Abstraction of Vessel Trajectories with LLMs

A recent study published on arXiv proposes a framework for abstracting vessel trajectories, combining AIS (Automatic Identification System) data with large language models (LLMs). The goal is to transform raw and often noisy data sequences into structured and semantically enriched representations, easily interpretable by humans and usable by machine reasoning systems.

Contextual Enrichment

The proposed framework segments trajectories into distinct trips, composed of episodes annotated with mobility information. Each episode is further enriched with contextual information from various sources, such as nearby geographic entities, offshore navigation features, and weather conditions. This enrichment allows for the generation of controlled natural language descriptions using LLMs.

Applications and Benefits

The research empirically examines the quality of descriptions generated by different LLMs using AIS data and open contextual features. Increasing semantic density and reducing spatiotemporal complexity facilitates downstream analytics and integration with LLMs for higher-level maritime reasoning tasks.