Elamur Research⌕
☰
home / papers / 2026 / MetaMiners at SMM4H-HeaRD 2026: A Semantic-Structural Knowledge-Enriched Ensemble for SARS-CoV-2 Metadata Identification

MetaMiners at SMM4H-HeaRD 2026: A Semantic-Structural Knowledge-Enriched Ensemble for SARS-CoV-2 Metadata Identification

Paper · 2026 · Claudia-Alexandra Ursu and Alecsandru-Florin Soare

Summary

The paper addresses SMM4H-HeaRD 2026 Task 5. The task is a binary classification of sentences from PubMed articles about SARS-CoV-2 sequences. The goal is to decide whether a sentence reports patient metadata, such as age, sex, location or symptoms, tied to the sequenced samples. Sentences that mention such data in unrelated, technical or prior-study contexts must be rejected. This is useful because repositories like GISAID and GenBank often have empty metadata fields, and manual curation cannot keep up during a pandemic. The corpus has 22,147 sentences, and only 13.3% are positive.

The authors enrich the input text with regex-based tags marking:

  • metadata values and their counts
  • "anti-metadata" technical cues
  • reporting versus procedural verbs
  • sentence focus and negation
  • scispaCy dependency frames that link attributes to a patient

On this enriched text they fine-tune four biomedical transformers, each with a different training trick. They then combine the models through a weighted average of logits. The ensemble reached F1 0.786 on the test set and ranked first, ahead of the earlier 0.776 BiomedBERT-Large baseline.

Key points

  • Data split: 15,504 train / 2,214 validation / 4,429 test sentences, with a strong class imbalance (13.3% positive).
  • Ensemble members and validation F1:
    • BioLinkBERT-Large with Focal Loss: 0.822
    • BiomedBERT-base-abstract with semantic frames and weighted cross-entropy: 0.815
    • PubMedBERT-fulltext with the bottom 8 layers frozen: 0.819
    • BioLinkBERT-Large with FGM adversarial training and label smoothing: 0.800
  • Ensemble result: weights of about 0.18–0.27 per model and a threshold of 0.55 gave validation F1 0.851 (P 0.836, R 0.867) and test F1 0.786, placing 1st.
  • Negation handling: the system scans a 60-character window after a negation cue and adds a [NEG] prefix to metadata found there. These negated items are excluded from the [META:N] count to cut false positives.
  • Error analysis: aggregate density tags ([META]/[ANTI]) were a main source of false positives. The authors plan to replace them with type-specific markers. No formal ablation study was run.

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 BucharestBucharest · Romania21 works · 516 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

Cite

Cite as Claudia-Alexandra Ursu and Alecsandru-Florin Soare. 2026. MetaMiners at SMM4H-HeaRD 2026: A Semantic-Structural Knowledge-Enriched Ensemble for SARS-CoV-2 Metadata Identification. Proceedings of the 11th Social Media Mining for Health Research and Applications (SMM4H-HeaRD 2026) Workshop and Shared Tasks.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.smm4h-1.45

@inproceedings{ursu2026metaminers,
  title = {{MetaMiners at SMM4H-HeaRD 2026: A Semantic-Structural Knowledge-Enriched Ensemble for SARS-CoV-2 Metadata Identification}},
  author = {Claudia-Alexandra Ursu and Alecsandru-Florin Soare},
  booktitle = {Proceedings of the 11th Social Media Mining for Health Research and Applications (SMM4H-HeaRD 2026) Workshop and Shared Tasks},
  year = {2026},
  doi = {10.18653/v1/2026.smm4h-1.45},
  url = {https://doi.org/10.18653/v1/2026.smm4h-1.45}
}

References

12 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Wendy W Chapman, Will Bridewell, Paul Hanbury, Gregory F Cooper, and Bruce G Buchanan. 2001. A simple algorithm for identifying negated findings and diseases in discharge summaries. Journal of Biomedical Informatics, 34(5):301–310. find
  2. Franck Dernoncourt and Ji Young Lee. 2017. Pubmed 200k rct: a dataset for sequential sentence classification in medical abstracts. In Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 2: Short Papers), pages 308–313, Taipei, Taiwan. Asian Federation of N…
  3. Nathan D. Grubaugh, Jason T. Ladner, Philippe Lemey, Oliver G. Pybus, Andrew Rambaut, Edward C. Holmes, and Kristian G. Andersen. 2019. Tracking virus outbreaks in the twenty-first century. Nat Microbiol, 4:10–19. find
  4. Mark Neumann, Daniel King, Iz Beltagy, and Waleed Ammar. 2019. ScispaCy: Fast and robust models for biomedical natural language processing. In Proceedings of the 18th BioNLP Workshop and Shared Task, pages 319–327, Florence, Italy. Association for Computational Linguistics.
  5. Eric W. Sayers, Mark Cavanaugh, Karen Clark, James Ostell, Kim D. Pruitt, and Ilene Karsch-Mizrachi. 2019. Genbank. Nucleic Acids Research, 47(D1):D94–D99. find
  6. Shanchan Wu and Yifan He. 2019. Enriching pre-trained language model with entity information for relation classification. Proceedings of the 28th ACM International Conference on Information and Knowledge Management. find
  7. Yu Gu, Robert Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong Huang, Kevin Gimpel, Wei-Chen Chen, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. 2021. Domain-specific language model pretraining for biomedical natural language processing. ACM Transactions on Computing for Healthcare, 3(1):1…
  8. Ari Z. Klein, Davy Weissenbacher, Karen O’Connor, Amir Elyaderani, Ivan Flores Amaro, Takeshi Onishi, Su Golder, Kaelen Spiegel, Matthew Scotch, and Graciela Gonzalez-Hernandez. 2025. Detection of patient metadata in published articles for genomic epidemiology using machine learning and large langua… find
  9. Guillermo Lopez-Garcia, Jose Miguel Acitores Cortina, Jacob Berkowitz, Joey Chan, Ganesh Chandrasekar, Sumon Kanti Dey, Ivan Flores Amaro, Fernando Gallego, Lauren Gryboski, Ari Z Klein, Martin Krallinger, Salvador Lima-López, Tomohiro Nishiyama, Lisa Raithel, Ahmad Rezaie Mianroodi, Roland Roller,… find
  10. Lluís Màrquez, Xavier Carreras, Kenneth C Litkowski, and Suzanne Stevenson. 2008. Semantic role labeling via framenet, verbnet and propbank. Natural Language Engineering, 14(1):59–109. find
  11. Takeru Miyato, Andrew M Dai, and Ian Goodfellow. 2017. Adversarial training methods for semi-supervised text classification. In International Conference on Learning Representations (ICLR).
  12. A. Mohammed and R. Kora. 2023. A comprehensive review on ensemble deep learning: Opportunities and challenges. Journal of King Saud University -Computer and Information Sciences, 35:757–774. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ursu-2026-metaminers-smm4h-heard-2026-semantic/