Elamur Research⌕
☰
home / papers / 2026 / A3S@C-DAC at #SMM4H-HeaRD 2026: Reasoning Meets Evidence: LLMs for Interpretable Insomnia Detection with Evidence Extraction in Clinical Notes

A3S@C-DAC at #SMM4H-HeaRD 2026: Reasoning Meets Evidence: LLMs for Interpretable Insomnia Detection with Evidence Extraction in Clinical Notes

Paper · 2026 · Abhishek Maity, Amol Shinde, Abhishek Suresh Kushare et al.

Summary

This shared-task paper from C-DAC Mumbai addresses insomnia detection in clinical notes, where a system must both classify notes and justify its decisions with text evidence. The data come from MIMIC-III, released for SMM4H-HeaRD 2026 Task 1. Every note is annotated with a binary insomnia label and evidence spans given as character offsets, along with separate labels for four rules: Rule B, Rule C, and Definitions 1 and 2. The authors compare two strategies. The first is parameter-efficient fine-tuning of small models: Gemma-3 1B with QLoRA and Qwen3-1.7B with LoRA. The second is few-shot prompting of Qwen3-8B to predict rule labels and extract spans jointly, in two prompt versions. The QLoRA-tuned Gemma model performed best on binary classification. The first prompting variant achieved the top rule-level micro-F1. The refined second prompt improved span extraction but slightly lowered classification scores. Exact character-level localization remained weak. This matters because it shows that small fine-tuned models can match or exceed larger ones on clinical classification. It also shows that reliably grounding predictions in exact evidence is still an unsolved problem.

Key points

  • Subtask 1 (binary classification): Gemma-3 1B (QLoRA) reached an F1 of 0.7333, with precision 1.00 and recall around 0.58. The model is conservative: it produced no false positives but missed some true cases. Qwen3-1.7B (LoRA) was more balanced but scored lower (F1 ≈ 0.57).
  • Subtask 2 (rule labels): few-shot Qwen3-8B v1 achieved the best micro-F1 of 0.6535, helped by high recall (≈0.75). The v2 prompt dropped to about 0.60.
  • Evidence spans: v2 reached a partial-match F1 of 0.5192. Exact-match F1 stayed considerably lower, starting at only 0.14 for v1. Post-processing parsed model outputs into JSON and checked offsets against the note text.
  • Error categories: the authors identify five main sources of error:
    • implicit symptoms (e.g., fatigue, restlessness);
    • ambiguous medications such as benzodiazepines, which are prescribed for many conditions;
    • inaccurate span boundaries;
    • negation;
    • evidence scattered across long notes.
  • Limitations and future work: the systems use no structured medical knowledge, outputs are sensitive to prompt wording, and evaluation covers a single dataset. Training ran on NVIDIA L4 GPUs.

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

Also on the byline: Centre for Development of Advanced Computing.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksPsychology · 1,649 worksExperimental and Cognitive Psychology · 930 worksSocial Psychology · 330 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,685 works

Cite

Cite as Abhishek Maity, Amol Shinde, Abhishek Suresh Kushare and Swapnil Pawar. 2026. A3S@C-DAC at #SMM4H-HeaRD 2026: Reasoning Meets Evidence: LLMs for Interpretable Insomnia Detection with Evidence Extraction in Clinical Notes. 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-1.1

@inproceedings{maity2026a3s,
  title = {{A3S@C-DAC at \#SMM4H-HeaRD 2026: Reasoning Meets Evidence: LLMs for Interpretable Insomnia Detection with Evidence Extraction in Clinical Notes}},
  author = {Abhishek Maity, Amol Shinde, Abhishek Suresh Kushare and Swapnil Pawar},
  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-1.1},
  url = {https://doi.org/10.18653/v1/2026.smm4h-1.1}
}

References

13 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Michael J Sateia. 2014. International Classification of Sleep Disorders-Third Edition. Chest, 146(5):1387– 1394. find
  2. Emily Alsentzer, John Murphy, William Boag, Wei-Hung Weng, Di Jindi, Tristan Naumann, and Matthew McDermott. 2019. Publicly Available Clinical BERT Embeddings. In Proceedings of the 2nd Clinical Natural Language Processing Workshop (ClinicalNLP), pages 72–78.
  3. Karan Singhal, Shekoofeh Azizi, Tao Tu, S Sara Mahdavi, Jason Wei, Hyung Won Chung, Nathan Scales, Ajay Tanwani, Heather Cole-Lewis, Stephen Pfohl, and 1 others. 2023. Large language models encode clinical knowledge. Nature, 620:172–180.
  4. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, and… find
  5. Yanshan Wang, Liwei Wang, Majid Rastegar-Mojarad, Sungrim Moon, Feichen Shen, Naveed Afzal, Sijia Liu, Yuqun Zeng, Saeed Mehrabi, Sunghwan Sohn, and 1 others. 2018. Clinical information extraction applications: A literature review. Journal of Biomedical Informatics, 77:34–49. find
  6. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, and 41 others. 2025. Qwen3 Technical Report.
  7. Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. 2023. QLoRA: Efficient fine-tuning of quantized LLMs. Advances in Neural Information Processing Systems (NeurIPS), 37:10088– 10115. find
  8. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of the conference of the North American chapter of the Association for Computational Linguistics (NAACL): Human Language Technologi…
  9. Jay DeYoung, Sarthak Jain, Nazneen Fatema Rajani, Eric Lehman, Caiming Xiong, Richard Socher, and Byron C Wallace. 2020. ERASER: A benchmark to evaluate rationalized NLP models. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL), pages 4443–4458.
  10. Carol Friedman, Philip O Alderson, John HM Austin, James J Cimino, and Stephen B Johnson. 1994. A General Natural-language Text Processor for Clinical Radiology. Journal of the American Medical Informatics Association, 1(2):161–174. find
  11. Gemma Team Gemma, Aishwarya Kamath, Johan Ferret, Shreya Pathak, Nino Vieillard, Ramona Merhej, Sarah Perrin, Tatiana Matejovicova, Alexandre Ramé, Morgane Rivière, Louis Rouillard, Thomas Mesnard, Geoffrey Cideron, Jean bastien Grill, Sabela Ramos, Edouard Yvinec, Michelle Casbon, Etienne Pot, Ivo… find
  12. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, and Weizhu Chen. 2022. LoRA: Low-Rank Adaptation of Large Language Models. In International Conference on Learning Representations (ICLR).
  13. Alistair Johnson, Tom Pollard, and Roger Mark. 2016. MIMIC-III Clinical Database. PhysioNet. Version 1.4. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/maity-2026-a3s-c-dac-smm4h-heard/