• large language model
• moderation
• simulation

Evaluation of moderation effects on the macro-level dynamics of Online Social Networks (OSNs) has traditionally relied on empirical, on-the-field methods that impose considerable limitations on both data collection and reliability, primarily concerning high costs and lack of control. The advent of Large Language Models (LLMs) and prompt learning frameworks has led to significant advancements in Agent Based Modeling (ABM), enabling the simulation of human-like behavior with unprecedented degree of believability. These latest developments pave the way for new experimental solutions. The present work takes the pioneering step of exploring simulation-based evaluation of moderation systems by introducing a novel OSN simulator powered by LLMs. Designed for the implementation of counterfactual-based experiments, this software tool supports robust, multi-dimensional analyses of moderation effects, with a keen focus on abusive and toxic discourse.