Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generalized Multi-agent Social Simulation Framework

Gang Li, Jie Lin, Yining Tang, Ziteng Wang, Yirui Huang, Junyu Zhang, Shuang Luo, Chao Wu, Yike Guo

TL;DR

The paper addresses scalability, reusability, and memory efficiency in multi-agent social simulations by introducing a modular, object-oriented framework with hierarchical base classes and a memory summarization mechanism. It presents a retrieval-then-summarization memory pipeline and demonstrates how derived classes can customize environments to model online social interactions. The approach yields improvements across Self-Knowledge, Reaction, General Response, and Memory System metrics in controlled benchmarks, with ablation confirming the value of memory summarization. End-to-end evaluations in online-discussion scenarios show increasingly nuanced and policy-aware agent behavior, aided by internet information access. Overall, the framework enhances realism and adaptability in large-scale social simulations and provides a pathway for reusable, configurable agent-based environments.

Abstract

Multi-agent social interaction has clearly benefited from Large Language Models. However, current simulation systems still face challenges such as difficulties in scaling to diverse scenarios and poor reusability due to a lack of modular design. To address these issues, we designed and developed a modular, object-oriented framework that organically integrates various base classes through a hierarchical structure, harvesting scalability and reusability. We inherited the framework to realize common derived classes. Additionally, a memory summarization mechanism is proposed to filter and distill relevant information from raw memory data, prioritizing contextually salient events and interactions. By selecting and combining some necessary derived classes, we customized a specific simulated environment. Utilizing this simulated environment, we successfully simulated human interactions on social media, replicating real-world online social behaviors. The source code for the project will be released and evolve.

Generalized Multi-agent Social Simulation Framework

TL;DR

The paper addresses scalability, reusability, and memory efficiency in multi-agent social simulations by introducing a modular, object-oriented framework with hierarchical base classes and a memory summarization mechanism. It presents a retrieval-then-summarization memory pipeline and demonstrates how derived classes can customize environments to model online social interactions. The approach yields improvements across Self-Knowledge, Reaction, General Response, and Memory System metrics in controlled benchmarks, with ablation confirming the value of memory summarization. End-to-end evaluations in online-discussion scenarios show increasingly nuanced and policy-aware agent behavior, aided by internet information access. Overall, the framework enhances realism and adaptability in large-scale social simulations and provides a pathway for reusable, configurable agent-based environments.

Abstract

Multi-agent social interaction has clearly benefited from Large Language Models. However, current simulation systems still face challenges such as difficulties in scaling to diverse scenarios and poor reusability due to a lack of modular design. To address these issues, we designed and developed a modular, object-oriented framework that organically integrates various base classes through a hierarchical structure, harvesting scalability and reusability. We inherited the framework to realize common derived classes. Additionally, a memory summarization mechanism is proposed to filter and distill relevant information from raw memory data, prioritizing contextually salient events and interactions. By selecting and combining some necessary derived classes, we customized a specific simulated environment. Utilizing this simulated environment, we successfully simulated human interactions on social media, replicating real-world online social behaviors. The source code for the project will be released and evolve.

Paper Structure

This paper contains 21 sections, 2 equations, 6 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Complex Social Simulation
  4. LLM-based Multi-Agent Framework
  5. LLM-based Multi-Agent Simulation
  6. Method
  7. Basic Framework Explanation
  8. Memory Summarization Mechanism
  9. Experiment
  10. Controlled Evaluation
  11. Result
  12. Ablation Study
  13. End-to-end Evaluation
  14. Basic Single-Event Discussion
  15. Multi-Event Discussion
  16. ...and 6 more sections

Figures (6)

  • Figure 1: The hierarchical structure of base classes.
  • Figure 2: The hierarchical structure of the proposed framework.
  • Figure 3: The prompt template for memory summary.
  • Figure 4: The prompt template for GPT evaluation.
  • Figure 5: The hierarchical structure of the customized environment.
  • ...and 1 more figures