Detoxify — YouTube Comment Toxicity Analyzer

Detoxify is a full-stack web application that analyzes YouTube video comments for toxicity, sexism, and offensive language using a custom-trained logistic regression model. It provides a beautiful, interactive dashboard to help users identify and report harmful content.

Features

• Paste any YouTube video URL and analyze the first 100 comments
• Custom AI-powered toxicity detection using a logistic regression model trained on your own dataset (see backend/model_train.py)
• Safety Analysis Dashboard — shows safe/flagged counts, severity breakdown (High/Medium/Low), and animated toxicity bar
• Filter tabs — view All, Safe, Flagged, or High Severity comments
• Severity badges on each comment card (color-coded: red for high, orange for medium, purple for low, green for safe)
• PDF Report Generation — download a branded evidence report with video info, summary stats, and a table of all flagged comments (username + comment + severity)
• Video info display (title, channel, views, likes, comment count)
• Modern glassmorphism UI with soft pinks, lavender & purple palette

Project Structure

Detoxify/
│   README.md
│   requirements.txt
│
├── backend/
│   ├── app.py              # Flask backend, serves API and frontend
│   ├── model_train.py      # Script to train and save the logistic regression model
│   ├── logreg_model.joblib # Saved model (generated by model_train.py)
│   └── vectorizer.joblib   # Saved vectorizer (generated by model_train.py)
│
└── frontend/
	├── index.html          # Main UI
	├── script.js           # Frontend logic (fetches API, renders dashboard)
	└── style.css           # Styles (glassmorphism, gradients, etc.)

How It Works

1. Frontend: User pastes a YouTube video URL and clicks Analyze. The UI sends a request to the backend.
2. Backend:
   • Extracts the video ID and fetches comments using the YouTube Data API.
   • Each comment is vectorized and classified using the custom logistic regression model (see backend/model_train.py).
   • Returns analysis results (flagged/safe, severity, confidence) and video info to the frontend.
3. Frontend: Displays results in a dashboard, allows filtering, and can generate a PDF report.

Setup Instructions

1. Get a YouTube API Key (Free)

1. Go to Google Cloud Console
2. Create a new project (or use an existing one)
3. Enable YouTube Data API v3 from the API Library
4. Go to Credentials → Create Credentials → API Key
5. Copy the API key

2. Configure the API Key

Create a .env file in the project root (next to README.md) and add:

YOUTUBE_API_KEY=your_actual_api_key_here

3. Install Dependencies

pip install -r requirements.txt

4. Train the Model (Optional: only if you want to retrain)

• Place your labeled dataset (CSV) in a known location and update the path in backend/model_train.py.
• Run:

cd backend
python model_train.py

This will generate logreg_model.joblib and vectorizer.joblib in the backend folder.

5. Run the Backend (serves both API and frontend)

cd backend
python app.py

Visit http://127.0.0.1:5000 in your browser.

Requirements

• Python 3.8+
• Flask, Flask-CORS, google-api-python-client, python-dotenv, scikit-learn, pandas, joblib
• (optionally transformers/torch if you want to use the old model)

Custom Model

• The backend uses a logistic regression model trained on your own dataset (see backend/model_train.py).
• You can retrain or swap the model by editing and rerunning model_train.py.
• The model and vectorizer are loaded at backend startup and used for all comment analysis.
• [Add here: Briefly describe your dataset and what the labels mean, e.g., 1=sexist, 0=safe, etc.]

Credits

• Built for Tink-Her-Hack hackathon.
• UI/UX: Modern glassmorphism, animated gradients, and PDF export.
• [Add here: Any additional contributors, dataset sources, or acknowledgments]

---

For questions or contributions, open an issue or pull request! The server starts at http://127.0.0.1:5000

Note: The AI model (~500MB) downloads automatically on first run. Subsequent starts are instant.

5. Open the Frontend

Open frontend/index.html in your browser (just double-click it, or use Live Server in VS Code).

Project Structure

Tink-Her-Hack/
├── backend/
│   └── app.py              # Flask API server + AI classifier
├── frontend/
│   ├── index.html           # Main page with dashboard & filters
│   ├── style.css            # Glassmorphism UI (women-friendly palette)
│   └── script.js            # Frontend logic + PDF generation
├── .env                     # Your API key (not committed)
├── .env.example             # Template for API key
├── .gitignore
├── requirements.txt
└── README.md

Tech Stack

• Backend: Python, Flask, YouTube Data API v3
• AI Model: cardiffnlp/twitter-roberta-base-offensive (HuggingFace Transformers, PyTorch)
• Frontend: HTML, CSS, JavaScript (no frameworks)
• PDF Generation: jsPDF + jsPDF-AutoTable (client-side)
• API: Google YouTube Data API (free tier — 10,000 units/day)