Topic Models from Crash Narrative Reports of Motorcycle Crash Causation Study
DOI: 10.1177/03611981211002523
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This study addresses the lack of in-depth analysis of unstructured textual data within the Motorcycle Crash Causation Study (MCCS), a large-scale matched case-control dataset sponsored by the National Highway Traffic Safety Administration. While previous research has examined structured variables in MCCS, such as rider demographics and vehicle conditions, the detailed crash narrative reports have remained largely unexplored. The authors aim to fill this gap by applying natural language processing (NLP) tools, specifically text mining and topic modeling, to extract insights from these narratives. The research seeks to answer three questions: whether MCCS data is representative of national statistics, what key insights emerge from the narratives, and how these insights can improve motorcycle safety. The methodology involved collecting crash narrative data for 351 injury crashes from the MCCS database. To assess representativeness, the authors compared key variable distributions in MCCS against the General Estimates System (GES), a nationally representative sample of police-reported crashes. For the textual analysis, the researchers manually converted PDF narratives into structured datasets, categorizing information into human, vehicle, environmental, and injury details. They performed data cleaning, including stop-word removal, lemmatization, and the removal of rare words. The study employed Rapid Automatic Keyword Extraction (RAKE) to identify frequent keywords and co-occurrence analysis to map term relationships. Finally, Latent Dirichlet Allocation (LDA) topic modeling was applied to separate corpora of fatal (40 crashes) and non-fatal (311 crashes) narratives, visualized using interactive multidimensional scaling and word clouds. The comparison with GES data revealed that MCCS is somewhat representative but exhibits specific biases: alcohol impairment, cloudy weather, younger riders, and certain roadway types are overrepresented. Text mining identified frequent keywords related to directional movement (e.g., "left," "right"), vehicle conditions, and road infrastructure. Co-occurrence analysis highlighted clusters linking injury terms with roadway types (e.g., "arterial," "surface") and rider demographics with motorcycle identifiers. The LDA topic models generated distinct themes for fatal and non-fatal crashes. In fatal crash narratives, topics frequently featured terms like "old," "failed," and "damage," suggesting a focus on rider age, mechanical failure, and severe vehicle impact. Non-fatal narratives showed different thematic distributions, allowing for a differentiated understanding of crash causation based on injury severity. The significance of this work lies in its demonstration that unstructured crash narratives contain valuable, hidden insights that complement structured data. By clustering fatal and non-fatal narratives separately, the study provides a framework for understanding the specific causation mechanisms associated with different injury levels. These findings contribute to ongoing MCCS research by offering a deeper, text-based perspective on crash dynamics, which can inform targeted safety interventions and countermeasures for motorcycle crashes.
Key finding
Applying topic modeling to MCCS crash narratives successfully identified distinct clusters of contributing factors related to rider age, vehicle damage, and roadway characteristics for both fatal and non-fatal crashes.
Methodology
dataset
Sample size: 351
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via author_sweep_intake on 2026-05-28.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-28 |
| archive | success | canonical_url | — | — | 7 | 2026-06-06 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | success | — | — | — | 1 | 2026-05-28 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 15 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
- causation analyses
- motorcycle crash typology
- in depth crash investigation
- vru crash typology
- pre crash contributing factors
- motorcyclist skill
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Empirical Findings: crash risk outcomes
- Methodological Resource: dataset resource