Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Ride-Hailing Forecasting at DiDi with Multi-View Geospatial Representation Learning from the Web

Xixuan Hao, Guicheng Li, Daiqiang Wu, Xusen Guo, Yumeng Zhu, Zhichao Zou, Peng Zhen, Yao Yao, Yuxuan Liang

TL;DR

This paper tackles the challenge of forecasting ride-hailing demand and supply under geospatial heterogeneity and external shocks. It introduces MVGR-Net, a two-stage framework that first learns region representations from semantic POI attributes and temporal mobility patterns, then conducts forecast generation through a prompt-empowered, LLM-based pipeline that integrates exogenous factors and textual prompts via LoRA fine-tuning. The approach delivers state-of-the-art performance on DiDi real-world data, with consistent improvements in Call and TSH and strong qualitative and deployment results, including an embedding vector library and online A/B tests. The work demonstrates practical impact by enabling more accurate demand-supply forecasting, smarter subsidy allocation, and scalable integration of geospatial priors into production systems.

Abstract

The proliferation of ride-hailing services has fundamentally transformed urban mobility patterns, making accurate ride-hailing forecasting crucial for optimizing passenger experience and urban transportation efficiency. However, ride-hailing forecasting faces significant challenges due to geospatial heterogeneity and high susceptibility to external events. This paper proposes MVGR-Net(Multi-View Geospatial Representation Learning), a novel framework that addresses these challenges through a two-stage approach. In the pretraining stage, we learn comprehensive geospatial representations by integrating Points-of-Interest and temporal mobility patterns to capture regional characteristics from both semantic attribute and temporal mobility pattern views. The forecasting stage leverages these representations through a prompt-empowered framework that fine-tunes Large Language Models while incorporating external events. Extensive experiments on DiDi's real-world datasets demonstrate the state-of-the-art performance.

Enhancing Ride-Hailing Forecasting at DiDi with Multi-View Geospatial Representation Learning from the Web

TL;DR

This paper tackles the challenge of forecasting ride-hailing demand and supply under geospatial heterogeneity and external shocks. It introduces MVGR-Net, a two-stage framework that first learns region representations from semantic POI attributes and temporal mobility patterns, then conducts forecast generation through a prompt-empowered, LLM-based pipeline that integrates exogenous factors and textual prompts via LoRA fine-tuning. The approach delivers state-of-the-art performance on DiDi real-world data, with consistent improvements in Call and TSH and strong qualitative and deployment results, including an embedding vector library and online A/B tests. The work demonstrates practical impact by enabling more accurate demand-supply forecasting, smarter subsidy allocation, and scalable integration of geospatial priors into production systems.

Abstract

The proliferation of ride-hailing services has fundamentally transformed urban mobility patterns, making accurate ride-hailing forecasting crucial for optimizing passenger experience and urban transportation efficiency. However, ride-hailing forecasting faces significant challenges due to geospatial heterogeneity and high susceptibility to external events. This paper proposes MVGR-Net(Multi-View Geospatial Representation Learning), a novel framework that addresses these challenges through a two-stage approach. In the pretraining stage, we learn comprehensive geospatial representations by integrating Points-of-Interest and temporal mobility patterns to capture regional characteristics from both semantic attribute and temporal mobility pattern views. The forecasting stage leverages these representations through a prompt-empowered framework that fine-tunes Large Language Models while incorporating external events. Extensive experiments on DiDi's real-world datasets demonstrate the state-of-the-art performance.
Paper Structure (36 sections, 18 equations, 11 figures, 3 tables)

This paper contains 36 sections, 18 equations, 11 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Preliminary
  3. Methodology
  4. Multi-View Geospatial Representation Learning
  5. Prompt-Empowered Forecasting
  6. Ride-Hailing Forecasting with Exogenous Factors
  7. Heterogeneity-Informed Prompt Learning
  8. Experiments
  9. Experimental Setup
  10. Datasets
  11. Baselines
  12. Metrics and Implementation
  13. RQ1: Overall Performance
  14. RQ2: Ablation Studies
  15. RQ3: Qualitative Analysis
  16. ...and 21 more sections

Figures (11)

  • Figure 1: Geospatial Heterogeneity and its two principal view in the context of ride-hailing services.
  • Figure 2: The Overall Framework of MVGR-Net.
  • Figure 3: The proportional distribution of top-10 and geographic distribution of top-5 POI primary categories.
  • Figure 4: Results of ablation studies on both WMAPE and MAE metrics. PGN stands for Prompt Generation Network, EV denotes External Variables.
  • Figure 5: Performance comparison for different hyperparameter setting on 2025.
  • ...and 6 more figures