Exploring trajectory embedding via spatial-temporal propagation for dynamic region representations

Chunyu LIU, Hongli ZHANG*, Guopu ZHU, Haotian GUAN, Sam KWONG

*Corresponding author for this work

Research output: Journal PublicationsJournal Article (refereed)peer-review

1 Citation (Scopus)

Abstract

In smart cities, the development of urban regions stands as a fundamental pillar in the planning process, significantly influencing the overall urban living experience. Effective representations of regions are essential for providing fundamental insights and enabling various applications in urban computing. While research on regional embeddings, especially in dynamic urban representations, has gained considerable attention, there is often a lack of in-depth investigation into the reciprocal impact of mobility trajectories and spatiotemporal interactions. To address this challenge, we present a novel Spatial-Temporal Dynamic Representation framework for urban regions (STDR) to uncover the dynamic functions and variation patterns. Our model leverages interaction information between human mobility and regional features based on motion trajectories, enabling time and geographic encoding for each region. It then combines temporal propagation and spatial proximity to aggregate dynamic function representations. Moreover, it implements a spatiotemporal gating mechanism addressing the imbalance issue in global spatiotemporal transmission. Compared with state-of-the-art research methods, our method can achieve more accurate performance in two downstream tasks.
Original languageEnglish
Article number120516
JournalInformation Sciences
Volume668
DOIs
Publication statusPublished - May 2024

Bibliographical note

Publisher Copyright:
© 2024 Elsevier Inc.

Keywords

  • Region representation
  • Spatial-temporal trajectory
  • Trajectory embedding
  • Urban computing

Fingerprint

Dive into the research topics of 'Exploring trajectory embedding via spatial-temporal propagation for dynamic region representations'. Together they form a unique fingerprint.

Cite this