🛡️ Comment Moderation using OpenRouter + LLM

A Python CLI tool that reads user comments from a CSV or JSON file, uses a Generative AI model via OpenRouter to detect offensive or inappropriate content, and generates a clean report with visual insights.

Uses reliable and open-source nvidia/llama-3.1-nemotron-nano-8b-v1:free model from OpenRouter.

---

✨ Features

• 🧠 LLM-powered moderation using OpenRouter
• 📁 Supports .csv and .json input
• 🚫 Fully LLM-based (no profanity filter)
• 🧹 Auto-fixes messy JSON responses
• 📝 Generates a detailed summary report
• 📊 Produces a pie chart of offense types
• 🧃 CLI support for flexible input/output

---

🚀 Installation

git clone https://github.com/tars-06/commentMOD.git
cd comment-moderation
pip install -r requirements.txt

Requirements (requirements.txt)

requests
matplotlib
python-dotenv

---

🔑 Setup

1. Create a .env file in the project root:

   OPENROUTER_API_KEY=sk-your-openrouter-api-key
   

2. Enable prompt training on your OpenRouter account:
   👉 https://openrouter.ai/settings/privacy

---

📂 Input Format

CSV (comments.csv)

comment_id,username,comment_text
1,alice,"I love the way you explained it!"
2,bob,"You're so dumb it's painful."
...

JSON (comments.json)

[
  {
    "comment_id": 1,
    "username": "alice",
    "comment_text": "I love the way you explained it!"
  },
  ...
]

---

🧪 Usage

Basic run:

python moderate_comments.py comments.csv

With JSON input:

python moderate_comments.py comments.json

With custom output folder:

python moderate_comments.py comments.csv --output_dir outputs/

---

📦 Output Files

File	Description
moderated_comments.csv	All comments + moderation results
moderation_report.txt	Summary report + top 5 most offensive comments
offense_type_pie_chart.png	Pie chart of offensive comment types

---

🔎 What the Model Returns

Each comment is enriched with the following fields:

{
  "comment_id": 42,
  "is_offensive": true,
  "offense_type": "harassment",
  "explanation": "Contains threatening and derogatory language."
}

---

🛠 Troubleshooting

• 401 Unauthorized
  → Check if your API key is valid and loaded via .env

• 404 from OpenRouter
  → Go to https://openrouter.ai/settings/privacy and enable "Allow prompt training"

• JSONDecodeError
  → Script automatically sanitizes bad output or skips malformed batches.

---

🧯 Notes

• Uses batch prompting (10 comments per API call)
• Automatically handles smart quotes, bad punctuation, and broken JSON
• Purely LLM-driven — no keyword-based filtering

---

📜 License

MIT © Aaditya Saraf