Incomplete mobility trajectories are a persistent problem: gaps in the data can hide important travel events and their endpoints. This paper asks whether compact, open-weight large language models—small enough to run locally—can recover those missing events from partial trajectory data.
The authors design a controlled evaluation around air trips, using incomplete trajectories to test how well the models infer what is missing. The emphasis on locally deployed models points to practical scenarios where sending sensitive movement data to external APIs is undesirable or infeasible.
The abstract does not disclose final results, so it remains an open question whether these compact models succeed at the task. What the paper offers is a framework for measuring that capability, and a concrete test bed for future work on privacy-preserving mobility analysis.