Abstract

Social simulation is transforming traditional social science research bymodeling human behavior through interactions between virtual individuals andtheir environments. With recent advances in large language models (LLMs), thisapproach has shown growing potential in capturing individual differences andpredicting group behaviors. However, existing methods face alignment challengesrelated to the environment, target users, interaction mechanisms, andbehavioral patterns. To this end, we introduce SocioVerse, an LLM-agent-drivenworld model for social simulation. Our framework features four powerfulalignment components and a user pool of 10 million real individuals. Tovalidate its effectiveness, we conducted large-scale simulation experimentsacross three distinct domains: politics, news, and economics. Resultsdemonstrate that SocioVerse can reflect large-scale population dynamics whileensuring diversity, credibility, and representativeness through standardizedprocedures and minimal manual adjustments.