$\text {MA}^{2}\text {Traj} $: Diffusion network with multi-attribute aggregation for trajectory generation

期刊论文
发表日期：

2025年5月8日

查看全文

Xingyu Zhao,
Xiao Zhang,
Bohan Zhang,
Jianpeng Qi,
Junyu Dong,
Yanwei Yu

摘要

With the rapid development of GPS devices and mobile internet technology, trajectory data has generated immense value but also raised concerns about privacy leakage. Existing trajectory generation models utilize users’ trip attributes to provide auxiliary associations for GPS trajectories, yet they overlook the spatial constraints of users’ trip patterns. Additionally, matching the road network with trajectories results in time consumption. To address these issues, we propose a diffusion network with multi-attribute aggregation for trajectory generation, named $\varvec{\text {MA}^{2}\text {Traj}}$ ${MA}^{2} Traj$ . Specifically, we introduce origin-destination information to provide spatial constraints for trip attributes. Meanwhile, we design a multi-attribute aggregation module, which integrates origin-destination information and trip attributes (such as trip distance and average speed) to effectively capture user trip patterns and spatial distribution, thereby enhancing the reliability of trajectory generation. Experimental results show that our model achieves a 4.8% performance improvement, significantly enhancing the accuracy of trajectory generation.

\(\text {MA}^{2}\text {Traj} \): Diffusion network with multi-attribute aggregation for trajectory generation

摘要