Reproduce “Beyond Isolated Investor: Predicting Startup Success via Roleplay-Based Collective Agents” 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
Zhongyang Liu, Haoyu Pei, Xiangyi Xiao, Xiaocong Du, Yihui Li, Suting Hong, Kunpeng Zhang, Haipeng Zhang
SimVC-CAS, a multi-agent framework, models startup success prediction by simulating collective decision-making among heterogeneous investor agents interacting within a network. The system achieved a 25.0% relative improvement in AP@10 and 13.1% in AP@30 over the strongest baseline on a PitchBook dataset, enhancing accuracy in identifying high-potential startups by accounting for real-world VC dynamics.
Key insights
- Modeling the collective decision-making dynamics of multiple interacting investors significantly improves the accuracy of startup success prediction compared to single-investor or non-interactive multi-agent approaches.
- The effectiveness of investor interactions is enhanced by explicitly considering co-investment network topology and startup context through the VGAT, outperforming simpler interaction models.
- Effective decision revisions by agents prioritize 'Authority Influence' and 'Information Completion' from their network over mere 'Consensus Pressure,' aligning with real-world VC expert evaluation processes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.