README.md

metadata

base_model: bert-base-multilingual-cased
tags:
  - persian-nlp
  - text-classification
  - traffic-crash-detection
  - crash-type-classification
  - bert
  - information-extraction
license: apache-2.0
language:
  - fa
pipeline_tag: text-classification
inference: false

BERT-Crashtype-Classification

Fine-tuned BERT for classifying Persian social media texts into 9 crash types.

📄 Paper: Extracting traffic crash information from social media: an LLM-based approach – Transportation Letters (2026)

---

🎯 What it does

Classifies a given Persian social media text into one of 9 types of traffic crashes.

🏷️ Crash Types (9 Classes)

1. vehicle with two-wheeled vehicle
2. Two-wheeled vehicle–pedestrian
3. vehicle with fixed object or ran off road
4. rollover or fall
5. multiple car
6. vehicle–animal
7. vehicle–pedestrian
8. vehicle with single other vehicle
9. two-wheeled vehicle with two-wheeled vehicle

⚙️ Fine-tuning

• Base Model: bert-base-multilingual-cased
• Data: Proprietary Persian social media crash dataset (Damavand County, Iran)

📊 Performance

Task	Metric	Score
Crash Type Classification (9 classes)	Accuracy	89.7%

🚀 Quick Start

from transformers import AutoModelForSequenceClassification, AutoTokenizer

model = AutoModelForSequenceClassification.from_pretrained("crash-information-extraction/BERT-Crashtype-Classification")
tokenizer = AutoTokenizer.from_pretrained("crash-information-extraction/BERT-Crashtype-Classification")

text = "تصادف دو خودرو در اتوبان"
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)
predicted_class = outputs.logits.argmax().item()
# predicted_class is an integer from 0 to 8