Elamur Research⌕
☰
home / papers / 2026 / Analyze Like a Venture Capitalist: Information-Gain and Knowledge Enhanced Graph Reasoning for Startup Success Prediction

Analyze Like a Venture Capitalist: Information-Gain and Knowledge Enhanced Graph Reasoning for Startup Success Prediction

Paper · 2026 · Haoyu Pei, Zhongyang Liu, Xiangyi Xiao et al.

Summary

The paper addresses early-stage startup success prediction. The target is whether a company that has closed a seed or angel round raises its next financing within one year. The authors argue that this is an "off-graph" problem: the investor–company network does not contain the answer but supplies evidence for it. They also identify three obstacles: path explosion in multi-hop retrieval, one-sided label noise (an unrecorded exit is not necessarily a failure), and missing domain expertise.

Their framework, MIRAGE-VC, combines three components:

  • Retriever: trained offline on information-gain signals, it grows a compact evidence subgraph around the target company step by step.
  • Knowledge base: a two-layer VC resource holding theory from books and academic work and practice from investment playbooks.
  • Manager LLM: a compact model tuned with SFT and then a noise-aware DPO, in which Positive-Unlabeled estimation reduces the weight of negatives that may be latent successes.

On proprietary PitchBook data, the system beats prior baselines. Results on public Crunchbase data point the same way, and VC experts judged its rationales to be of professional quality. The selection-by-marginal-utility idea may transfer to other tasks where graphs serve as evidence, such as recommendation or credit risk.

Key points

  • Data:
    • PitchBook Global VC covers 2005 to November 2023, with 263,729 startups and 1,014,157 individuals.
    • The Crunchbase 2013 snapshot has 196,553 companies and 226,708 individuals.
  • Headline results: over state-of-the-art baselines, MIRAGE-VC reports +5.9% F1 and +22.1% Precision@5. In AP@5 it also outperforms Llama 3.3 70B by 30.0% and GPT-5 by 15.1%.
  • Ablations, evidence side: the full model reaches 35.92% AP@5. Removing subgraph retrieval causes the largest drop (−6.82 AP@5). Random selection instead of IG scoring costs −4.32, and dropping VC knowledge costs −2.12.
  • Ablations, training side: noise-aware DPO beats standard DPO by 1.42 AP@5, and standard DPO beats SFT alone.
  • Stated limitations: the data is mostly from the US and Europe, and the retriever treats all historical edges equally regardless of recency. Code and reasoning outputs are released on GitHub (ZhangDataLab/MIRAGE-VC).

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of Maryland, College ParkCollege Park · United States132 works · 25,380 citations🎓ShanghaiTech UniversityShanghai · China13 works · 1,974 citations🎓Xi’an Jiaotong-Liverpool UniversitySuzhou · China10 works · 319 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksBusiness, Management and Accounting · 32 worksManagement Information Systems · 4 worksAccounting · 5 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,426 works

Cite

Cite as Haoyu Pei, Zhongyang Liu, Xiangyi Xiao, Xiaocong Du, Suting Hong, Kunpeng Zhang and Haipeng Zhang. 2026. Analyze Like a Venture Capitalist: Information-Gain and Knowledge Enhanced Graph Reasoning for Startup Success Prediction. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1555

@inproceedings{pei2026analyze,
  title = {{Analyze Like a Venture Capitalist: Information-Gain and Knowledge Enhanced Graph Reasoning for Startup Success Prediction}},
  author = {Haoyu Pei, Zhongyang Liu, Xiangyi Xiao, Xiaocong Du, Suting Hong, Kunpeng Zhang and Haipeng Zhang},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1555},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1555}
}

References

32 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Javier Arroyo, Francesco Corea, Guillermo Jimenez-Diaz, and Juan A Recio-Garcia. 2019. Assessment of machine learning performance for decision support in venture capital investments. Ieee Access, 7:124233– 124243. find
  2. Francisco Ramadas da Silva Ribeiro Bento. 2017. Predicting start-up success with machine learning. find
  3. Mas-Ryuichi Kiryo, Gang Niu, Marthinus C Du Plessis, and Masashi Sugiyama. 2017. Positive-unlabeled learning with non-negative risk estimator. Advances in neural information processing systems, 30. find
  4. Hyungjin Ko and Jaewook Lee. 2024. Can chatgpt improve investment decisions? from a portfolio management perspective. Finance Research Letters, 64:105433. find
  5. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, and 1 others. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems, 33:9459–…
  6. Xiao-Yang Liu, Guoxuan Wang, Hongyang Yang, and Daochen Zha. 2023. Fingpt: Democratizing internet-scale data for financial large language models. arXiv preprint arXiv:2307.10485. — [arXiv:2307.10485](https://arxiv.org/abs/2307.10485) find
  7. Zhongyang Liu, Haoyu Pei, Xiangyi Xiao, Xiaocong Du, Yihui Li, Suting Hong, Kunpeng Zhang, and Haipeng Zhang. 2025. Llm agents as vc investors: Predicting startup success via roleplay-based collective simulation. arXiv preprint arXiv:2512.22608. — [arXiv:2512.22608](https://arxiv.org/abs/2512.22608) find
  8. Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct preference optimization: Your language model is secretly a reward model. Advances in neural information processing systems, 36:53728–53741.
  9. Parth Sarthi, Salman Abdullah, Aditi Tuli, Shubh Khanna, Anna Goldie, and Christopher D Manning. 2024. Raptor: Recursive abstractive processing for tree-organized retrieval. In The Twelfth International Conference on Learning Representations.
  10. Boris Sharchilev, Michael Roizner, Andrey Rumyantsev, Denis Ozornin, Pavel Serdyukov, and Maarten De Rijke. 2018. Web-based startup success prediction. In Proceedings of the 27th ACM international conference on information and knowledge management, pages 2283–2291. find
  11. Linhao Luo, Yuan-Fang Li, Gholamreza Haffari, and Shirui Pan. 2023. Reasoning on graphs: Faithful and interpretable large language model reasoning. arXiv preprint arXiv:2310.01061. find
  12. Hwanjun Song, Minseok Kim, Dongmin Park, Yooju Shin, and Jae-Gil Lee. 2022. Learning from noisy labels with deep neural networks: A survey. IEEE transactions on neural networks and learning systems, 34(11):8135–8153. find
  13. Shiwei Lyu, Xiaofeng Li, Suting Hong, Qing Ke, Jinjie Gu, Kunpeng Zhang, and Haipeng Zhang. 2025. Help me screen: Analyzing and predicting the success of start-ups in dynamic venture capital networks. ACM Transactions on Intelligent Systems and Technology. find
  14. Jiashuo Sun, Chengjin Xu, Lumingyuan Tang, Saizhuo Wang, Chen Lin, Yeyun Gong, Lionel M Ni, Heung-Yeung Shum, and Jian Guo. 2023. Think-on-graph: Deep and responsible reasoning of large language model on knowledge graph. arXiv preprint arXiv:2307.07697. find
  15. Shiwei Lyu, Shuai Ling, Kaihao Guo, Haipeng Zhang, Kunpeng Zhang, Suting Hong, Qing Ke, and Jinjie Gu. 2021. Graph neural network based vc investment success prediction. arXiv preprint arXiv:2105.11537. — [arXiv:2105.11537](https://arxiv.org/abs/2105.11537) find
  16. Heng Wang, Shangbin Feng, Tianxing He, Zhaoxuan Tan, Xiaochuang Han, and Yulia Tsvetkov. 2023. Can language models solve graph problems in natural language? Advances in Neural Information Processing Systems, 36:30840–30861. find
  17. Abdurahman Maarouf, Stefan Feuerriegel, and Nicolas Pröllochs. 2025. A fused large language model for predicting startup success. European Journal of Operational Research, 322(1):198–214. find
  18. Xisen Wang, Yigit Ihlamur, and Fuat Alican. 2025. Ssff: Investigating llm predictive capabilities for startup success through a multi-agent framework with enhanced explainability and performance. Preprint, arXiv:2405.19456. — [arXiv:2405.19456](https://arxiv.org/abs/2405.19456) find
  19. Ian C MacMillan, Robin Siegel, and PN Subba Narasimha. 1985. Criteria used by venture capitalists to evaluate new venture proposals. Journal of Business venturing, 1(1):119–128. find
  20. Dat Mai. 2024. Stockgpt: A genai model for stock prediction and trading. arXiv preprint arXiv:2404.05101. — [arXiv:2404.05101](https://arxiv.org/abs/2404.05101) find
  21. Costas Mavromatis and George Karypis. 2024. Gnn-rag: Graph neural retrieval for large language model reasoning. arXiv preprint arXiv:2405.20139. find
  22. Junlang Qian, Zixiao Zhu, Hanzhang Zhou, Zijian Feng, Zepeng Zhai, and Kezhi Mao. 2025. Beyond the next token: Towards prompt-robust zero-shot classification via efficient multi-token prediction. arXiv preprint arXiv:2504.03159. find
  23. J. Ross Quinlan. 1986. Induction of decision trees. Machine learning, 1(1):81–106. find
  24. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824– 24837.
  25. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025. Qwen3 technical report. arXiv preprint arXiv:2505.09388.
  26. Junchi Yu, Tingyang Xu, Yu Rong, Yatao Bian, Junzhou Huang, and Ran He. 2021. Recognizing predictive substructures with subgraph information bottleneck. IEEE transactions on pattern analysis and machine intelligence, 46(3):1650–1663. find
  27. Yangyang Yu, Zhiyuan Yao, Haohang Li, Zhiyang Deng, Yuechen Jiang, Yupeng Cao, Zhi Chen, Jordan Suchow, Zhenyu Cui, Rong Liu, and 1 others. 2024. find
  28. Fincon: A synthesized llm multi-agent system with conceptual verbal reinforcement for enhanced financial decision making. Advances in Neural Information Processing Systems, 37:137010–137045. Liangliang Zhang, Zhuorui Jiang, Hongliang Chi, Haoyang Chen, Mohammed Elkoumy, Fali Wang, Qiong Wu, Zhengyi… — [arXiv:2505.23495](https://arxiv.org/abs/2505.23495) find
  29. Shengming Zhang, Hao Zhong, Zixuan Yuan, and Hui Xiong. 2021. Scalable heterogeneous graph neural networks for predicting high-potential early-stage startups. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, pages 2202–2211. find
  30. Wentao Zhang, Lingxuan Zhao, Haochong Xia, Shuo Sun, Jiaze Sun, Molei Qin, Xinyi Li, Yuqing Zhao, Yilei Zhao, Xinyu Cai, and 1 others. 2024. A multimodal foundation agent for financial trading: Tool-augmented, diversified, and generalist. In Proceedings of the 30th acm sigkdd conference on knowledge… find
  31. Han Zhou, Xingchen Wan, Lev Proleev, Diana Mincu, Jilin Chen, Katherine Heller, and Subhrajit Roy. 2023. Batch calibration: Rethinking calibration for in-context learning and prompt engineering. arXiv preprint arXiv:2309.17249. — [arXiv:2309.17249](https://arxiv.org/abs/2309.17249) find
  32. Bob Zider. 1998. How venture capital works. Harvard business review, 76(6):131–139. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/pei-2026-analyze-like-venture-capitalist-information/