TreeReview: A Dynamic Tree of Questions Framework for Deep and Efficient LLM-based Scientific Peer Review
Open accessAbstract
Yuan Chang, Ziyue Li, Hengyuan Zhang, Yuanbo Kong, Yanru Wu, Hayden Kwok-Hay So, Zhijiang Guo, Liya Zhu, Ngai Wong. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The annotation is bidirectional: AI and human review items are subject to the same scrutiny rather than human reviews being treated as the implicit reference for AI.
“The annotation is bidirectional, in that AI and human review items are subject to the same scrutiny rather than human reviews being treated as the implicit reference for AI [Chang et al., 2025, Garg et al., 2025, Idahl and Ahmadi, 2025, Bougie and Watanabe, 2025].”
From On the limits and opportunities of AI reviewers: Reviewing the reviews of Nature-family papers with 45 expert scientists (2026) · cites this paper as Chang et al., 2025
Section 4.3 treats human reviewer comments as the reference for evaluating AI-generated feedback: it merges them into a 'comprehensive ground truth.' Section 4.4 describes human evaluators comparing reviews generated by different methods, not subjecting human and AI review items to the same scrutiny. The paper therefore describes the opposite of the claimed bidirectional annotation. (LLM, conf=0.90)
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredTreeReview models peer review as a hierarchical question-answering process, beginning with top-down decomposition into a tree of sub-questions.
“TreeReview (Chang et al., 2025) models peer review as a hierarchical question-answering process.The process begins with a top-down decomposition stage, where a high-level review task is recursively broken down into a tree of more finegrained sub-questions.”
From Can AI Be a Good Peer Reviewer? A Survey of Peer Review Process, Evaluation, and the Future (2026) · cites this paper as Chang et al., 2025
TreeReview evaluates individual feedback comments through embedding-based and LLM-based matching for coverage and specificity.
“TreeReview evaluates individual feedback comments through embedding-based and LLM-based matching for coverage and specificity [Chang et al., 2025]”
From On the limits and opportunities of AI reviewers: Reviewing the reviews of Nature-family papers with 45 expert scientists (2026) · cites this paper as Chang et al., 2025
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
The publisher hasn’t deposited this paper’s reference list.
Research works printed but not linked: 13
- Zhong-Zhi Li, Duzhen Zhang, Ming-Liang Zhang, Ji- axin Zhang, Zengyan Liu, Yuxuan Yao, Haotian Xu, Junhao Zheng, Pei-Jie Wang, Xiuyi Chen, and 1 oth- ers. 2025b. From system 1 to system 2: A survey of reasoning large language models. arXiv preprint arXiv:2502.17419. · arXiv
- Ziyue Li, Yuan Chang, and Xiaoqiu Le. 2024d. Simulat- ing expert discussions with multi-agent for enhanced scientific problem solving. In Proceedings of the Fourth Workshop on Scholarly Document Process- ing (SDP 2024), pages 243-256, Bangkok, Thailand. Association for Computational Linguistics.
- Chin-Yew Lin. 2004. ROUGE: A package for auto- matic evaluation of summaries. In Text Summariza- tion Branches Out, pages 74-81, Barcelona, Spain. Association for Computational Linguistics.
- Renze Lou, Hanzi Xu, Sijia Wang, Jiangshu Du, Ryo Kamoi, Xiaoxin Lu, Jian Xie, Yuxuan Sun, Yusen Zhang, Jihyun Janice Ahn, Hongchao Fang, Zhuoyang Zou, Wenchao Ma, Xi Li, Kai Zhang, Con- gying Xia, Lifu Huang, and Wenpeng Yin. 2024. AAAR-1.0: Assessing AI's potential to assist re- search. In 2nd AI4Research Workshop: Towards a Knowledge-grounded Scientific Research Lifecycle.
- Qusay H Mahmoud and 1 others. 2024. Evaluating the efficacy of large language models in automat- ing academic peer reviews. In 2024 International Conference on Machine Learning and Applications (ICMLA), pages 1208-1213. IEEE.
- Guanchao Wang, Pawin Taechoyotin, Tong Zeng, Bradley Sides, and Daniel Acuna. 2024a. MAMORX: Multi-agent multi-modal scien- tific review generation with external knowledge. In Neurips 2024 Workshop Foundation Models for Science: Progress, Opportunities, and Challenges.
- Yixuan Weng, Minjun Zhu, Guangsheng Bao, Hongbo Zhang, Jindong Wang, Yue Zhang, and Linyi Yang. 2025. Cycleresearcher: Improving automated re- search via automated review. In The Thirteenth Inter- national Conference on Learning Representations.
- Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2020. Bertscore: Evalu- ating text generation with BERT. In 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020. OpenRe- view.net.
- Wenqing Zheng, SP Sharan, Ajay Kumar Jaiswal, Kevin Wang, Yihan Xi, Dejia Xu, and Zhangyang Wang. 2023a. Outline, then details: Syntactically guided coarse-to-fine code generation. In International Con- ference on Machine Learning, pages 42403-42419.
- PMLR.
- Denny Zhou, Nathanael Schärli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuurmans, Claire Cui, Olivier Bousquet, Quoc V Le, and Ed H. Chi. 2023. Least-to-most prompting enables com- plex reasoning in large language models. In The Eleventh International Conference on Learning Rep- resentations.
- You specialize in providing precise, evidence-based answers to review questions for submitted paper. You operate at the leaf-node level of a peer-review question tree. Your answers will directly support higher-level critique synthesis. TASK REQUIREMENTS:
- Provide Detailed Evidence: For each comment, include specific evidence from the given context (e.g., quotes, section references, or data points) to justify your point. INPUT: -Review Question: {QUESTION} -Relevant Context: {CONTEXT} OUTPUT FORMAT: A single string containing only the answer to the review question. Your final answer: Figure 14: Prompt for the Answer Synthesizer in answering leaf questions.
Cited by
Sorted by citations
Torrielli, F., Locci, S., Rapp, A., & Di, L. (2026) · Scientometrics · doi:10.1007/s11192-026-05695-x · Cited by 4 · see citing works
Kim, S., Yoon, D., Gashteovski, K., Suk, J., Baek, J., Aggarwal, P., Wu, I., Zaverkin, V., Petkoski, S., Schrider, D. R., et al. (2026) · arXiv · doi:10.48550/arxiv.2605.20668 · Cited by 1 · see citing works
Can AI Be a Good Peer Reviewer? A Survey of Peer Review Process, Evaluation, and the Future
Wu, S., Jiang, O., Zhao, Y., Hu, T., Ma, Y., Zhang, K., Patwardhan, M., & Cohan, A. (2026) · Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2026.acl-long.1504 · Cited by 1 · see citing works
AI-Assisted Peer Review at Scale: The AAAI-26 AI Review Pilot
Biswas, J., Schoepp, S., Vasan, G., Opipari, A., Zhang, A., Hu, Z., Joseph, S., Lease, M., Li, J. J., Stone, P., et al. (2026) · arXiv · doi:10.48550/arxiv.2604.13940 · Cited by 1 · see citing works
Dycke, N., & Gurevych, I. (2025) · arXiv · doi:10.48550/arxiv.2508.21422 · Cited by 1
PRISM: A Multi-Dimensional Benchmark for Evaluating LLM Peer Reviewers
Loc, N. P. P., La Viet, T. H., Khanh, T. T., Nguyen, D. A., Pham, T. A. N., Nguyen, T., Chawla, N. V., Buntine, W., Wong, K.-S., Doan, K. D., et al. (2026) · arXiv · doi:10.48550/arxiv.2605.26730
CoCoReviewBench: A Completeness- and Correctness-Oriented Benchmark for AI Reviewers
Deng, H., Ke, X., Li, Y., Hu, R., Huang, D., Wong, D. F., Wang, Y., Liu, X., & Zhang, M. (2026) · arXiv · doi:10.48550/arxiv.2605.07905
Beyond Final Decisions: A Process-Centric Benchmark for Transparent AI-Assisted Peer Review
Yuan, S., Zhang, X., Ni, W., Xiao, T., Di, S., Zhu, J., Jiang, Z., Tan, R., Chen, L., & Ren, K. (2026) · arXiv · doi:10.48550/arxiv.2609.05947
(2344960), Y. G., Yu, L., & Li, H. (2026) · Journal of Superintelligence · doi:10.3390/superintelligence1020008
Kim, E., & Singh, V. (2026) · Publications · doi:10.3390/publications14040063
- Authors
- Chang, Y., Li, Z., Zhang, H., Kong, Y., Wu, Y., So, H. K.-H., Guo, Z., Zhu, L., & Wong, N.
- Affiliations
- Chinese Academy of Sciences (CN)
- University of Hong Kong (HK)
- Tsinghua University (CN)
- The Hong Kong University of Science and Technology (Guangzhou) (CN)
- Venue
- Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing
- Published
- 2025
- Pages
- 15662-15693
- OpenAlex
- W4416036292
- Cited by
- 2
- References
- 0
- Citations by year
- 2026: 2
- FWCI
- 1.91
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Conference on Empirical Methods in Natural Language Processing (EMNLP) (publishedVersion, cc-by)
- Funders
- Chinese Academy of Sciences
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Social Sciences › Arts and Humanities › History and Philosophy of Science › Academic Writing and Publishing
- Physical Sciences › Computer Science › Information Systems › Expert finding and Q&A systems
- Keywords
- scientific peer review, large language models, question generation
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
