Reproduce “GRAPHIA: Harnessing Social Graph Data to Enhance LLM-Based Social Simulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiarui Ji, Zehua Zhang, Zhewei Wei, Bin Tong, Guan Wang, Bo Zheng
GRAPHIA, a reinforcement learning framework developed by researchers at Renmin University of China and Alibaba, enhances LLM-based social simulations by explicitly leveraging social graph data during post-training. This framework improves the realism of simulated social interactions and network structures, achieving quantitative gains in both micro-level interaction fidelity and macro-level structural and phenomenological alignment across various real-world datasets.
Key insights
- Explicitly leveraging social graph data through reinforcement learning can bridge the gap between individual LLM agent behaviors and emergent network dynamics, significantly improving simulation realism.
- Integrating GNN-based structural priors into LLM reward functions during training guides message category generation towards more structurally coherent interactions, aligning micro-level actions with macro-level properties.
- A unified, quantitative evaluation paradigm encompassing both micro-level interaction quality and macro-level structural and phenomenological fidelity is crucial for systematically assessing and advancing LLM-based social simulations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.