LLM-Agent-based Social Simulation for Attitude Diffusion
Summary: arXiv:2604.03898v1 Announce Type: new
Abstract
This paper introduces discourse_simulator, an open-source framework that combines Large Language Models (LLMs) with agent-based modeling. It offers a new way to simulate how public attitudes toward immigration change over time in response to salient events such as protests, controversies, or policy debates.
Key Features of discourse_simulator
The framework utilizes LLMs to:
- Generate social media posts.
- Interpret diverse opinions.
- Model the diffusion of ideas through social networks.
Unlike traditional agent-based models that rely on fixed, rule-based opinion updates and lack the ability to generate natural language or consider current events, discourse_simulator integrates multidimensional sociological belief structures and real-world event timelines.
Technical Overview
This innovative framework is encapsulated in an open-source Python package that features:
- Generative agents operating within a small-world network topology.
- A live news retrieval system to keep the simulation contextually relevant.
discourse_sim is specifically designed as a social science research instrument aimed at studying attitude dynamics, polarization, and belief evolution following real-world critical events.
A New Approach to Social Science Research
One of the most significant distinctions of discourse_sim lies in its epistemological stance. Unlike other LLM Agent Swarm frameworks that often treat simulations as a prediction black box, discourse_sim is intended as a theory-testing instrument. This fundamentally alters the approach to studying social science problems, providing researchers with a more nuanced understanding of how public attitudes can shift in response to various stimuli.
Case Study: The Dublin Anti-Immigration March
The paper further illustrates the capabilities of the framework by modeling the Dublin anti-immigration march that occurred on April 26, 2025. In this simulation, a total of 100 agents were utilized over a 15-day period to observe how public sentiment evolved in response to the event.
Conclusion
The discourse_simulator presents a significant advancement in the field of social simulation, combining the strengths of LLMs with agent-based modeling to provide a comprehensive tool for understanding the dynamics of public attitudes. Researchers interested in exploring this innovative framework can access the Python package through the following link: