Elamur Research⌕
☰
home / papers / 2025 / ReAgent: Reversible Multi-Agent Reasoning for Knowledge-Enhanced Multi-Hop QA

ReAgent: Reversible Multi-Agent Reasoning for Knowledge-Enhanced Multi-Hop QA

Paper · 2025 · Zhao Xinjie, Fan Gao, Xingyu Song et al.

Summary

Multi-hop QA requires chaining evidence from several sources. Standard LLM pipelines, including CoT prompting, retrieval-augmented generation and many multi-agent systems, move only forward. An error in an early hop carries through to the end, and contradictions found later have no structured way to force a revision. The authors introduce ReAgent, a multi-agent framework in which reasoning steps can be undone. An Execution Layer decomposes the question, retrieves evidence from both text and graph sources, verifies it and assembles an answer. A Supervisory Layer monitors overall consistency, and an Interaction Layer handles messaging and logging. When a conflict appears, agents return to an earlier valid state. This happens either locally, within one sub-question, or globally, after the system isolates a minimal set of conflicting claims. ReAgent was tested on HotpotQA, 2WikiMultiHopQA and Musique. It gains about 6% accuracy on average over the strongest forward-only baselines and produces more interpretable reasoning traces. It even beats some dedicated reasoning models while running on a lightweight non-reasoning LLM. The results suggest that revisable, non-monotonic inference is a practical lever for knowledge-intensive QA.

Key points

  • Agent roles: Decomposer (A_Q), Retriever (A_R), Verifier (A_V), Answer-Assembler (A_A), Supervisor (A_S) and Controller (A_C). They communicate through a messaging channel that keeps a persistent log and a temporal tracker of states.
  • Two rollback levels:
    • Local backtracking is triggered by the Verifier, for example when two sources give conflicting populations for Sacramento.
    • Global backtracking is triggered by the Supervisor, which identifies a minimal conflict set and escalates.
  • Evaluation: the system was tested on HotpotQA, 2WikiMultiHopQA and Musique, with an average gain of about 6% over the strongest baselines. The code is released on GitHub.
  • Case study: the question asks for the official fight song of the University of Kansas.
    • O3 and the CoA majority-vote system answer with the popular rally tune "I'm a Jayhawk".
    • ReAgent detects the conflict, backtracks and re-weights the evidence. It selects "Kansas Song" (adopted 1928), whose source explicitly marks it as official.
  • Stated limitations: repeated backtracking increases inference time. The more complex architecture may also be less robust when resources are limited or inputs are noisy.

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

🎓National University of SingaporeSingapore · Singapore180 works · 34,304 citations🎓Hong Kong University of Science and TechnologyHong Kong · China142 works · 25,136 citations🎓The University of TokyoTokyo · Japan90 works · 9,156 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 2Research subjects and languages the work is about.Reasoning · 1,426 worksQuestion answering · 441 works

Cite

Cite as Zhao Xinjie, Fan Gao, Xingyu Song, Yingjian Chen, Rui Yang, Yanran Fu, Yuyang Wang, Yusuke Iwasawa, Yutaka Matsuo and Irene Li. 2025. ReAgent: Reversible Multi-Agent Reasoning for Knowledge-Enhanced Multi-Hop QA. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.emnlp-main.202

@inproceedings{xinjie2025reagent,
  title = {{ReAgent: Reversible Multi-Agent Reasoning for Knowledge-Enhanced Multi-Hop QA}},
  author = {Zhao Xinjie, Fan Gao, Xingyu Song, Yingjian Chen, Rui Yang, Yanran Fu, Yuyang Wang, Yusuke Iwasawa, Yutaka Matsuo and Irene Li},
  booktitle = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
  year = {2025},
  doi = {10.18653/v1/2025.emnlp-main.202},
  url = {https://doi.org/10.18653/v1/2025.emnlp-main.202}
}

References

48 references, 20 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Xiaohe Bo, Zeyu Zhang, Quanyu Dai, Xueyang Feng, Lei Wang, Rui Li, Xu Chen, and Ji-Rong Wen. 2024. Reflective multi-agent collaboration based on large language models. arXiv preprint arXiv:2409.14880. Presented as a poster at NeurIPS 2024. — [arXiv:2409.14880](https://arxiv.org/abs/2409.14880) find
  2. Rajarshi Das, Shehzaad Dhuliawala, Manzil Zaheer, Luke Vilnis, Ishan Durugkar, Akshay Krishnamurthy, Alex Smola, and Andrew McCallum. 2018. Go for a walk and arrive at the answer: Reasoning over paths in knowledge bases using reinforcement learning. In International Conference on Learning Representa… find
  3. DeepSeek-AI. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. Preprint, arXiv:2501.12948.
  4. Jon Doyle. 1979. A truth maintenance system. Artificial Intelligence, 12(3):231–272. find
  5. Fan Gao, Hang Jiang, Rui Yang, Qingcheng Zeng, Jinghui Lu, Moritz Blum, Dairui Liu, Tianwei She, Yuang Jiang, and Irene Li. 2023. Evaluating large language models on wikipedia-style survey generation. In Annual Meeting of the Association for Computational Linguistics. find
  6. Gemini, Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, and 1 others. 2023. Gemini: a family of highly capable multimodal models. arXiv preprint arXiv:2312.11805.
  7. Google DeepMind. 2025. Gemini 2.5: Our most intelligent ai model. Accessed: 2025-05-16. find
  8. Bernal Jiménez Gutiérrez, Yiheng Shu, Yu Gu, Michihiro Yasunaga, and Yu Su. 2024. Hipporag: Neurobiologically inspired long-term memory for large language models. In The Thirty-eighth Annual Conference on Neural Information Processing Systems.
  9. Bernal Jiménez Gutiérrez, Yiheng Shu, Yu Gu, Michihiro Yasunaga, and Yu Su. 2025. Hipporag: Neurobiologically inspired long-term memory for large language models. Preprint, arXiv:2405.14831.
  10. Gaole He, Yunshi Lan, Jing Jiang, Wayne Xin Zhao, and Ji-Rong Wen. 2021. Improving multi-hop knowledge base question answering by learning intermediate supervision signals. In Proceedings of the 14th ACM International Conference on Web Search and Data Mining (WSDM), pages 553–561. ACM. find
  11. Xanh Ho, Anh-Khoa Duong Nguyen, Saku Sugawara, and Akiko Aizawa. 2020. Constructing a multi-hop QA dataset for comprehensive evaluation of reasoning steps. In Proceedings of the 28th International Conference on Computational Linguistics, pages 6609–6625, Barcelona, Spain (Online). International Comm…
  12. Siyuan Huang, Zhiyuan Ma, Jintao Du, Changhua Meng, Weiqiang Wang, and Zhouhan Lin. 2024. Mirror-consistency: Harnessing inconsistency in majority voting. arXiv preprint arXiv:2410.10857. find
  13. Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, and 1 others. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276.
  14. Naoya Inoue, Pontus Stenetorp, and Kentaro Inui. 2020. R4c: A benchmark for evaluating rc systems to get the right answer for the right reason. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL), pages 6740–6750, Online. Association for Computational Lin… find
  15. Yuhe Ke, Rui Yang, Sui An Lie, Taylor Xin Yi Lim, Yilin Ning, Irene Li, Hairil Rizal Abdullah, Daniel Shu Wei Ting, and Nan Liu. 2024. Mitigating cognitive biases in clinical decision-making through multi-agent conversations using large language models: Simulation study. J Med Internet Res, 26:e5943… find
  16. Akbir Khan, John Hughes, Dan Valentine, Laura Ruis, Kshitij Sachan, Ansh Radhakrishnan, Edward Grefenstette, Samuel R. Bowman, Tim Rocktäschel, and Ethan Perez. 2024. Debating with More Persuasive LLMs Leads to More Truthful Answers. arXiv preprint arXiv:2402.06782. find
  17. Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin, and Bernard Ghanem. 2023. CAMEL: Communicative Agents for "Mind" Exploration of Large Language Model Society. In Advances in Neural Information Processing Systems (NeurIPS).
  18. Lei Liang, Mengshu Sun, Zhengke Gui, Zhongshu Zhu, Zhouyu Jiang, Ling Zhong, Yuan Qu, Peilong Zhao, Zhongpu Bo, Jin Yang, Huaidong Xiong, Lin Yuan, Jun Xu, Zaoyang Wang, Zhiqiang Zhang, Wen Zhang, Huajun Chen, Wenguang Chen, and Jun Zhou. 2024. Kag: Boosting llms in professional domains via knowledg… — [arXiv:2409.13731](https://arxiv.org/abs/2409.13731) find
  19. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, and 1 others. 2024. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437. find
  20. Dairui Liu, Boming Yang, Honghui Du, Derek Greene, Neil J. Hurley, Aonghus Lawlor, Ruihai Dong, and Irene Li. 2023. Recprompt: A self-tuning prompting framework for news recommendation using large language models. In International Conference on Information and Knowledge Management. find
  21. Alex Long, Joel Mason, Alan Blair, and Wei Wang. 2019. Multi-hop reading comprehension via deep reinforcement learning based document traversal. arXiv preprint arXiv:1905.09438. — [arXiv:1905.09438](https://arxiv.org/abs/1905.09438) find
  22. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Nicholas Lourie, Gabriel Ilharco, Daphne Ippolito, Sam Singh, and 1 others. 2023. Self-refine: Iterative refinement with self-feedback. In Advances in Neural Information Processing Systems (NeurIPS). find
  23. Meta AI. 2024. Introducing llama 4: Advancing multimodal intelligence. find
  24. OpenAI. 2024. Openai o1 system card. Preprint, arXiv:2412.16720.
  25. OpenAI. 2025. Introducing openai o3 and o4-mini. Accessed: 2025-05-16. find
  26. Elham Parhizkar, Mohammad Hossein Nikravan, Robert C. Holte, and Sandra Zilles. 2020. Combining direct trust and indirect trust in multi-agent systems. In Proceedings of the 29th International Joint Conference on Artificial Intelligence (IJCAI), pages 311–317. find
  27. Ofir Press, Muru Zhang, Sewon Min, Ludwig Schmidt, Noah Smith, and Mike Lewis. 2023. Measuring and narrowing the compositionality gap in language models. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 5687–5711, Singapore. Association for Computational Linguistics. find
  28. Haritz Puerto, Gözde Gül Şahin, and Iryna Gurevych. 2023. Metaqa: Combining expert agents for multi-skill question answering. In Proceedings of the 17th Conference of the European Chapter of the ACL (EACL), pages 3566–3580, Dubrovnik, Croatia. Association for Computational Linguistics. find
  29. Qwen3. 2025. Qwen3 technical report. arXiv:2505.09388.
  30. Preprint, Noah Shinn, Federico Cassano, Edward Berman, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language Agents with Verbal Reinforcement Learning. In Advances in Neural Information Processing Systems (NeurIPS).
  31. Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal. 2022. MuSiQue: Multi-hop questions via single-hop question composition. Transactions of the Association for Computational Linguistics, 10:539–554. find
  32. Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal. 2023. Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), p…
  33. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed H. Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023. Self-consistency improves chain of thought reasoning in language models. In Proceedings of the International Conference on Learning Representations (ICLR).
  34. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. 2023. Chain-of-thought prompting elicits reasoning in large language models. Preprint, arXiv:2201.11903.
  35. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc Le, and Denny Zhou. 2022. Chain-of-thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems (NeurIPS), volume 35, pages 24824–24837.
  36. Johannes Welbl, Pontus Stenetorp, and Sebastian Riedel. 2018. Constructing datasets for multi-hop reading comprehension across documents. Transactions of the Association for Computational Linguistics, 6:287– 302.
  37. Kai Xiong, Xiao Ding, Yixin Cao, Ting Liu, and Bing Qin. 2023. Examining Inter-Consistency of Large Language Models Collaboration: An In-depth Analysis via Debate. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 7572–7590.
  38. An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, and 1 others. 2024a. Qwen2. 5 technical report. arXiv preprint arXiv:2412.15115. find
  39. Rui Yang, Haoran Liu, Edison Marrese-Taylor, Qingcheng Zeng, Yuhe Ke, Wanxin Li, Lechao Cheng, Qingyu Chen, James Caverlee, Yutaka Matsuo, and Irene Li. 2024b. KG-rank: Enhancing large language models for medical QA with knowledge graphs and ranking techniques. In Proceedings of the 23rd Workshop on… find
  40. Rui Yang, Yilin Ning, Emilia Keppo, Mingxuan Liu, Chuan Hong, Danielle S Bitterman, Jasmine Chiat Ling Ong, Daniel Shu Wei Ting, and Nan Liu. 2025a. Retrieval-augmented generation for generative artificial intelligence in health care. npj Health Systems, 2(1):2. find
  41. Rui Yang, Boming Yang, Aosong Feng, Sixun Ouyang, Moritz Blum, Tianwei She, Yuang Jiang, Freddy Lecue, Jinghui Lu, and Irene Li. 2024c. Graphusion: a rag framework for knowledge graph construction with a global perspective. arXiv preprint arXiv:2410.17600. — [arXiv:2410.17600](https://arxiv.org/abs/2410.17600) find
  42. Rui Yang, Boming Yang, Xinjie Zhao, Fan Gao, Aosong Feng, Sixun Ouyang, Moritz Blum, Tianwei She, Yuang Jiang, Freddy Lecue, Jinghui Lu, and Irene Li. 2025b. Graphusion: A rag framework for scientific knowledge graph construction with a global perspective. In Proceedings of the NLP4KGC Workshop at t… find
  43. Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William W. Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. 2018. Hotpotqa: A dataset for diverse, explainable multi-hop question answering. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP),…
  44. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. ReAct: Synergizing Reasoning and Acting in Language Models. In Proceedings of the International Conference on Learning Representations (ICLR).
  45. Yusen Zhang, Ruoxi Sun, Yanfei Chen, Tomas Pfister, Rui Zhang, and Sercan Ö. Arik. 2024. Chain of agents: Large language models collaborating on long-context tasks. Preprint, arXiv:2406.02818. find
  46. Zhuosheng Zhang, Aston Zhang, Mu Li, Hai Zhao, George Karypis, and Alexander J. Smola. 2023. Multimodal Chain-of-Thought Reasoning in Language Models. arXiv preprint arXiv:2302.00923.
  47. Jun Zhao, Can Zu, Xu Hao, Yi Lu, Wei He, Yiwen Ding, Tao Gui, Qi Zhang, and Xuanjing Huang. 2024. Longagent: Achieving question answering for 128k-token-long documents through multi-agent collaboration. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP)… find
  48. Yongchao Zhou, Andrei Ioan Muresanu, Ziwen Han, Keiran Paster, Silviu Pitis̆, Harris Chan, and Jimmy Ba. 2023. Large Language Models Are Human-Level Prompt Engineers. arXiv preprint arXiv:2211.01910.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xinjie-2025-reagent-reversible-multi-agent-reasoning/