Lockstep uses machine learning models to determine whether a user interacting on a social media site is displaying behavior that is similar to the behavior displayed by known bots. The focal idea of Lockstep is in the simplicity of and the ability to scale the architecture used in lockstep. A benefit of this simplicity is being closer to the position of being able to 'interpret' decisions made by the chosen algorithms.
Lockstep was trained on Twibot-22.
@ https://github.com/LuoUndergradXJTU/TwiBot-22
Lockstep utilizes fine-tuned Random Forest models and a graph neural network 'GraphSage' to make predictions on features.
The original model performed well on it's task versus competitors.
