A Multi-Dimensional Supervised Machine Learning Framework for Cybersecurity-Focused Social Media Bot Detection

Abstract

Social media bots are a significant threat to the integrity of social platforms, privacy and the authenticity of information. We propose a new supervised multi-perspective centric social media bot detection framework to perform crossplatform analysis on Instagram, Twitter or TikTok in this paper. Labeled datasets with human and bot accounts available on the Internet were crawled and analyzed to investigate behavior patterns of different bot types, as well as characteristics of their activity. We implemented and fit many different supervised learning models such as Random Forest, XGBoost, k-Nearest Neighbors, Decision Tree and Neural Networks. We evaluated model performance using standard metrics, which include accuracy, precision, recall, and F1-score after tackling platform-specific challenges. Experimental results show that Random Forest and XGBoost are able to outperform all other models under each platform in an extremely consistent way. Random Forest gave accuracies of 95% for Instagram, 91% for Twitter, and 97% for TikTok, and XGBoost got the following accuracies: Instagram (95%), Twitter (90%), and TikTok (97%). The Neural Network model demonstrated consistent performance with a trade-off between precision and recall. Bot Detection on TikTok and Results: TikTok bot detection has the highest overall accuracy of our evaluated platforms. This study presents the benefits of using supervised machine learning for social media bot detection while offering insights towards improving platform security, user privacy, and future studies through ensemble and advanced detection techniques.