Simulating Social Networks with Large Language Models: The SNSim Framework

ACL ARR 2026 January Submission8490 Authors

Keywords: Social Network Simulation, Opinion Dynamics, Large Language Models, Agent-based Modeling

Abstract: While large language models (LLMs) provide a promising technical foundation for social network simulation, existing approaches often implicitly equate simulation with text generation, thereby overlooking the dynamic interplay between opinion evolution and network structure. To address this limitation, we propose \textbf{SNSim}, a modular framework that formalizes social network simulation as the \textbf{joint and co-evolving} modeling of \textbf{language}, \textbf{opinion}, and \textbf{network structure}. SNSim adopts an \textbf{individual-community dual-channel} prompt design to systematically integrate personal attributes and community context into LLM-driven agents. Experiments on three real-world datasets demonstrate that SNSim consistently outperforms baseline approaches, successfully reproducing complex network topologies and temporally evolving opinion trajectories. Further analysis shows that the proposed modular design effectively mitigates linguistic homogenization while preserving structural realism. Taken together, these results indicate that SNSim provides a practical and operational framework to simulate social networks.

Paper Type: Long

Research Area: Computational Social Science, Cultural Analytics, and NLP for Social Good

Research Area Keywords: NLP tools for social analysis, human behavior analysis, quantitative analyses of news and/or social media, stance detection, multi-agent systems

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Data analysis

Languages Studied: English

Submission Number: 8490