Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Grid and Road Expressions Are Complementary for Trajectory Representation Learning

Silin Zhou, Shuo Shang, Lisi Chen, Peng Han, Christian S. Jensen

TL;DR

A novel multimodal TRL method, dubbed GREEN, is proposed to jointly utilize Grid and Road trajectory Expressions for Effective representatioN learning and is compared with 7 state-of-the-art TRL methods for 3 downstream tasks, finding that GREEN consistently outperforms all baselines and improves the accuracy of the best-performing baseline.

Abstract

Trajectory representation learning (TRL) maps trajectories to vectors that can be used for many downstream tasks. Existing TRL methods use either grid trajectories, capturing movement in free space, or road trajectories, capturing movement in a road network, as input. We observe that the two types of trajectories are complementary, providing either region and location information or providing road structure and movement regularity. Therefore, we propose a novel multimodal TRL method, dubbed GREEN, to jointly utilize Grid and Road trajectory Expressions for Effective representatioN learning. In particular, we transform raw GPS trajectories into both grid and road trajectories and tailor two encoders to capture their respective information. To align the two encoders such that they complement each other, we adopt a contrastive loss to encourage them to produce similar embeddings for the same raw trajectory and design a mask language model (MLM) loss to use grid trajectories to help reconstruct masked road trajectories. To learn the final trajectory representation, a dual-modal interactor is used to fuse the outputs of the two encoders via cross-attention. We compare GREEN with 7 state-of-the-art TRL methods for 3 downstream tasks, finding that GREEN consistently outperforms all baselines and improves the accuracy of the best-performing baseline by an average of 15.99\%.

Grid and Road Expressions Are Complementary for Trajectory Representation Learning

TL;DR

A novel multimodal TRL method, dubbed GREEN, is proposed to jointly utilize Grid and Road trajectory Expressions for Effective representatioN learning and is compared with 7 state-of-the-art TRL methods for 3 downstream tasks, finding that GREEN consistently outperforms all baselines and improves the accuracy of the best-performing baseline.

Abstract

Trajectory representation learning (TRL) maps trajectories to vectors that can be used for many downstream tasks. Existing TRL methods use either grid trajectories, capturing movement in free space, or road trajectories, capturing movement in a road network, as input. We observe that the two types of trajectories are complementary, providing either region and location information or providing road structure and movement regularity. Therefore, we propose a novel multimodal TRL method, dubbed GREEN, to jointly utilize Grid and Road trajectory Expressions for Effective representatioN learning. In particular, we transform raw GPS trajectories into both grid and road trajectories and tailor two encoders to capture their respective information. To align the two encoders such that they complement each other, we adopt a contrastive loss to encourage them to produce similar embeddings for the same raw trajectory and design a mask language model (MLM) loss to use grid trajectories to help reconstruct masked road trajectories. To learn the final trajectory representation, a dual-modal interactor is used to fuse the outputs of the two encoders via cross-attention. We compare GREEN with 7 state-of-the-art TRL methods for 3 downstream tasks, finding that GREEN consistently outperforms all baselines and improves the accuracy of the best-performing baseline by an average of 15.99\%.

Paper Structure

This paper contains 22 sections, 15 equations, 8 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Definitions
  5. Problem Statement
  6. The GREEN Model
  7. Grid Encoder
  8. Road Encoder
  9. Dual-modal Interactor
  10. Loss Functions
  11. Training and Inference
  12. Experimental Evaluation
  13. Experiment Settings
  14. Main Results (RQ1)
  15. Micro Experiments
  16. ...and 7 more sections

Figures (8)

  • Figure 1: Comparison of three types of trajectories.
  • Figure 2: Overview of GREEN.
  • Figure 3: Comparison of grid and road mask recovery.
  • Figure 4: Comparison of the top-3 similar trajectories retrieved by GREEN and JGRM from Porto dataset for a query trajectory.
  • Figure 5: The effect of GREEN pre-training on Chengdu. X-axis is the ratio of the data used for training in the dataset.
  • ...and 3 more figures