Social Media Bot Detection

This project implements a machine learning pipeline to classify social media tweets as bot-generated or human-generated using tweet-level linguistic and structural features.

Problem Statement

Automated bot accounts pose challenges on social media platforms by spreading spam and misinformation. The goal of this project is to detect bot-generated tweets using machine learning techniques without relying on temporal metadata.

Dataset

• Source: Public bot detection dataset
• Each row represents a single tweet
• Labels:
  • 0 → Human
  • 1 → Bot

Approach

1. Performed exploratory data analysis to understand class distribution and feature variance
2. Cleaned tweet text and removed noise such as URLs, mentions, and hashtags
3. Engineered tweet-level features including:
   • Tweet length
   • Hashtag count
   • Follower-based metrics
   • Sentiment polarity
4. Trained an XGBoost classifier and evaluated performance using ROC-AUC and F1-score

Technologies Used

• Python
• Pandas, NumPy
• Scikit-learn
• XGBoost
• TextBlob
• Matplotlib

Results

The model demonstrates the effectiveness of combining text-based features with numerical metadata for bot detection.