🚨 Toxic Comment Detection (Binary Classification)

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

---

📌 Project Architecture

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

Components:

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

---

🚀 Features

✔ Binary Toxic Comment Classification

Predicts:

• 0 → Non-Toxic
• 1 → Toxic

✔ FastAPI Backend

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

✔ Streamlit Frontend

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

✔ Dockerized Microservices

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

---

🔧 How to Run the Project

1️⃣ Clone Repository

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

---

2️⃣ Run With Docker (Recommended)

docker-compose up --build

Access:

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

---

3️⃣ Run Without Docker (Local Mode)

Start FastAPI:

uvicorn fastapp.main:app --reload 

Start Streamlit:

cd frontend
streamlit run frontend/frontend.py

---

🧠 Model Details

This project uses a binary deep learning classifier.

Pipeline:

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

Output:

• 0 → Non-Toxic
• 1 → Toxic

---

📡 API Example

Request:

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

Response:

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

---

🎨 Streamlit UI

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

---

📦 Tech Stack

Backend:

• FastAPI
• Uvicorn
• TensorFlow / Keras
• Pickle

Frontend:

• Streamlit
• Requests

DevOps:

• Docker
• Docker Compose

---

🛠 Future Enhancements

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

---