Cyberbullying Detection Discord Bot

A Discord bot that automatically detects and moderates cyberbullying messages in real-time using Machine Learning (SVM classifier with TF-IDF vectorization) and a curated suspicious words list.

Features

• Real-time message analysis — Every message is scanned for cyberbullying content
• Dual detection system — Combines a suspicious words list with a trained SVM model for accurate detection
• Auto-moderation — Deletes offensive messages automatically
• User tracking — Tracks offenders using SQLite, flags users after 10+ violations, and auto-kicks after 20+
• Logging channel — Dedicated channel for admins to see all violations with detailed embeds
• Offense history — Full history of each user's violations stored in database
• Comprehensive admin commands — Check users, view history, reset data, list flagged users, and more

How It Works

1. A message is sent in the server
2. The bot preprocesses the text (removes URLs, mentions, normalizes whitespace)
3. Checks the message against a list of known suspicious/offensive words
4. If no match, the message is vectorized using TF-IDF and classified by the SVM model
5. If cyberbullying is detected:
   • The message is deleted
   • The user is warned
   • The incident is logged to the database and log channel
   • The user's offense count is incremented
6. After exceeding thresholds, users are flagged (10+) or kicked (20+) automatically

Project Structure

├── bot.py                  # Main bot script
├── bot.log                 # Bot activity log (auto-generated)
├── discord_bully.db        # SQLite database (auto-generated)
├── server_config.json      # Server configurations (auto-generated)
├── data/
│   ├── cyberbullying.csv   # Training dataset
│   ├── CyberTrollIEEE.csv  # Training dataset (IEEE)
│   └── sus_words.csv       # Curated list of suspicious/offensive words
├── ipynb/
│   └── main.ipynb          # Jupyter notebook for model training & analysis
├── models/                 # Pre-trained models (not included in repo)
│   ├── svc_model.pkl       # Trained SVM classifier
│   └── tfidf_vectorizer.pkl # Fitted TF-IDF vectorizer
├── requirements.txt        # Python dependencies
├── .env                    # Environment variables (not included in repo)
└── LICENSE

Setup

1. Clone the repository

git clone https://github.com/<your-username>/<repo-name>.git
cd <repo-name>

2. Download the pre-trained models

The models/ folder is not included in this repository. Download it from Google Drive:

Download Models

Place the downloaded files (svc_model.pkl and tfidf_vectorizer.pkl) inside a models/ directory at the project root.

3. Install dependencies

pip install -r requirements.txt

4. Configure environment variables

Create a .env file in the project root:

DISCORD_BOT_TOKEN=your_discord_bot_token_here

5. Run the bot

python bot.py

Requirements

• Python 3.8+
• A Discord Bot Token (from the Discord Developer Portal)
• CUDA-compatible GPU (optional, for model training)

Bot Commands

All moderation commands require Administrator permission.

Command	Description
!help_mod	Show all moderation commands
!check_user <username>	Check a user's offense record
!user_history <username> [limit]	View a user's recent offenses (default: 5)
!reset_user <username>	Reset a user's offense data
!flagged_users	List all flagged users
!top_offenders [limit]	Show top offenders (default: 10)
!set_log_channel [#channel]	Set the logging channel for violations
!bot_stats	Show bot statistics

Logging Channel

Set up a dedicated logging channel to receive detailed reports of all detected violations:

1. Create a private channel for moderators (e.g., #mod-logs)
2. Run !set_log_channel #mod-logs in that channel
3. The bot will now send detailed embeds for every:
   • Cyberbullying detection (with message content, detection method, user info)
   • User kick
   • User reset

Configuration

The following values can be adjusted in bot.py:

FLAG_THRESHOLD = 10   # Offenses before user is flagged
KICK_THRESHOLD = 20   # Offenses before user is auto-kicked

Database Schema

user_data

Column	Type	Description
username	TEXT	Primary key, Discord username
user_id	INTEGER	Discord user ID
count	INTEGER	Total offense count
flag	INTEGER	1 if flagged, 0 otherwise
kicked	INTEGER	1 if kicked, 0 otherwise
last_offense	TEXT	ISO timestamp of last offense

offense_log

Column	Type	Description
id	INTEGER	Auto-increment primary key
user_id	INTEGER	Discord user ID
username	TEXT	Discord username
guild_id	INTEGER	Server ID
channel_id	INTEGER	Channel ID
message_content	TEXT	The offensive message (truncated to 500 chars)
detection_method	TEXT	"word_list" or "ml_model"
timestamp	TEXT	ISO timestamp

Datasets

The model was trained on the following datasets:

• CyberTrollIEEE.csv — IEEE cybertroll dataset with content and annotation columns (1 = cyberbullying, 0 = safe)
• cyberbullying.csv — Additional cyberbullying dataset

Tech Stack

• Discord API — discord.py
• ML Model — SVM (Support Vector Machine) via scikit-learn
• Feature Extraction — TF-IDF Vectorizer
• Database — SQLite
• Training — PyTorch, Transformers, CUDA (GPU-accelerated)

License

This project is licensed under the MIT License. See the LICENSE file for details.