AI-Driven Analysis of Crash Narratives: Evaluating Open-Source LLMs for Structured Information Extraction
The project aim is to develop and evaluate an open-source Large Language Model
(LLM), trained on the NHTSA’s Crash Investigation Sampling System (CISS)
dataset, capable of extracting missing details from textual crash reports.
The primary objective is to assess the effectiveness of LLMs in understanding
and interpreting accident narratives, enabling them to answer both binary and
categorical questions about crash dynamics.