Context-Enriched Natural Language Descriptions of Vessel Trajectories

From Vessel Trajectories to Context-Enriched Natural Language Trip Descriptions

Kostas Patroumpas Alexandros Troupiotis-Kapeliaris Giannis Spiliopoulos Panagiotis Betchavas Dimitrios Skoutas Dimitris Zissis Nikos Bikakis

Archimedes/Athena RC · University of the Aegean · Athena RC · Hellenic Mediterranean University

Abstract

This paper addresses the gap between raw AIS vessel data and the structured representations needed for human interpretation and machine reasoning. The proposed framework segments noisy trajectories into trips and mobility episodes, then enriches each episode with contextual information from geographic, maritime, and meteorological sources.

The enriched representations are used as controlled inputs to LLMs, which generate natural-language descriptions of vessel trips. The paper studies whether these descriptions remain relevant, faithful, and correct, and shows how semantic trajectories can support maritime analytics, voyage reports, anomaly analysis, and trajectory reasoning.

From AIS Points to Trip Narratives

1

Annotate mobility

Detect stops, gaps, speed changes, slow motion, turns, and noisy positions in raw AIS streams.

2

Segment trips

Split long trajectories into trips and episodes such as sailing, stopped, turning, maneuvering, and communication gap.

3

Add context

Enrich each episode with ports, coastal features, offshore areas, traffic schemes, weather, and bathymetry.

4

Generate descriptions

Pass context-enriched JSON to LLMs to produce human-readable trip narratives and summary statistics.

BibTeX

@inproceedings{ptsb2026,
  title  = {Context-Enriched Natural Language Descriptions of Vessel Trajectories},
  author = {Patroumpas, Kostas and Troupiotis-Kapeliaris, Alexandros and Spiliopoulos, Giannis and Betchavas, Panagiotis and Skoutas, Dimitrios and Zissis, Dimitris and Bikakis, Nikos},
  booktitle = {IEEE International Conference on Mobile Data Management (MDM)},
  year   = {2026}
}