colleagues, Jakub Marecek develops solvers for optimisation and control problems at IBM Research -- Ireland. Jakub joined IBM Research from the School of Mathematics at the University of Edinburgh in August 2012. Prior to his post-doc in Edinburgh, Jakub had presented an approach to generalpurpose integer programming in his dissertation at the University of Nottingham and worked in two start-up companies.