Delayed Matching Considering User Patience in Ride-Sourcing System

Ride-sourcing system reshaped the way people travel. Online matching between the idle drivers and passengers is an important problem in a ride-sourcing system. This paper constructs a new multi-stage learning framework, which is used to determine the optimal delay time in ride-sourcing system, and achieve a better matching result while considering the user's patience. We conducted experiments to evaluate the proposed framework and verified its effectiveness, highlighting the importance of patience. This paper presents a novel application of reinforcement learning, showcasing examples of how to apply reinforcement learning in scenarios with large action spaces. In the example of this paper, it achieved good results and can be extended to other application scenarios.