Reproduce “ResearchTown: Simulator of Human Research Community” 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
Haofei Yu, Zhaochen Hong, Zirui Cheng, Kunlun Zhu, Keyang Xuan, Jinwei Yao, Tao Feng, Jiaxuan You
Researchers at the University of Illinois Urbana-Champaign developed RESEARCHTOWN, a multi-agent framework using an Agent-Data Graph and TextGNN, to simulate human scientific communities and generate research papers and reviews. The system achieves a 67.51% average similarity for simulated papers to real ones and demonstrates the value of collaboration in generating interdisciplinary research ideas.
Key insights
- Simulating collaborative interactions between multiple LLM agents and leveraging graph-structured data (cited papers, co-authors) significantly improves the quality and realism of generated research outputs.
- The Agent-Data Graph and TextGNN paradigm effectively extends graph neural networks from static analysis to dynamic simulation, allowing for the generation and refinement of novel research artifacts.
- The framework can inspire new, interdisciplinary research ideas by facilitating interactions between agents from disparate fields, demonstrating potential beyond merely replicating existing work.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.