GTR: A General, Multi-View, and Dynamic Framework for Trajectory Representation Learning

Xiangheng Wang, Ziquan Fang, Chenglong Huang, Danlei Hu, Lu Chen, Yunjun Gao

Abstract: Trajectory representation learning aims to transform raw trajectory data into compact and low-dimensional vectors that are suitable for downstream analysis. However, most existing methods adopt either a free-space view or a road-network view during the learning process, which limits their ability to capture the complex, multi-view spatiotemporal features inherent in trajectory data. Moreover, these approaches rely on task-specific model training, restricting their generalizability and effectiveness for diverse analysis tasks. To this end, we propose GTR, a general, multi-view, and dynamic Trajectory Representation framework built on a pre-train and fine-tune architecture. Specifically, GTR introduces a multi-view encoder that captures the intrinsic multi-view spatiotemporal features. Based on the pre-train and fine-tune architecture, we provide the spatio-temporal fusion pre-training with a spatio-temporal mixture of experts to dynamically combine spatial and temporal features, enabling seamless adaptation to diverse trajectory analysis tasks. Furthermore, we propose an online frozen-hot updating strategy to efficiently update the representation model, accommodating the dynamic nature of trajectory data. Extensive experiments on two real-world datasets demonstrate that GTR consistently outperforms 15 state-of-the-art methods across 6 mainstream trajectory analysis tasks. All source code and data are available at https://github.com/ZJU-DAILY/GTR.

Lay Summary:

Smartphones and GPS devices constantly record our movements, generating massive amounts of trajectory data—sequences of location points that reveal how people and vehicles move through cities. This data holds great potential, helping urban planners optimize traffic, navigation apps suggest faster routes, and researchers study mobility patterns. However, current methods analyze trajectories from just a single perspective—either raw GPS points or road maps—missing the bigger picture. They also require retraining for each new task, making them inefficient for real-life development.

We present GTR, a unified and more flexible way to process trajectory data. GTR combines multiple perspectives, looking at both free movement (like GPS traces) and road networks to better understand travel patterns. It’s designed as a general-purpose tool, meaning it can adapt to different tasks—such as predicting travel times, finding similar routes, or even trajectory data generation, without needing full retraining each time. Additionally, GTR keeps improving over time, automatically adjusting to new movement data, which is crucial for real-world applications where traffic and travel behaviors constantly change. In evaluations, GTR outperforms 15 SOTA methods across 6 common trajectory analysis tasks, proving its versatility and superiority.

Link To Code: https://github.com/ZJU-DAILY/GTR

Primary Area: Deep Learning->Other Representation Learning

Keywords: Trajectory representation learning, mobility learning, spatio-temporal learning

Submission Number: 15229