Understanding and Predicting the Behavioural Evolution of Promotional Spambots on Social Media

Abstract

Social media bots are rapidly evolving, rendering traditional detection tools increasingly ineffective as these bots adapt their strategies. This research introduces a dynamic and predictive framework for modelling the behavioural evolution of online promotional spambots. Inspired by biological DNA, bot activities are encoded into behavioural sequences, with each block capturing seven distinct post-level features. Techniques such as sequence alignment, cosine similarity, and hierarchical clustering are used to group bots into behaviourally similar “families.” These families serve as the foundation for identifying behavioural mutations—insertions, deletions, substitutions, and alterations—that signal adaptive strategy changes. The model evaluates how these mutations propagate within and across bot families and investigates their predictive power through mutation transfer analysis and an event-driven case study. Results show that bots within the same family are significantly more likely to share and adopt behavioural mutations than those from different families. Closely related bots achieved high precision and F1 scores (up to 0.97) in mutation transfer prediction. These findings support the feasibility of a behavioural evolution model as a scalable, interpretable, and adaptive tool for anticipating future bot activity and offering a proactive approach to combating evolving threats on social media platforms.