Tutorial YSocial: A Sandbox for Computational Social Science

Social media platforms are complex socio-technical systems in which individual behaviors, network structures, information diffusion, and platform mechanisms interact to produce emergent collective phenomena. Recent advances in large language models offer new opportunities to study these systems through simulations populated by agents capable of generating content, interacting with others, and adapting their behavior to evolving social contexts.
This tutorial introduces YSocial, a framework for simulating social media environments with LLM-powered agents. We will present the main concepts underlying YSocial and guide participants through the design, configuration, execution, and analysis of social media simulations. Particular attention will be devoted to modeling agent populations and their characteristics, defining interaction and content-generation mechanisms, constructing dynamic social networks, and collecting data describing the resulting individual and collective behaviors.

Through practical examples, participants will learn how YSocial can be used to investigate questions related to computational social science, including opinion dynamics, homophily and polarization, information diffusion, and the emergence of network structures. We will also discuss methodological challenges associated with generative-agent simulations, including experimental design, reproducibility, scalability, and the interpretation and validation of synthetic social behavior.

By the end of the tutorial, participants will be able to design and run their own YSocial experiments and critically assess the opportunities and limitations of LLM-driven social simulation as a methodology for studying online social systems.