Imitative transportation trajectory generation with urban knowledge-enhancement

Abstract

Designing effective tools for modeling and analyzing transportation trajectories help improve various smart city services, e.g., location prediction, business recommendation and traffic forecasting. Yet with limited historical travel data, conventional tools are faced with challenges in representing the sophisticated contextual relationships inherent in transportation trajectories, while such trajectories have strong functional, geographical, and time-specific characteristics. In this thesis, I focus on the task of urban trajectory generation with multi-source urban knowledge enhancement, where I propose a unified framework for knowledge aggregation, and keep the spatiotemporal characteristics to enable accurate and effective trajectory analysis. Specifically, I start by introducing the challenge of multiple data analysis for traffic modeling, then investigate the relevant research studies on knowledge fusion. Based on the previous research, I build a large-scale urban knowledge graph by combining vast data from multiple sources. Then I begin to study the effective knowledge graph representation learning methods under high temporal-dependent relations. After that, I further propose a knowledge-enhanced model for spatial and temporal trajectory analysis. Finally, in addition to cross-domain knowledge, I explore the possibility of utilizing meta knowledge to facilitate efficient generalization across tasks. Overall, I present a knowledge-fusion system that can deal with multiple urban knowledge for imitative transportation trajectory generation. Extensive simulations on real-world data validate the effectiveness of our proposed system. In the end, I conclude this thesis and suggest several potential research opportunities for further study.

Date of Award	2022
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Hao LIU (Supervisor) & Yize CHEN (Supervisor)