Regensburg 2025 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

SOE: Fachverband Physik sozio-ökonomischer Systeme

SOE 10: Focus Session: Large Language Models, Social Dynamics, and Assessment of Complex Systems

SOE 10.4: Talk

Thursday, March 20, 2025, 16:00–16:15, H45

Dialogue React: Enhancing Conversation Synthesis for Social Science Simulations — •Ruggero Marino Lazzaroni — RWTH Aachen

Recent advancements in Large Language Models (LLMs) have spurred significant progress in conversation synthesis and social dynamics modeling. However, existing approaches often fall short in generating human-like dialogues suitable for complex social science simulations and opinion formation studies. This research introduces DialogueReact, a novel framework that builds upon previous work in conversation synthesis by incorporating react prompting, dialogue acts and agentic behaviour across different scales. By leveraging LLMs' ability to learn communication statistics, DialogueReact generates qualitatively better conversations compared to previous state-of-the-art methods. Our findings demonstrate the potential of DialogueReact to improve social science research by providing an extra tool for simulating complex social interactions and understanding collective behavior in social systems.

Keywords: LLM; ABM; Dialogue acts; multi-agent; agentic