Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GTG: Generalizable Trajectory Generation Model for Urban Mobility

Jingyuan Wang, Yujing Lin, Yudong Li

TL;DR

The paper tackles cross-city trajectory generation under privacy and data-access constraints by proposing GTG, a framework that learns invariant mobility patterns across cities. It fuses Space Syntax-based topological features, disentangled travel-cost prediction with adversarial domain adaptation, and a travel-preference learning component guided by shortest-path search. Empirical results on three real-city datasets show that GTG generalizes well to unseen cities, surpassing baselines on macro and micro trajectory similarity metrics and benefiting downstream tasks and targeted fine-tuning. The approach offers practical value for privacy-preserving data augmentation and urban mobility analytics by enabling realistic trajectory synthesis in new cities with limited or no target-city data.

Abstract

Trajectory data mining is crucial for smart city management. However, collecting large-scale trajectory datasets is challenging due to factors such as commercial conflicts and privacy regulations. Therefore, we urgently need trajectory generation techniques to address this issue. Existing trajectory generation methods rely on the global road network structure of cities. When the road network structure changes, these methods are often not transferable to other cities. In fact, there exist invariant mobility patterns between different cities: 1) People prefer paths with the minimal travel cost; 2) The travel cost of roads has an invariant relationship with the topological features of the road network. Based on the above insight, this paper proposes a Generalizable Trajectory Generation model (GTG). The model consists of three parts: 1) Extracting city-invariant road representation based on Space Syntax method; 2) Cross-city travel cost prediction through disentangled adversarial training; 3) Travel preference learning by shortest path search and preference update. By learning invariant movement patterns, the model is capable of generating trajectories in new cities. Experiments on three datasets demonstrates that our model significantly outperforms existing models in terms of generalization ability.

GTG: Generalizable Trajectory Generation Model for Urban Mobility

TL;DR

The paper tackles cross-city trajectory generation under privacy and data-access constraints by proposing GTG, a framework that learns invariant mobility patterns across cities. It fuses Space Syntax-based topological features, disentangled travel-cost prediction with adversarial domain adaptation, and a travel-preference learning component guided by shortest-path search. Empirical results on three real-city datasets show that GTG generalizes well to unseen cities, surpassing baselines on macro and micro trajectory similarity metrics and benefiting downstream tasks and targeted fine-tuning. The approach offers practical value for privacy-preserving data augmentation and urban mobility analytics by enabling realistic trajectory synthesis in new cities with limited or no target-city data.

Abstract

Trajectory data mining is crucial for smart city management. However, collecting large-scale trajectory datasets is challenging due to factors such as commercial conflicts and privacy regulations. Therefore, we urgently need trajectory generation techniques to address this issue. Existing trajectory generation methods rely on the global road network structure of cities. When the road network structure changes, these methods are often not transferable to other cities. In fact, there exist invariant mobility patterns between different cities: 1) People prefer paths with the minimal travel cost; 2) The travel cost of roads has an invariant relationship with the topological features of the road network. Based on the above insight, this paper proposes a Generalizable Trajectory Generation model (GTG). The model consists of three parts: 1) Extracting city-invariant road representation based on Space Syntax method; 2) Cross-city travel cost prediction through disentangled adversarial training; 3) Travel preference learning by shortest path search and preference update. By learning invariant movement patterns, the model is capable of generating trajectories in new cities. Experiments on three datasets demonstrates that our model significantly outperforms existing models in terms of generalization ability.

Paper Structure

This paper contains 36 sections, 41 equations, 5 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Definitions
  4. Problem Statement
  5. Methodology
  6. Framework
  7. Topological Feature Extraction
  8. Space Syntax Feature Extraction
  9. Topological Feature Aggregation
  10. Travel Cost Prediction with Disentangled Representation
  11. Travel Cost Prediction
  12. Domain Discrimination
  13. Disentangled Adversarial Training
  14. Travel Preference Learning
  15. Experiments
  16. ...and 21 more sections

Figures (5)

  • Figure 1: Similar local topological structures in New York (left) and Chicago (right).
  • Figure 2: Overview of the framework
  • Figure 3: Travel cost (speed and time) distribution in Xi'an and Beijing.
  • Figure 4: The visualization of road segment visit frequency on the XA dataset, brighter color means higher frequency.
  • Figure 5: Four types of Space Syntax concepts: (a) node $i$ has a larger Total Depth than node $j$; (b) node $i$ is in the center of network, with larger Integration than node $j$; (c) Connectivity is only related with neighborhood nodes, node $i$ 's Connectivity is $5$; (d) node $i$ is a transportation hub in this network, so it has larger Choice than other nodes.

Theorems & Definitions (2)

  • Definition 1: Road Network
  • Definition 2: Trajectory