This project is an end-to-end Toxic vs Non-Toxic Comment Classification System built using:

• Deep Learning (CNN / LSTM model)
• FastAPI for backend inference
• Streamlit for frontend UI
• Docker for containerized deployment

The system classifies any user comment into:

• 🟥 Toxic
• 🟩 Non-Toxic

User → Streamlit UI → FastAPI API → Toxicity Model → Prediction

• frontend/ – Streamlit user interface
• fastapp/ – FastAPI backend with model inference
• model/ – tokenizer + saved deep learning model
• docker-compose.yml – runs both apps together

Predicts:

• 0 → Non-Toxic
• 1 → Toxic

• /predict endpoint
• Returns prediction + confidence
• Handles tokenization, padding, inference

• Clean textbox for entering comments
• Color-coded output
• Real-time classification

• Run using docker-compose up
• Streamlit + FastAPI communicate internally

git clone <repo-url>
cd Toxic-Comment-Classifier

docker-compose up --build

• 🌐 Streamlit UI: http://localhost:8501
• 🧪 FastAPI Docs (Swagger): http://localhost:8000/docs

uvicorn fastapp.main:app --reload 

cd frontend
streamlit run frontend/frontend.py

This project uses a binary deep learning classifier.

• Text cleaning
• Tokenization
• Sequence padding
• Embedding layer
• CNN
• Dense layer with sigmoid output

• 0 → Non-Toxic
• 1 → Toxic

{
  "text": "You are so stupid!"
}

{
  "prediction": "Toxic",
  "confidence": 0.91
}

• Simple text input
• Classification result with confidence score
• User-friendly layout

• FastAPI
• Uvicorn
• TensorFlow / Keras
• Pickle

• Streamlit
• Requests

• Docker
• Docker Compose

• Add multi-label toxicity detection
• Add explanation (LIME / SHAP)