Traditional multi-criteria selection methods are the leading approach for selecting a set of candidates when multiple criteria determine selection relevancy. For instance, hiring platforms combine candidates' proximity, skills, and years of experience to build shortlists for recruiters. While these methods succeed in efficiently selecting candidates, their chosen set may unfairly affect marginalized candidate groups (e.g., race or gender). Bridging the gap between traditional fairness-unaware multi-criteria selection and contemporary fairness interventions, we characterize the open problem of fair multi-criteria selection. We design FAIR&SHARE the first efficient fairness-tunable multi-criteria selection method. FAIR&SHARE supports several fair representation notions. The key to FAIR&SHARE is the design of its group-aware utility objective. FAIR&SHARE uses a novel fairness calibration component to provide a user-friendly tuning mechanism for controlling the balance between selection relevancy (utility) and representation fairness. Our fairness-focused selection policy iteratively builds the result set by prioritizing candidates as aiding either the fair representation or the share-d overall utility goals. We prove the optimality of FAIR&SHARE, meaning that FAIR&SHARE selects the best possible candidates such that the desired fair representation is achieved. Our experimental study demonstrates that FAIR&SHARE achieves the best fairness and utility performance of state-of-the-art alternatives adapted to this new problem while taking a fraction of the time.
