Bot Detection System for Social Media

Overview

This project is a machine learning-based bot detection system for social media. It leverages advanced algorithms, including hierarchical models, to differentiate between human and bot accounts based on behavioral patterns, text analysis, and engagement metrics.

Features

• Hierarchical Model: Implements a multi-stage classification system using behavioral analysis, text-based analysis, and a meta-classifier for enhanced accuracy.
• Automated Detection: Uses machine learning models (Random Forest, NLP, and Meta-classification) to detect bots on social media platforms.
• Scalability: Can be extended to various social media platforms.
• Customizable: Allows fine-tuning of detection thresholds and model parameters.
• Adaptability: Can handle dynamic bot behaviors using advanced anomaly detection techniques.
• Visualization: Generates feature importance rankings and confusion matrices for model evaluation.
• Interactive Interface: Integrates with Gradio for real-time bot prediction.

Prerequisites

Before running the project, ensure you have the following installed:

• Python: Version 3.8+
• Libraries:

  numpy
  pandas
  scikit-learn
  seaborn
  matplotlib
  gradio
  requests
  beautifulsoup4
  torch
  transformers
  

Usage

1. Preprocess Data: Prepare your dataset using the preprocessing.py script.
2. Train the Model: Use the HierarchicalBotDetector class to train the behavioral model, tweet content model, and meta-classifier.
3. Test the Model: Evaluate the trained models using test data and analyze performance metrics (accuracy, confusion matrix, feature importance).
4. Deploy the Model: Use the provided Gradio interface for real-time bot detection.

Model Explanation

• Uses a hierarchical approach combining:
  • Behavioral Analysis: Features like account age, engagement metrics, and profile details.
  • Text-Based Analysis: NLP techniques (TF-IDF and transformers) to analyze tweet content.
  • Meta-Classifier: Integrates outputs from base models for improved accuracy.
• Feature Engineering:
  • Extracts features like activity patterns, text analysis, and network behavior.
  • Implements anomaly detection techniques to identify suspicious engagement patterns.
• Training Approach:
  • Trained on a labeled dataset of bots and human accounts.
  • Uses supervised learning methods (Random Forest, SVM, and Neural Networks).
  • Supports real-time adaptation to new bot behaviors.
• Visualization:
  • Generates confusion matrices and feature importance plots for better interpretability.

Interactive Interface

The project includes a Gradio-based web interface for real-time bot detection. Users can input profile details and tweets to obtain a classification result with confidence scores.

Contributing

Contributions are welcome! Feel free to open issues or submit pull requests.

Contact

For any queries, reach out at singhash058@gmail.com.