"If It Looks Like a User": Measuring Real-Time Moderation Effects via Social Media Simulation
2026-08-17 • Social and Information Networks
Social and Information NetworksComputers and SocietyMultiagent Systems
AI summaryⓘ
The authors improved a computer model that mimics how information, especially about vaccines during the COVID-19 pandemic, spreads on social media. They made the model more realistic by fitting it to real data and tested it to ensure it behaves like true social networks. Using this model, they showed that removing top users spreading misinformation lowers bad content in both real and simulated data. They also found that banning users in real time is less effective than previously thought because others can compensate by sharing more. Their work shows the importance of using realistic simulations to test social media policies.
agent-based modelinformation diffusionsocial networksmisinformationcontent moderationsimulation calibrationCMA-ESvaccine discourseuser bansreal-time moderation
Authors
Enrico Verdolotti, Gianluca Nogara, Luca Luceri, Silvia Giordano
Abstract
Agent-based social media simulators offer a controlled environment to study content moderation, yet their value hinges on how faithfully they reproduce real platform dynamics. We develop a calibrated extension of SimSoM, an agent-based model of information diffusion on social networks, grounded in a real-world dataset of online vaccine discourse during the COVID-19 pandemic. Our approach replaces ad-hoc parametrisations with empirically fitted distributions, optimised via CMA-ES (Covariance Matrix Adaptation Evolution Strategy) and validated against real data across temporal, distributional, and structural dimensions. Using this validated simulator, we provide three key contributions. First, we show that the calibrated model reproduces key statistical signatures of the empirical data, including activity distributions, post/reshare ratios, and temporal patterns. Second, we apply established misinformation-spreader detection and prevention methods to both empirical and simulated data, progressively removing top-ranked users and showing that the resulting decline in low-quality content is consistent across the two. Third, comparing static (retroactive) and dynamic (in-simulation) moderation across 30 network realisations, we show that static evaluation significantly overestimates the effectiveness of user bans for the most effective detectors: when moderation is applied in real time, compensatory resharing by the remaining users dampens the expected reduction in low-quality content, so static estimates should be read as an upper bound. These findings highlight the necessity of simulation-based evaluation for content moderation policies and contribute a reusable, empirically grounded simulation framework.