Social media research has advanced the understanding of individual-level processes in online communication, yet these insights are rarely linked to meso- and macro-level outcomes, including polarization dynamics, misinformation spread, and public opinion formation. Agent-based modeling (ABM) offers a powerful approach to bridging this gap by linking micro-level interactions to emergent macro-level phenomena. While ABM has been widely adopted in computer science, engineering, mathematics, and economics to model social media communication, such applications often rely on theoretical assumptions and empirical findings outside of social media research. In this study, we systematically review how ABM has been applied across various fields to study social media phenomena and offer a comprehensive foundation for building on existing models. We identify common practices, research gaps, and opportunities for theoretical, methodological, and open science advancements. Through this project, we aim to encourage a broader adoption of ABM in the study of social media and provide a practical entry point for those new to modeling.

Discussions of social issues on social media have a capacity to form social movements reaching far beyond the virtual setting, such as in the cases of the #BlackLivesMatter and the #MeToo movements. In this paper, I present the social issue emergence agent-based model (SIE ABM) that consists of a network of social media users that dynamically choose topics to pay atten-tion to. The model explores how the competitive information environment and the users’ limited attention impact the spread of competing topics. The model exploration results in three scenarios of dynamic emergent patterns: the “focus” scenario, in which all actors focus on the same topic; the “di-vide” scenario, in which the actors focus on two or three topics; and the “fragmentation” scenario, in which the model never reaches an equilibrium, and the actors keep changing the topics of their interest without letting any of them to take over the majority.

Social media platforms have become essential for interpersonal communication, accessing news and other information. There are growing concerns about the presence of social bots that mimic human behaviour while pushing their own agenda distorting online communities and the news environment. Previous research has demonstrated social networks and social media platforms play a key role in the emergence of polarisation. The selective exposure of filter bubbles limits the information users get exposed to and promotes certain content. Echo chambers, or highly homophilous communities, can also limit the exposure to information and people that disagree with the majority view. Social bots are an additional layer of complexity in this process. To better understand their behaviour and consequences for data quality this paper proposes an agent-based simulation model (ABM) to explore how social bots distort the opinion dynamics on social media platforms. It uses survey data on political attitudes and media trust in Germany in combination with web scrapping data on internet use to calibrate some of the model parameters. Simulation conditions separately manipulated (1) the number of social bots, (2) their information strategies, (3) the homophily or group similarity in personal networks and (4) the level of selective exposure through filter bubbles on social media. We are currently implementing the social bots on the model and designing their behaviours. Once we collect the simulation data, we will compare the synthetic and empirical opinion data to determine the effects social bots had in terms of both, data quality and opinion dynamics. We will present our findings at the conference.

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, we compare static (retroactive) and dynamic (in-simulation) moderation, revealing that static evaluations systematically overestimate the effectiveness of user bans: when moderation is applied in real time, network feedback loops (such as compensatory resharing by remaining users) dampen the expected reduction in low-quality content. These findings highlight the necessity of simulation-based evaluation for content moderation policies and contribute a reusable, empirically grounded simulation framework.

ABM research into social networks regularly emphasizes its potential policy impact. Little is known, however, about how ABM can support platform regulators, and under what conditions platform regulators would be able to rely on ABM-produced evidence. This significantly limits the potential poli-cy impact of ABM research into social networks. This paper aims to reduce this uncertainty in two ways. First, it analyses how ABM fits into existing platform regulation, drawing on literature on platform risk assessment, the use of agent-based modelling in other fields, and anticipatory governance and evidence-based policymaking. Second, to clarify the conditions under which platform regulators can rely on evidence produced by agent-based models, the article reports the results of interviews with 18 regulators. The article finds that agent-based simulations have significant potential to com-plement other sources of evidence used in platform risk assessment. Their main added value is that they allow researchers and regulators to more easi-ly, ethically, and independently explore the potential impacts of changes to platform design, and provide interactive and visual scenarios that concretize abstract platform dynamics. The article closes by identifying how ABM re-search can increase its uptake by platform regulators.