Cyber-Bullying Tweet Detector

A Streamlit web app for detecting cyberbullying in tweets using a trained XGBoost classifier.

The project preprocesses tweet text, transforms it with a TF-IDF vectorizer, and predicts one of the following categories:

• gender (gender-based bullying)
• religion (religion-based bullying)
• age (age-based bullying)
• ethnicity (ethnicity-based bullying)
• other_cyberbullying (other bullying content)
• not_cyberbullying (not cyberbullying)

Features

• Clean and normalize tweet text
• Remove URLs, mentions, hashtags, punctuation, numbers, and stop words
• Lemmatize tokens using NLTK
• Predict cyberbullying category with model confidence visualization
• Display model metadata and metrics

Model Information

• Model: XGBoost Classifier
• Accuracy: 83.52%
• F1 Macro: 0.8316

These values are loaded from model_metadata.json.

Live Demo

Getting Started

Prerequisites

• Python 3.10+ (recommended)
• pip package manager

Install dependencies

pip install -r requirements.txt

Run the app

streamlit run app.py

Then open the local Streamlit URL shown in the terminal.

Repository Files

• app.py - Streamlit application entry point
• requirements.txt - Python dependencies
• best_model.pkl - serialized trained model
• tfidf_vectorizer.pkl - serialized TF-IDF vectorizer
• label_encoder.pkl - serialized label encoder
• model_metadata.json - metadata for the trained model

Notes

• The app downloads NLTK resources for stop words and lemmatization at startup.
• If the NLTK data is not already installed, the app will fetch it automatically.

License

This repository does not include a license file. Add one if you want to publish or share the project publicly.