Large Language Model Driven Agents for Simulating Echo Chamber Formation

TL;DR

This work addresses how echo chambers form in social networks by jointly modeling opinion evolution and network rewiring with Large Language Models as generative agents. The method uses prompts to implement the opinion update $O_i(t+1) = O_i(t) + \frac{1}{|N_i(t)|} \sum_{j\in N_i(t)} f(O_j(t), O_i(t))$ and rewiring probabilities $P_{\text{unfollow}}(i,j) \propto 1 - g(O_i,O_j)$, while $f$ and $g$ are produced via LLM prompts and the model also generates new posts $y_i(t)$ from contextual content $C_i$ and $\{C_j\}$. Real-world data from Twitter/X on COVID-19 vaccine discourse and Ukraine-related polarization are used to validate the simulations against observed structural (modularity, clustering) and semantic (language distribution) patterns, with six LLMs and a traditional baseline evaluated. The findings show that LLM-driven simulations more accurately reproduce both the structural properties and semantic content of observed echo chambers, offering a robust tool for studying polarization dynamics and evaluating interventions in online communities.

Abstract

The rise of echo chambers on social media platforms has heightened concerns about polarization and the reinforcement of existing beliefs. Traditional approaches for simulating echo chamber formation have often relied on predefined rules and numerical simulations, which, while insightful, may lack the nuance needed to capture complex, real-world interactions. In this paper, we present a novel framework that leverages large language models (LLMs) as generative agents to simulate echo chamber dynamics within social networks. The novelty of our approach is that it incorporates both opinion updates and network rewiring behaviors driven by LLMs, allowing for a context-aware and semantically rich simulation of social interactions. Additionally, we utilize real-world Twitter (now X) data to benchmark the LLM-based simulation against actual social media behaviors, providing insights into the accuracy and realism of the generated opinion trends. Our results demonstrate the efficacy of LLMs in modeling echo chamber formation, capturing both structural and semantic dimensions of opinion clustering. %This work contributes to a deeper understanding of social influence dynamics and offers a new tool for studying polarization in online communities.

Figures (5)

• Figure 1: The framework is divided into three main components: (1) Data Preparation, where social network data is collected and used to build an user network; (2) Simulation Process, where an LLM generates user posts and adjusts connection dynamically; and (3) Analysis and Validation, which analyzes the simulated structure of user interactions and echo chamber effect.
• Figure 2: Prompt template example
• Figure 3: Simulation results using ChatGPT and Gemini. Opinion dynamics at different time steps illustrate the gradual formation of echo chambers.
• Figure 4: Comparison of stance accuracy across different models in COVID-19 dataset
• Figure 5: Clustering of real-world data (left) and simulated data (right) illustrated in t-SNE embedding space.