🛑 YouTube Comment Toxicity Detector

A YouTube Comment Toxicity Detector built using TensorFlow, NLP, and Gradio that allows users to check the toxicity of individual comments or analyze the toxicity of comments from any YouTube video.

---

🚀 Features

• Custom Comment Checker: Enter any text and check its toxicity.
• YouTube Comment Analyzer: Fetch comments from a YouTube video and analyze toxicity.
• Multi-Label Classification: Detects multiple types of toxicity, including:
  • Toxic
  • Severe Toxic
  • Obscene
  • Threat
  • Insult
  • Identity Hate
• Interactive UI with Gradio
• Color-Coded Output: 🚨 Red for toxic labels, ✅ Green for non-toxic labels.

---

🖥️ Tech Stack

• Python
• TensorFlow / Keras (Deep Learning Model)
• NLP (Natural Language Processing)
• Pandas & NumPy (Data Handling)
• Google YouTube API (Fetching YouTube Comments)
• Gradio (User Interface)

---

🛠️ Installation & Setup

1️⃣ Clone the Repository

git clone https://github.com/Codebreaker42/comment-toxicity.git
cd youtube-toxicity-detector

#️⃣ Create a Virtual Environment (Recommended)

python -m venv your_virtual_env_name
source your_virtual_env_name/bin/activate  # On Mac/Linux
your_virtual_env_name\Scripts\activate     # On Windows

3️⃣ Install Dependencies

pip install -r requirements.txt

4️⃣ Download the Pre-trained Model & Tokenizer

Ensure you have the required toxicity.keras model and vectorizer.pkl file in the comment_checkpoint/ directory.

⚙️ Running the Application

🔹 Start the Gradio App

python app.py

🔹 Access the Interface Once running, open the URL (shown in the terminal) in your browser.

🎯 Usage Guide

✅ Check Custom Comment

• Enter a comment in the Text Box.
• Click Analyze.
• See toxicity labels with color indicators
• 🔴 Red → Toxic
• ✅ Green → Not Toxic

✅ Analyze YouTube Comments

• Paste the YouTube Video URL. Click Analyze.
• View a table of comments with toxicity analysis.
• View a table of comments with toxicity analysis.

📸 Screenshots

💡 How It Works

1.Preprocessing:

• Text is tokenized using a saved vectorizer (vectorizer.pkl).
• The input is padded to MAX_LEN = 1800.

Model Prediction:

• A deep learning model (toxicity.keras) predicts six toxicity labels.

YouTube Comments:

• Uses Google YouTube API to fetch comments.
• Each comment is analyzed for toxicity.

🔧 Troubleshooting

Issue: Model Not Found

• Ensure toxicity.keras and vectorizer.pkl exist in the comment_checkpoint/ directory.

Issue: YouTube API Key Error

• Replace the API key in app.py with your own Google API key.

📜 License

This project is open-source and available under the MIT License.

🤝 Connect with Me

• 💼 GitHub: https://github.com/Codebreaker42
• 📧 Email: nitinbdkt777@gmail.com

Name

🚀 Star this repo if you find it useful! ⭐