Advanced text-based transformer architecture for malicious social bots detection
Zineb Ellaky, Faouzia Benabbou · Mathematical Modeling and Computing · 2025
The increasing prevalence of automated social media accounts, or Social Media Bots (SMBs), presents significant challenges in maintaining authentic online discourse and preventing disinformation campaigns on social platforms. This research introduces a novel multiclass classification framework for detecting and categorizing SMBs, leveraging fine-tuned transformer-based models. In this study, we conducted a comprehensive comparative analysis of various transformer variants, including BERT, DistilBERT, RoBERTa, DeBERTa, XLNet, and ALBERT, to evaluate their efficacy in recognizing diverse types of social bots, such as spambots, politically motivated SMBs, Sybil-type accounts, fraudulent and fake accounts, as well as legitimate human users. The empirical findings indicate that the proposed methodology substantially outperforms traditional machine learning and deep learning approaches. Notably, the DistilBERT architecture demonstrated exceptional performance metrics, achieving 96.83% accuracy and 96.85% precision in social bot classification.