Elamur Research⌕
☰
home / papers / 2026 / Beyond One-Size-Fits-All: Multi-Agent Refinement Framework for Persona-Based Biomedical Summarization

Beyond One-Size-Fits-All: Multi-Agent Refinement Framework for Persona-Based Biomedical Summarization

Paper · 2026 · Rohan Charudatt Salvi, Chirag Chawla, Md. Shad Akhtar et al.

Summary

Most biomedical lay summarization treats non-experts as one homogeneous group. In practice, a layperson, a pre-medical student, a researcher from another field and a medical expert differ in how much detail and terminology they need. The authors introduce MAPS (Multi-Agent Persona-based Summarization), which works in three stages:

  1. Several LLM agents each draft a persona-conditioned summary of an abstract.
  2. Over several rounds, each agent receives critique from another agent in a round-robin scheme and revises its draft. The critique follows either persona-centered guidance (optionally with Wikipedia lookup) or the PERCS error taxonomy.
  3. A final adjudication step uses a judge, voting, or a judge that merges the candidates.

On the PERCS dataset, human raters judged MAPS clearly better than single-model baselines for the more advanced personas, and roughly equal for laypeople. Standard automatic metrics did not reflect these gains, and LLM judges gave inflated scores and detected errors poorly. The work matters because it shows audience-tailored summarization benefits from multi-agent refinement, while current evaluation tools are not reliable for this task.

Key points

  • Setup: PERCS contains 500 abstracts with 2,000 expert-written summaries for four personas. Experiments used the 150-abstract test split, k=4 agents and T=3 feedback rounds. Agents were drawn from GPT-4o, Mixtral-8x7B-Instruct, Gemini-2.0 Flash Lite and LLaMA-3 70B.
  • Human results: For the Researcher persona, comprehensiveness climbed to 4.74 compared with Mixtral's 3.36. For the Expert persona, it reached 4.85 compared with GPT-4o's 4.33.
  • Human evaluation design: Two persona-matched Upwork raters per persona scored 30 abstracts each. Inter-rater agreement (Krippendorff's α) exceeded 0.84.
  • Automatic and LLM evaluation: ROUGE, SARI, FKGL, DCRS, CLI and LENS showed no consistent advantage for MAPS. LLaMA-3 was chosen as the LLM judge because its error detection matched human judgments most closely.
  • Cost: MAPS uses about 49.7K tokens (~$0.0575) per sample, versus roughly 3–9K tokens for few-shot and self-refine baselines.

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 Illinois ChicagoChicago · United States61 works · 20,107 citations🎓Indian Institute of Technology BHUVaranasi · India10 works · 548 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksBiochemistry, Genetics and Molecular Biology · 234 worksMolecular Biology · 231 works

🧭 Topics 1Research subjects and languages the work is about.Summarization · 232 works

Cite

Cite as Rohan Charudatt Salvi, Chirag Chawla, Md. Shad Akhtar and Shweta Yadav. 2026. Beyond One-Size-Fits-All: Multi-Agent Refinement Framework for Persona-Based Biomedical Summarization. Proceedings of the Third Workshop on Patient-Oriented Language Processing (CL4Health) @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/22uei8f8k9za

@inproceedings{salvi2026beyond,
  title = {{Beyond One-Size-Fits-All: Multi-Agent Refinement Framework for Persona-Based Biomedical Summarization}},
  author = {Rohan Charudatt Salvi, Chirag Chawla, Md. Shad Akhtar and Shweta Yadav},
  booktitle = {Proceedings of the Third Workshop on Patient-Oriented Language Processing (CL4Health) @ LREC 2026},
  year = {2026},
  doi = {10.63317/22uei8f8k9za},
  url = {https://doi.org/10.63317/22uei8f8k9za}
}

References

28 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Biaoyan Fang, Xiang Dai, and Sarvnaz Karimi. 2024. Understanding faithfulness and reasoning of large language models on plain biomedical summaries. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 9890–9911, Miami, Florida, USA. Association for Computational Linguistic… find
  2. Dengzhao Fang, Jipeng Qiang, Xiaoye Ouyang, Yi Zhu, Yunhao Yuan, and Yun Li. 2025. Collaborative document simplification using multi-agent systems. In Proceedings of the 31st International Conference on Computational Linguistics, pages 897–912. find
  3. Tomas Goldsack, Carolina Scarton, and Chenghua Lin. 2025. Leveraging large language models for zero-shot lay summarisation in biomedicine and beyond. arXiv preprint arXiv:2501.05224. — [arXiv:2501.05224](https://arxiv.org/abs/2501.05224) find
  4. Tomas Goldsack, Zhihao Zhang, Chen Tang, Carolina Scarton, and Chenghua Lin. 2023. Enhancing biomedical lay summarisation with external knowledge graphs. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 8016–8032, Singapore. Association for Computation… find
  5. Yue Guo, Tal August, Gondy Leroy, Trevor Cohen, and Lucy Lu Wang. 2024. APPLS: Evaluating evaluation metrics for plain language summarization. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 9194–9211, Miami, Florida, USA. Association for Computationa… find
  6. Yue Guo, Wei Qiu, Yizhong Wang, and Trevor Cohen. 2021. Automated lay language summarization of biomedical scientific reviews. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 160–168. find
  7. Sebastian Joseph, Lily Chen, Jan Trienes, Hannah Göke, Monika Coers, Wei Xu, Byron Wallace, and Junyi Jessy Li. 2024. FactPICO: Factuality evaluation for plain language summarization of medical evidence. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volu… find
  8. Stuart RF King, Emma Pewsey, and Sarah Shailes. 2017. An inside guide to elife digests. ELife, 6:e25410. find
  9. Chin-Yew Lin. 2004. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out, pages 74–81.
  10. Zheheng Luo, Qianqian Xie, and Sophia Ananiadou. 2022. Readability controllable biomedical document summarization. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 4667–4680, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics. find
  11. Chen Lyu and Gabriele Pergola. 2024. Society of medical simplifiers. In Proceedings of the Third Workshop on Text Simplification, Accessibility and Readability (TSAR 2024), pages 61– 68, Miami, Florida, USA. Association for Computational Linguistics. find
  12. Mounica Maddela, Yao Dou, David Heineman, and Wei Xu. 2022. Lens: A learnable evaluation metric for text simplification. arXiv preprint arXiv:2212.09739.
  13. Shufan Ming, Yue Guo, and Halil Kilicoglu. 2025. Towards knowledge-guided biomedical lay summarization using large language models. In Proceedings of the Second Workshop on Patient-Oriented Language Processing (CL4Health), pages 285–297, Albuquerque, New Mexico. Association for Computational Linguis… find
  14. Kaijie Mo and Renfen Hu. 2024. Expertease: A multi-agent framework for grade-specific document simplification with large language models. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 9080–9099. find
  15. Ankan Mullick, Sombit Bose, Rounak Saha, Ayan Bhowmick, Pawan Goyal, Niloy Ganguly, Prasenjit Dey, and Ravi Kokku. 2024. On the persona-based summarization of domain-specific documents. In Findings of the Association for Computational Linguistics: ACL 2024, pages 14291– 14307, Bangkok, Thailand. Ass… find
  16. Rohan Charudatt Salvi, Chirag Chawla, Dhruv Jain, Swapnil Panigrahi, Md Shad Akhtar, and Shweta Yadav. 2025a. Percs: Persona-guided controllable biomedical summarization dataset. arXiv preprint arXiv:2512.03340. — [arXiv:2512.03340](https://arxiv.org/abs/2512.03340) find
  17. Rohan Charudatt Salvi, Swapnil Panigrahi, Dhruv Jain, Shweta Yadav, and Md. Shad Akhtar. 2025b. Towards understanding LLM-generated biomedical lay summaries. In Proceedings of the Second Workshop on Patient-Oriented Language Processing (CL4Health), pages 260–268, Albuquerque, New Mexico. Association… find
  18. Junjie Song, Yiwen Liu, Dapeng Li, Yin Sun, Shukun Fu, Siqi Chen, and Yuji Cao. 2025. Balancing rewards in text summarization: Multi-objective reinforcement learning via hypervolume optimization. arXiv preprint arXiv:2510.19325. — [arXiv:2510.19325](https://arxiv.org/abs/2510.19325) find
  19. Sue Stableford and Wendy Mettger. 2007. Plain language: a strategic response to the health literacy challenge. Journal of public health policy, 28(1):71–93. find
  20. Hieu Tran, Zonghai Yao, Won Seok Jang, Sharmin Sultana, Allen Chang, Yuan Zhang, and Hong Yu. 2025a. Medreadctrl: Personalizing medical text generation with readability-controlled instruction learning. arXiv preprint arXiv:2507.07419. — [arXiv:2507.07419](https://arxiv.org/abs/2507.07419) find
  21. Hieu Tran, Zonghai Yao, Lingxi Li, and Hong Yu. 2025b. ReadCtrl: Personalizing text generation with readability-controlled instruction learning. In Proceedings of the Fourth Workshop on Intelligent and Interactive Writing Assistants (In2Writing 2025), pages 19–36, Albuquerque, New Mexico, US. Associ… find
  22. Manya Wadhwa, Xinyu Zhao, Junyi Jessy Li, and Greg Durrett. 2024. Learning to refine with fine-grained natural language feedback. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 12281–12308, Miami, Florida, USA. Association for Computational Linguistics. find
  23. Lucy Lu Wang, Julia Otmakhova, Jay DeYoung, Thinh Hung Truong, Bailey Kuehl, Erin Bransom, and Byron C Wallace. 2023. Automated metrics for medical multi-document summarization disagree with human evaluations. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics… find
  24. Zhihui Xie, Jie Chen, Liyu Chen, Weichao Mao, Jingjing Xu, and Lingpeng Kong. 2025. Teaching language models to critique via reinforcement learning. arXiv preprint arXiv:2502.03492. — [arXiv:2502.03492](https://arxiv.org/abs/2502.03492) find
  25. Wei Xu, Courtney Napoles, Ellie Pavlick, Quanze Chen, and Chris Callison-Burch. 2016. Optimizing statistical machine translation for text simplification. Transactions of the Association for Computational Linguistics, 4:401–415.
  26. Zhiwen You, Shruthan Radhakrishna, Shufan Ming, and Halil Kilicoglu. 2024. UIUC_BioNLP at BioLaySumm: An extract-then-summarize approach augmented with Wikipedia knowledge for biomedical lay summarization. In Proceedings of the 23rd Workshop on Biomedical Natural Language Processing, pages 132–143,… find
  27. Readability. Readability is measured based on the use of medical jargon, sentence structure, and whether explanations are provided for technical terms. 1. There is little difference between the plain-text summary and the original abstract. find
  28. Yinghao Zhu, Ziyi He, Haoran Hu, Xiaochen Zheng, Xichen Zhang, Zixiang Wang, Junyi Gao, Liantao Ma, and Lequan Yu. 2025. Medagentboard: Benchmarking multi-agent collaboration with conventional methods for diverse medical tasks. arXiv preprint arXiv:2505.12371. — [arXiv:2505.12371](https://arxiv.org/abs/2505.12371) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/salvi-2026-beyond-one-size-fits-all/