Spatio-Temporal Hierarchical Adaptive Dispatching for Ridesharing Systems

TL;DR

This paper introduces a hierarchical adaptive dispatching approach for ridesharing systems that optimizes order matching by dynamically adjusting dispatch intervals based on spatio-temporal demand patterns, significantly improving platform profit.

Contribution

It proposes a novel hierarchical adaptive dispatching framework with online algorithms that adapt to demand variability, outperforming existing uniform-interval methods.

Findings

Algorithms improve profit over existing methods

Hierarchical clustering effectively captures demand patterns

Extensive experiments on large-scale dataset validate effectiveness

Abstract

Nowadays, ridesharing has become one of the most popular services offered by online ride-hailing platforms (e.g., Uber and Didi Chuxing). Existing ridesharing platforms adopt the strategy that dispatches orders over the entire city at a uniform time interval. However, the uneven spatio-temporal order distributions in real-world ridesharing systems indicate that such an approach is suboptimal in practice. Thus, in this paper, we exploit adaptive dispatching intervals to boost the platform's profit under a guarantee of the maximum passenger waiting time. Specifically, we propose a hierarchical approach, which generates clusters of geographical areas suitable to share the same dispatching intervals, and then makes online decisions of selecting the appropriate time instances for order dispatch within each spatial cluster. Technically, we prove the impossibility of designing constant-competitive-ratio algorithms for the online adaptive interval problem, and propose online algorithms under partial or even zero future order knowledge that significantly improve the platform's profit over existing approaches. We conduct extensive experiments with a large-scale ridesharing order dataset, which contains all of the over 3.5 million ridesharing orders in Beijing, China, received by Didi Chuxing from October 1st to October 31st, 2018. The experimental results demonstrate that our proposed algorithms outperform existing approaches.