A Large Language Model-Driven Agent-Based Modeling Framework With Multi-Round Communication for Simulating Vaccine Opinion Dynamics

Jul 16, 2026

A framework integrates a large language model (Qwen3-8B) into agent-based modeling to study how cognitive modules shape decisions and emergent opinion dynamics. Using vaccination opinion dynamics as a case study, the method simulates agents with heterogeneous profiles and social networks, activating different cognitive mechanisms. Scenarios compare a memory module and a prompt diversity module, showing opposite effects on emergent opinions. The framework also reproduces non-linear social-influence patterns and supports validation at the agent-based modeling level 3.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages11

Reading Time28 min

Methodology

Overview

Introduction

Simulation and Results

Conclusion

arXiv :2607 .07387v 1 [ cs .MA] 8 Jul 2026 A LARGE LANGUAGE MODEL-DRIVEN AGENT-BASED MODELING FRAMEWORK WITH MULTI-ROUND COMMUNICATION FOR SIMULATING VACCINE OPINION DYNAMICS Bo Zhang 1,2 , Na Jiang 1 1Thrust of Urban Governance and Design, The Hong Kong University of Science and Technology (Guangzhou) Guangzhou, Guangdong, CHINA 2Jinhe Center for Economic Research, Xi’an Jiaotong University Xi’an, Shaanxi, CHINA ABSTRACT Recently, Large Language Models (LLMs) have been utilized in various applications of computational social science and provide the possibility to integrate such models into agent-based modeling to explore the cognitive processes. However, how specific cognitive modules drive individual decisions and macro-level opinion dynamics remains unclear. Therefore, this study introduces a framework that integrates an LLM (Qwen3-8B) into agent-based modeling to investigate this problem, using vaccination opinion dynamics asa case study. We utilize this framework to simulate opinion dynamics among agents with heterogeneous profiles and social networks, evaluating scenarios by enabling different cognitive modules: a memory module and a prompt diversity module. The simulation results reveal that different cognitive modules have opposite impacts on our emergent opinion. Furthermore, the framework reproduces the non-linear behavior patterns of social influence observed in existing research, demonstrating our framework’s validity and potential to reach the level 3 validation of agent-based models. 1 INTRODUCTION Large Language Models (LLMs) have significantly advanced computational social science through applications in reasoning, zero-shot learning, and role-playing (Berti et al. 2025 ; Zhang et al. 2023 ; Zhou et al. 2024) . Among these applications, LLMs as computational proxies that mimic human behavior have transformed agent-based modeling from a static, rule-based heuristic method to a more complex approach (Horton, Filippas, and Manning 2023) . Specifically, such an approach allows us to capture more realistic human decision-making by utilizing LLMs’ generative capabilities for human-like behaviors, such as communication via dialogues and making decisions through these dialogues (Liu et al. 2025) . Consequently, agents can leverage an LLM to generate natural language to enrich their interaction, allowing researchers to simulate complex social systems and observe how macro-level phenomena emerge from micro-level language-based interactions (Larooij and Törnberg 2025; Lu et al. 2024 ; Vanhée et al. 2025) . With respect to integrating LLMs into agent-based modeling, one of the current efforts has been placed to extend core modules within the cognitive process by diversifying identity profiles, incorporating memory, reflection and react modules (Kuroki et al. 2025 ; Park et al. 2023 ; Shinn et al. 2023 ; Yan et al. 2025) . Such modules enable agents to engage in complex dialogues and tasks (Yao et al. 2023) and are theoretically important for an individual’s cognitive process (Chu et al. 2024 ; Zhang et al. 2025) . However, it remains unclear which module from the individuals’ cognitive processes contributes to their decision-making and further leads to the emergence of social phenomena on the macro-level (Li et al. 2025 ; Taillandier et al. 2026 ; Zhou et al. 2026) . Therefore, the goal of this study is to explore how different cognitive modules shape macro-level social phenomena by simulating agents to engage in sustained and multi-round interactions driven by an LLM. To achieve the goal above, opinion dynamics provides an ideal test environment for this study. Traditionally, simulations of opinion dynamics often treat peer influence as a simple transmission of states (Flache, Mäs, Feliciani, Chattoe-Brown, Deffuant, Huet, and Lorenz 2017; Yin, Crooks, and Yin Zhang and Jiang 2024) . However, real-world opinion formation, especially in high-stakes contexts like public health and vaccination, is rar

FAQs

What problem does the proposed LLM-driven agent-based framework address?

It targets the unclear relationship between individual cognitive modules and how those modules drive decisions, leading to macro-level opinion dynamics.

How is multi-round communication implemented in the simulations?

Agents interact through sustained, multi-round dialogues generated by the Qwen3-8B language model, and then update their vaccination opinions based on reflection after each dialogue.

Which cognitive modules are compared, and what is their impact?

The study enables different cognitive modules, specifically a memory module and a prompt diversity module, and finds that they produce opposite impacts on the emergent opinion.

Jul 16, 2026

Upload by Seraphina