Algorithmic bias in rankings

Abstract

In this talk we refer to bias in its everyday sense, as a prejudice against a person or a group, and ask whether an algorithm, particularly a ranking algorithm, can be biased. We begin by defining under which conditions this can happen. Next, we describe key results from research on algorithmic fairness, much of which studies automatic classification by a supervised learning method. Finally, we attempt to map these concepts to rankings and to introduce new, ranking-specific ways of looking at algorithmic bias.