Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Integrating LLM and Diffusion-Based Agents for Social Simulation

Xinyi Li, Zhiqiang Guo, Qinglang Guo, Hao Jin, Weizhi Ma, Min Zhang

TL;DR

This work tackles scalable social information diffusion prediction by fusing LLM-driven agents for a core subset of users with diffusion-model agents for the remaining population. The approach seeds the diffusion process with semantically rich, personalized decisions from LLMs and completes diffusion using a multi-scale diffusion model that integrates global and local user representations via a fusion gate. It introduces a structured, prompts-based LLM agent with a three-stage decision framework and a LHGN-based global-scale learner paired with a local sequence model, achieving superior predictive accuracy on Weibo and Zhihu datasets. The proposed hybrid framework demonstrates strong performance gains over traditional ABMs and full LLM-based approaches, offering scalable, topic-aware diffusion modeling with practical implications for policy design and platform optimization.

Abstract

Agent-based social simulation provides a valuable methodology for predicting social information diffusion, yet existing approaches face two primary limitations. Traditional agent models often rely on rigid behavioral rules and lack semantic understanding of textual content, while emerging large language model (LLM)-based agents incur prohibitive computational costs at scale. To address these challenges, we propose a hybrid simulation framework that strategically integrates LLM-driven agents with diffusion model-based agents. The framework employs LLM-based agents to simulate a core subset of users with rich semantic reasoning, while a diffusion model handles the remaining population efficiently. Although the two agent types operate on disjoint user groups, both incorporate key factors including user personalization, social influence, and content awareness, and interact through a coordinated simulation process. Extensive experiments on three real-world datasets demonstrate that our framework outperforms existing methods in prediction accuracy, validating the effectiveness of its modular design.

Integrating LLM and Diffusion-Based Agents for Social Simulation

TL;DR

This work tackles scalable social information diffusion prediction by fusing LLM-driven agents for a core subset of users with diffusion-model agents for the remaining population. The approach seeds the diffusion process with semantically rich, personalized decisions from LLMs and completes diffusion using a multi-scale diffusion model that integrates global and local user representations via a fusion gate. It introduces a structured, prompts-based LLM agent with a three-stage decision framework and a LHGN-based global-scale learner paired with a local sequence model, achieving superior predictive accuracy on Weibo and Zhihu datasets. The proposed hybrid framework demonstrates strong performance gains over traditional ABMs and full LLM-based approaches, offering scalable, topic-aware diffusion modeling with practical implications for policy design and platform optimization.

Abstract

Agent-based social simulation provides a valuable methodology for predicting social information diffusion, yet existing approaches face two primary limitations. Traditional agent models often rely on rigid behavioral rules and lack semantic understanding of textual content, while emerging large language model (LLM)-based agents incur prohibitive computational costs at scale. To address these challenges, we propose a hybrid simulation framework that strategically integrates LLM-driven agents with diffusion model-based agents. The framework employs LLM-based agents to simulate a core subset of users with rich semantic reasoning, while a diffusion model handles the remaining population efficiently. Although the two agent types operate on disjoint user groups, both incorporate key factors including user personalization, social influence, and content awareness, and interact through a coordinated simulation process. Extensive experiments on three real-world datasets demonstrate that our framework outperforms existing methods in prediction accuracy, validating the effectiveness of its modular design.
Paper Structure (35 sections, 11 equations, 3 figures, 4 tables)

This paper contains 35 sections, 11 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Social Information Diffusion Prediction
  4. Traditional Model-driven Agent-based Social Simulation
  5. Social Simulation with LLMs
  6. Simulation Framework
  7. Task Definition
  8. Overall Framework
  9. LLM-based Agent
  10. User Description
  11. User Profile
  12. User historical interactions
  13. Multi-stage Decision Framework
  14. Decision Strategies
  15. Multi-stage Desicion-making Process
  16. ...and 20 more sections

Figures (3)

  • Figure 1: The LLM-empowered Hybrid Simulation Agent Framework
  • Figure 2: Ablation Experiments on LLM-based Agents over three datasets
  • Figure 3: Ablation Experiments on LLM-based Agents over three datasets