@inproceedings{charan-etal-2026-thunderbolts,
    title = "Thunderbolts at {\#}{SMM}4{H}-{H}ea{RD} 2026: Detection of Insomnia in Clinical Notes using Transformers",
    author = "Charan, Guddanti Venkata Sree  and
      Nama{\_}Ss@Cs.Iitr.Ac.In, Nama{\_}Ss@Cs.Iitr.Ac.In  and
      Sharma, Raksha  and
      Murthy, Rudra",
    editor = "Lopez-Garcia, Guillermo  and
      Gonzalez-Hernandez, Graciela",
    booktitle = "Proceedings of the 11th Social Media Mining for Health Research and Applications ({SMM}4{H}-{H}ea{RD} 2026) Workshop and Shared Tasks",
    month = jul,
    year = "2026",
    address = "San Diego, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.smm4h-1.43/",
    pages = "264--267",
    ISBN = "979-8-89176-432-3",
    abstract = "We present the SuSh system for Subtask 1 of the MultiClinAI shared task at the 11th SMM4H and HeaRD Workshop (ACL 2026), which addresses multilingual clinical named entity recognition (NER) across seven languages. Our system adopts a fully zero-shot approach using GLiNER-biomed-large-v1.0, a span-based NER model pre-trained on biomedical text, requiring no task-specific fine-tuning or labeled data in target languages. We apply a character-level sliding window strategy to handle long clinical documents that exceed the model{'}s token limit and incorporate a post processing pipeline including threshold optimization via F1-max sweep, entity-specific gazetteer lookup derived from DisTEMIST and SympTEMIST terminology lists, span boundary correction, and negation filtering. Our official submission achieves a Strict F1 of 0.5175, Strict Precision of 0.5536, Strict Recall of 0.4859, and CHR F1 of 0.6130 on the English disease subtask, demonstrating that domain adapted zero-shot biomedical NER models can serve as competitive baselines for multilingual026 clinical entity recognition without any task specific training data."
}