Mitigating Fairness and Efficiency Tradeoff in Vehicle-Dispatch Problems

Abstract

We propose a fair-assignment algorithm between vehicles and passengers to mitigate the efficiency and fairness tradeoff for on-demand ride-hailing platforms. Ride-hailing platforms connect passengers and drivers in real time. While most studies focused on developing an optimally efficient assignment method for maximizing the profit of the platform, optimal efficiency may lead to profit inequality for drivers. Therefore, fair-assignment algorithms have begun to attract attention from artificial-intelligence researchers. While a fair-assignment algorithm based on max-min fairness, which is a representative concept of fairness, has been proposed, profit inequality among drivers still remains when assignments are made multiple times. To address such inequality, we develop a fair-assignment algorithm called the priority assignment algorithm PA(k) to give priority to drivers with low cumulative profit then generate an optimally efficient assignment for the remaining drivers and passengers. We also develop a method of dynamically determining the number of priorities at each assignment. We experimentally demonstrated that PA(k) outperforms the existing fair assignment algorithms in both efficiency and fairness in the case of excess supply by using a real-world dataset.