Harnessing Natural Language Processing for Disaster Response and Crisis Management

This repository contains the complete source code, model configurations, and documentation for the study
"Harnessing Natural Language Processing for Disaster Response and Crisis Management"
submitted to the National High School Journal of Science (NHSJS), 2025.

---

📖 Overview

This project demonstrates how Natural Language Processing (NLP) can be applied to classify disaster-related tweets by urgency.
It includes:

• Fine-tuning BERT for binary urgency classification (urgent vs not urgent).
• Evaluation using classification metrics, confusion matrices, and ROC curves.
• Sample data to illustrate the expected schema; full datasets can be re-scraped or provided upon request.

All code is lightweight and designed for reproducibility and review.

---

📁 Repository Structure

NLP-Disaster-Response/ ├── data/ # Sample CSV data and preprocessing scripts ├── models/ # Model training scripts (BERT and Naive Bayes) ├── evaluation/ # Model evaluation and ROC curve generation ├── figures/ # Publication figures (ROC curves, tables, etc.) ├── results/ # Output metrics and model checkpoints └── requirements.txt # Python dependencies

---

⚙️ Reproduction Instructions

Clone the repository

git clone https://github.com/AbhineethDuddela/NLP-Disaster-Response.git
cd NLP-Disaster-Response
pip install -r requirements.txt
python models/train_bert.py
python evaluation/evaluate.py

## NOTE
Original tweets are not included due to Twitter policy. Scripts to re-scrape and annotate are provided.

The sample_tweets.csv file illustrates the CSV format expected by the training and evaluation scripts.

To reproduce figures from the paper, run python evaluation/evaluate.py.

The repository is structured to make it easy for reviewers and other researchers to reproduce the results.