Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Opinion dynamics and mutual influence with LLM agents through dialog simulation

Yulong He, Dutao Zhang, Sergey Kovalchuk, Pengyi Li, Artem Sedakov

TL;DR

The paper addresses the scarcity of longitudinal opinion data by introducing LLM-based dialog simulations that emulate opinion dynamics under classical updating rules. It maps text outputs to numerical opinions via sentiment analysis and preserves initial opinions to capture anchoring, enabling updates according to $x(t+1)=W x(t)$ (DG) or $x(t+1)=S W x(t)+(I-S)x(1)$ (FJ) across configurable network topologies. Contributions include bridging classical DeGroot and Friedkin–Johnsen theory with modern multi-agent LLM systems, providing a scalable experimental testbed for data-scarce settings, and benchmarking multiple LLMs to characterize mutual trust and updating behavior. The framework offers linguistically grounded, scalable insights into social influence and opinion formation across networks, with broad relevance to computational social science and policy-relevant simulations.

Abstract

A fundamental challenge in opinion dynamics research is the scarcity of real-world longitudinal opinion data, which complicates the validation of theoretical models. To address this, we propose a novel simulation framework using large language model (LLM) agents in structured multi-round dialogs. Each agent's dialog history is iteratively updated with its own previously stated opinions and those of others analogous to the classical DeGroot model. Furthermore, by retaining each agent's initial opinion throughout the dialog, we simulate anchoring effects consistent with the Friedkin-Johnsen model of opinion dynamics. Our framework thus bridges classical opinion dynamics models and modern multi-agent LLM systems, providing a scalable tool for simulating and analyzing opinion formation when real-world data is limited or inaccessible.

Opinion dynamics and mutual influence with LLM agents through dialog simulation

TL;DR

The paper addresses the scarcity of longitudinal opinion data by introducing LLM-based dialog simulations that emulate opinion dynamics under classical updating rules. It maps text outputs to numerical opinions via sentiment analysis and preserves initial opinions to capture anchoring, enabling updates according to (DG) or (FJ) across configurable network topologies. Contributions include bridging classical DeGroot and Friedkin–Johnsen theory with modern multi-agent LLM systems, providing a scalable experimental testbed for data-scarce settings, and benchmarking multiple LLMs to characterize mutual trust and updating behavior. The framework offers linguistically grounded, scalable insights into social influence and opinion formation across networks, with broad relevance to computational social science and policy-relevant simulations.

Abstract

A fundamental challenge in opinion dynamics research is the scarcity of real-world longitudinal opinion data, which complicates the validation of theoretical models. To address this, we propose a novel simulation framework using large language model (LLM) agents in structured multi-round dialogs. Each agent's dialog history is iteratively updated with its own previously stated opinions and those of others analogous to the classical DeGroot model. Furthermore, by retaining each agent's initial opinion throughout the dialog, we simulate anchoring effects consistent with the Friedkin-Johnsen model of opinion dynamics. Our framework thus bridges classical opinion dynamics models and modern multi-agent LLM systems, providing a scalable tool for simulating and analyzing opinion formation when real-world data is limited or inaccessible.
Paper Structure (10 sections, 3 figures, 2 tables)

This paper contains 10 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Initial opinion generation
  4. Modeling social influence
  5. Analytical pipeline
  6. Experiments and analysis
  7. Homogeneous LLM agents
  8. Heterogeneous LLM agents
  9. Conclusion
  10. Conflict of interest

Figures (3)

  • Figure 1: Overview of the multi-agent LLM opinion dynamics simulation framework.
  • Figure 2: Components of the original opinion (sentiment) vector $x(t)$ and fitted opinion vector $\widehat{x}(t)$ for homogeneous populations of LLM agents under different opinion dynamics models. The horizontal axis represents the round number.
  • Figure 3: Components of the original opinion (sentiment) vector $x(t)$ and fitted opinion vector $\widehat{x}(t)$ for heterogeneous populations of LLM agents under two opinion dynamics models. The components of the vectors correspond to the opinions of the LLM agents DeepSeek V3, GPT-4o mini, Qwen2.5, Mistral Large, and Llama 3.3, respectively. The horizontal axis represents the round number.