Generation of Synthetic Trajectory Microdata from Language Models

Releasing and sharing mobility data, and specifically trajectories, is necessary for many applications, from infrastructure planning to epidemiology. Yet, trajectories are highly sensitive data, because the points visited by an individual can be identifying and also confidential. Hence, trajectories must be anonymized before releasing or sharing them. While most contributions to the trajectory anonymization literature take statistical approaches, deep learning is increasingly being used. We observe that natural language sentences and trajectories share a sequential nature that can be exploited in similar ways. In this paper, we present preliminary work on generating synthetic trajectories using machine learning models typically used for natural language processing. Our empirical results attest to the quality of the generated synthetic trajectories. Furthermore, our methods allow discovering natural neighborhoods based on trajectories.