CoChat
Sign Up
Cited by 56Topic ModelingArtificial Intelligence in Healthcare and EducationInformation Retrieval and Search Behavior

MEGA-RAG: a retrieval-augmented generation framework with multi-evidence guided answer refinement for mitigating hallucinations of LLMs in public health

Open access
Xu, Shan, Yan, Zhaokun, Dai, Chengxiao, Wu, Fan·2025·Frontiers in Public Health
DOI 10.3389/fpubh.2025.1635381PMC PMC12540348
Open this paper in CoChat — free accountOpen PDF

Abstract

Introduction: The increasing adoption of large language models (LLMs) in public health has raised significant concerns about hallucinations-factually inaccurate or misleading outputs that can compromise clinical communication and policy decisions. Methods: We propose a retrieval-augmented generation framework with multi-evidence guided answer refinement (MEGA-RAG), specifically designed to mitigate hallucinations in public health applications. The framework integrates multi-source evidence retrieval (dense retrieval via FAISS, keyword-based retrieval via BM25, and biomedical knowledge graphs), employs a cross-encoder reranker to ensure semantic relevance, and incorporates a discrepancy-aware refinement module to further enhance factual accuracy. Results: Experimental evaluation demonstrates that MEGA-RAG outperforms four baseline models [PubMedBERT, PubMedGPT, standalone LLM, and LLM with standard retrieval-augmented generation (RAG)], achieving a reduction in hallucination rates by over 40%. It also achieves the highest accuracy (0.7913), precision (0.7541), recall (0.8304), and F1 score (0.7904). Discussion: These findings confirm that MEGA-RAG is highly effective in generating factually reliable and medically accurate responses, thereby enhancing the credibility of AI-generated health information for applications in health education, clinical communication, and evidence-based policy development.

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.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.3389/fpubh.2025.1635381

Cites

4 of 4 references verified0 citation errors · 0 not checkable

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.

A few free checks a day · papers already verified open instantly · results are public

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

Flagged first, then by citations

Research works printed but not linked: 31

  1. [1]Kraemer MU Tsui JLH Chang SY Lytras S Khurana MP Vanderslott S. Artificial intelligence for modelling infectious disease epidemics. Nature. (2025) 638: 623– 35. 10.1038/s41586-024-08564-w 39972226 PMC11987553 · DOI
  2. [2]Hossain MI Zamzmi G Mouton PR Salekin MS Sun Y Goldgof D. Explainable AI for medical data: current methods, limitations, and future directions. ACM Comput Surv. (2025) 57: 1– 46. 10.1145/3637487 · DOI
  3. [3]Wang J Weiss JC. A large-language model framework for relative timeline extraction from pubmed case reports. arXiv preprint arXiv:250412350 (2025). PMC12150726 40502242
  4. [4]Goh E Bunning B Khoong EC Gallo RJ Milstein A Centola D. Physician clinical decision modification and bias assessment in a randomized controlled trial of AI assistance. Commun Med. (2025) 5: 59. 10.1038/s43856-025-00781-2 40038550 PMC11880198 · DOI
  5. [6]Pal A Umapathi LK Sankarasubbu M. Med-halt: Medical domain hallucination test for large language models. arXiv preprint arXiv:230715343 (2023). 10.18653/v1/2023.conll-1.21 · DOI · Paper page
  6. [7]Zhou J Chen AZ Shah D Reese LS De Choudhury M. "It's a conversation, not a quiz": a risk taxonomy and reflection tool for LLM adoption in public health. arXiv preprint arXiv:241102594 (2024).
  7. [9]Beaglehole R Bonita R. Public Health at the Crossroads: Achievements and Prospects. New York: Cambridge University Press (2004).
  8. [11]Friel R Sanyal A. Chainpoll: a high efficacy method for LLM hallucination detection. arXiv preprint arXiv:231018344 (2023).
  9. [12]He L Li K. Mitigating hallucinations in LLM using k-means clustering of synonym semantic relevance. TechRxiv. (2024). 10.36227/techrxiv.171822241.11082054/v1 · DOI
  10. [13]Kim Y Jeong H Chen S Li SS Lu M Alhamoud K. Medical hallucinations in foundation models and their impact on healthcare. arXiv preprint arXiv:250305777 (2025). 10.1101/2025.02.28.25323115 · DOI
  11. [14]Galitsky B. Truth-o-meter: collaborating with LLM in fighting its hallucinations. In: Interdependent Human-Machine Teams (Elsevier) (2025). p. 175– 210. 10.1016/B978-0-443-29246-0.00004-3 · DOI
  12. [15]Chakraborty N Ornik M Driggs-Campbell K. Hallucination detection in foundation models for decision-making: a flexible definition and review of the state of the art. ACM Comput Surv. (2025) 57: 1– 35. 10.1145/3716846 · DOI
  13. [16]Kumar A Agarwal C Srinivas S Li AJ Feizi S Lakkaraju H. Certifying LLM safety against adversarial prompting. arXiv preprint arXiv:230902705 (2023).
  14. [17]Wang C Xu Y Peng Z Zhang C Chen B Wang X. keqing: knowledge-based question answering is a nature chain-of-thought mentor of LLM. arXiv preprint arXiv:240100426 (2023).
  15. [18]Wang X Yang Q Qiu Y Liang J He Q Gu Z. Knowledgpt: Enhancing large language models with retrieval and storage access on knowledge bases. arXiv preprint arXiv:230811761 (2023).
  16. [19]Sriramanan G Bharti S Sadasivan VS Saha S Kattakinda P Feizi S. Llm-check: Investigating detection of hallucinations in large language models. Adv Neural Inf Process Syst. (2024) 37: 34188– 216.
  17. [20]Heo S Son S Park H. HaluCheck: Explainable and verifiable automation for detecting hallucinations in LLM responses. Expert Syst Applic. (2025) 272: 126712. 10.1016/j.eswa.2025.126712 · DOI
  18. [22]Nguyen H He Z Gandre SA Pasupulety U Shivakumar SK Lerman K. Smoothing out hallucinations: mitigating LLM hallucination with smoothed knowledge distillation. arXiv preprint arXiv:250211306 (2025).
  19. [23]Azamfirei R Kudchadkar SR Fackler J. Large language models and the perils of their hallucinations. Critical Care. (2023) 27: 120. 10.1186/s13054-023-04393-x 36945051 PMC10032023 · DOI
  20. [24]Zhou X Zhou J Wang C Xie Q Ding K Mao C. PH-LLM: public health large language models for infoveillance. medRxiv. (2025). p. 2025-02. 10.1101/2025.02.08.25321587 · DOI
  21. [25]De Angelis L Baglivo F Arzilli G Privitera GP Ferragina P Tozzi AE. ChatGPT and the rise of large language models: the new AI-driven infodemic threat in public health. Front Public Health. (2023) 11: 1166120. 10.3389/fpubh.2023.1166120 37181697 PMC10166793 · DOI · Paper page
  22. [27]Matsumoto N Moran J Choi H Hernandez ME Venkatesan M Wang P. KRAGEN: a knowledge graph-enhanced RAG framework for biomedical problem solving using large language models. Bioinformatics. (2024) 40: btae353. 10.1093/bioinformatics/btae353 38830083 PMC11164829 · DOI
  23. [28]Soman K Langdon A Villouta C Agrawal C Salta L Peetoom B. Zebra-Llama: a context-aware large language model for democratizing rare disease knowledge. arXiv preprint arXiv:2411.02657 (2024). · arXiv
  24. [29]Zhao X Liu S Yang SY Miao C. MedRAG: enhancing retrieval-augmented generation with knowledge graph-elicited reasoning for healthcare copilot. In: Proceedings of the ACM on Web Conference 2025. WWW '25. New York, NY, USA: Association for Computing Machinery (2025). p. 4442– 4457. 10.1145/3696410.3714782 · DOI · Paper page
  25. [30]Hogan A Blomqvist E Cochez M D'amato C Melo GD Gutierrez C. Knowledge graphs. ACM Comput Surv. (2021) 54: 1– 37. 10.1145/3447772 · DOI
  26. [31]Douze M Guzhva A Deng C Johnson J Szilvasy G Mazaré PE. The Faiss library. arXiv preprint arXiv:2401.08281 (2025). · arXiv
  27. [32]Li X Lipp J Shakir A Huang R Li J. BMX: Entropy-weighted similarity and semantic-enhanced lexical search. arXiv preprint arXiv:2408.06643 (2024). · arXiv
  28. [33]Rosa G Bonifacio L Jeronymo V Abonizio H Fadaee M Lotufo R. In defense of cross-encoders for zero-shot retrieval. arXiv preprint arXiv:2212.06121 (2022). · arXiv
  29. [34]Steck H Ekanadham C Kallus N. Is cosine-similarity of embeddings really about similarity? In: Companion Proceedings of the ACM Web Conference 2024. ACM (2024). p. 887– 890. 10.1145/3589335.3651526 · DOI
  30. [35]Hanna M Bojar O. A fine-grained analysis of BERTScore. In: Proceedings of the Sixth Conference on Machine Translation (2021). p. 507– 517.
  31. [37]Wang X Wei J Schuurmans D Le Q Chi E Narang S. Self-consistency improves chain of thought reasoning in language models. arXiv preprint arXiv:2203.11171 (2023). · arXiv

Cited by

Sorted by citations

Authors
Xu, S., Yan, Z., Dai, C., & Wu, F.
Affiliations
  • Fudan University (CN)
  • China Academy of Information and Communications Technology (CN)
  • The University of Sydney (AU)
Corresponding author
Wu, Fan
ORCID
Venue
Frontiers in Public Health
Publisher
Frontiers Media SA
Published
2025-10-08
Volume
13
Pages
1635381-1635381
ISSN
2296-2565 (ISSN-L)
DOI
10.3389/fpubh.2025.1635381
PMC
PMC12540348
OpenAlex
W4414925442
Cited by
56
References
4
Citations by year
2026: 53 · 2025: 3
FWCI
18.25
Type
journal article
Language
en
Open access
Licence
Locations
  • Frontiers in Public Health (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Physical Sciences › Computer Science › Information Systems › Information Retrieval and Search Behavior
Keywords
retrieval-augmented generation, hallucination mitigation, large language models, dense retrieval, biomedical knowledge graphs, cross-encoder reranking, factual accuracy, public health, clinical communication, health education, evidence-based policy
MeSH terms
Hallucinations, Humans, Language, Public Health, Information Storage and Retrieval

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