A Practical Framework for Online Driver Repositioning to Balance Supply and Demand in Ride‑Hailing Platforms
This article presents a three‑stage, data‑driven framework for online driver repositioning that generates candidate dispatch tasks, scores them using a marginal gain model, and selects optimal tasks via a minimum‑cost flow planning algorithm, demonstrating significant improvements in driver efficiency and experience through large‑scale A/B experiments.
With the rise of mobile internet, ride‑hailing has become a common travel option, yet platforms often suffer from supply‑demand imbalance, causing passengers to miss rides and drivers to idle. The paper defines the driver dispatch problem: the platform interrupts a driver’s self‑directed cruising and guides them to a destination with higher order‑receiving probability.
A dispatch task consists of four elements—driver, target location, expiration time, and compensation amount—and follows three possible successful outcomes (status 2 and 3) or a failure (status 4). Failed tasks trigger compensation to maintain driver trust.
The proposed solution consists of three stages. Stage 1: Candidate Generation filters idle drivers, selects candidate target cells (nearby grids, historically frequent grids, city hotspots), picks a POI within each cell, and sets expiration based on ETA. A failure‑probability model discards high‑risk tasks and determines compensation proportional to travel distance.
Stage 2: Task Scoring models the relationship between answer rate and supply‑demand ratio with a piecewise linear function, computes the marginal gain of adding an idle driver to a spatio‑temporal cell, and derives a driver‑gap metric for each cell.
Stage 3: Planning formulates the selection of final dispatch tasks as a constrained optimization problem that maximizes global platform benefit while respecting driver experience constraints; this is transformed into a minimum‑cost flow problem (see Figure 5).
The framework was deployed on the Didi platform and evaluated through multiple online A/B experiments. Results show increased driver efficiency, higher total income, and positive driver feedback (NPS 27.0%, 64.6% willingness to accept future tasks). The paper also discusses future directions such as reinforcement‑learning‑based end‑to‑end solutions and direct route optimization using road‑network data.
Reference: "When Recommender Systems Meet Fleet Management: Practical Study in Online Driver Repositioning System" (https://dl.acm.org/doi/abs/10.1145/3366423.3380287).
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
DataFunTalk
Dedicated to sharing and discussing big data and AI technology applications, aiming to empower a million data scientists. Regularly hosts live tech talks and curates articles on big data, recommendation/search algorithms, advertising algorithms, NLP, intelligent risk control, autonomous driving, and machine learning/deep learning.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
