Comment Classifier

A machine learning project that automatically identifies spam and toxic comments from YouTube using natural language processing and pre-trained transformer models.

Overview

This project analyzes YouTube comments to classify them as spam or legitimate content by combining:

• Toxicity detection using a pre-trained transformer model
• Pattern matching for common spam indicators
• Text preprocessing and cleaning techniques

Features

• Automated Text Cleaning: Removes URLs, mentions, hashtags, and special characters
• Dual Classification Approach:
  • Toxicity scoring using martin-ha/toxic-comment-model
  • Pattern-based spam detection
• Comprehensive Analysis: Statistical reporting and visualization
• Flexible Thresholds: Customizable spam detection sensitivity

Installation

Requirements

pip install transformers torch torchvision
pip install scikit-learn pandas numpy matplotlib seaborn
pip install nltk wordcloud textblob

Setup

1. Clone this repository
2. Install the required packages
3. Prepare your YouTube comments dataset in CSV format
4. Run the Jupyter notebook

Usage

1. Data Preparation

Ensure your CSV file contains a CONTENT column with the comment text:

CONTENT,category
"Great video! Thanks for sharing",ham
"Subscribe to my channel! Check out www.spam.com",spam

2. Running the Classifier

# Load and preprocess data
df = pd.read_csv('Youtube-Spam-Dataset.csv')
df['CLEAN_CONTENT'] = df['CONTENT'].apply(clean_text)

# Initialize the classifier
classifier = pipeline('text-classification', model='martin-ha/toxic-comment-model')

# Generate spam labels
df['CLASS_LABEL'] = create_spam_labels(df)

3. View Results

The classifier outputs:

• CLASS_LABEL: Binary classification (0 = Clean, 1 = Spam)
• Statistical summary of spam detection rates
• Visualization charts showing comment distribution

Classification Logic

The spam detection algorithm considers a comment as spam if:

1. High Pattern Match: 2+ spam patterns detected
2. Medium Pattern + Toxicity: 1+ patterns AND toxicity score > 0.3
3. High Toxicity: Toxicity score > 0.7

Spam Patterns Detected:

• subscribe, channel, check out, follow me
• my channel, visit, website
• URLs (.com, www)

Text Preprocessing

The clean_text() function performs:

• Converts to lowercase
• Removes URLs (http://, www., https://)
• Removes mentions (@username) and hashtags (#tag)
• Strips special characters and extra whitespace
• Handles missing/null values

Model Performance

The classifier uses the pre-trained martin-ha/toxic-comment-model which provides:

• Toxicity probability scores
• Binary toxic/non-toxic classification
• Robust performance on social media content

Output Visualization

The project generates:

• Pie Chart: Distribution of clean vs spam comments
• Statistical Summary: Total counts and spam percentage
• Color-coded Results: Green for clean, red for spam

Customization

Adjusting Sensitivity

Modify the thresholds in create_spam_labels():

# More sensitive detection
if pattern_matches >= 1:  # Lower threshold
    spam_label = 1
elif toxicity > 0.5:  # Lower toxicity threshold
    spam_label = 1

Adding Custom Patterns

Extend the spam patterns list:

spam_patterns = [
    r'subscribe', r'channel', r'check out',
    r'like and subscribe',  # Add custom patterns
    r'hit the bell',
    r'free download'
]

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/improvement)
3. Commit your changes (git commit -am 'Add new feature')
4. Push to the branch (git push origin feature/improvement)
5. Create a Pull Request

License

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

Acknowledgments

• martin-ha/toxic-comment-model for the pre-trained toxicity classifier
• Hugging Face Transformers for the pipeline infrastructure
• YouTube Spam Dataset contributors

Support

If you encounter any issues or have questions:

1. Check the Issues section
2. Review the code comments and documentation
3. Submit a new issue with detailed information

---

Note: This classifier is designed for educational and research purposes. Always review automated classifications and consider implementing human moderation for production systems.