DigestAI news desk
Research updated

Retrieval‑Augmented LLM Boosts Intersection Safety Recommendations from Crash Narratives

A new study demonstrates how large language models can turn unstructured crash narratives into concrete, site‑specific safety countermeasures. The researchers built a retrieval‑augmented generation (RAG) pipeline that extracts key mechanism attributes—such as traffic control, signal indication, driver fault, vehicle movement, and travel direction—from accident reports and links them to…

1 source primary source

Key points

  • RAG framework extracts crash mechanism attributes and maps them to FHWA‑validated countermeasures
  • Evaluated on 312 crashes at 115 Florida intersections, achieving 0.82 precision, 0.85 recall
  • Average of 3.91 recommended countermeasures per site, closely matching real‑world average of 3.86

The framework was tested on 312 fatal and serious‑injury crashes across 115 intersections in Lake and Sumter Counties, Florida, using five‑fold cross‑validation. It achieved a precision of 0.82, recall of 0.85, and an F1‑score of 0.82, recommending an average of 3.91 countermeasures per site with 3.14 matching the actual average of 3.86. The results suggest that retrieval‑augmented LLMs can provide interpretable, scalable decision‑support for transportation agencies, reducing reliance on scarce expert judgment.

By automating the translation of narrative crash data into actionable safety measures, the approach promises faster, more consistent interventions that could lower intersection‑related injuries and fatalities nationwide.

Read the original at arXiv cs.CL · by Abu Saif Md Nasim Uddin, Mohamed Abdel-Aty, Zubayer Islam, Parvez Anowar, Chenzhu Wang primary source Open source ↗
Topics · follow one to build your own front page

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us.

Comments

via GitHub Discussions

More in Research

All →

Related stories