@inproceedings{hsiao-etal-2026-llatmu,
title = "{LLATMU} at {\#}{SMM}4{H}-{H}ea{RD} 2026: Clinical Text Structuring with {QL}o{RA}-based Generation and Partial-Label {TNM} Classification",
author = "Hsiao, Eric and
Ku, Min-Hsuan and
Shao, Hsuan-Lei",
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://aclanthology.org/2026.smm4h-1.10/",
pages = "56--60",
ISBN = "979-8-89176-432-3",
abstract = "We describe the LLATMU systems submitted to the {\#}SMM4H-HeaRD 2026 shared tasks, covering two related clinical text structuring problems: dialogue-to-SOAP note generation (Task 4) and TNM staging classification from pathology reports (Task 6). Although the two tasks differ in modeling paradigm (text generation versus supervised classification), both require transforming unstructured clinical narratives into structured representations.For Task 4, we instruction-tuned LLMs with parameter-efficient adaptation and submitted a QLoRA-based Ministral-3B system, achieving an official blind test average score of 0.53 and outperforming the task-wide mean and median. For Task 6, we formulate TNM prediction as a three-head classification problem using BioClinical-ModernBERT-large with long-context encoding, class-weighted loss, and normalized partial-label training. The model achieves a validation average macro-F1 of 0.9196 and continues to outperform the official baseline on the more challenging tie-break test set.Across both tasks, our results suggest that robust data handling, stable fine-tuning, and task-appropriate supervision are important for practical clinical NLP under constrained and imperfect shared-task settings."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hsiao-etal-2026-llatmu">
<titleInfo>
<title>LLATMU at #SMM4H-HeaRD 2026: Clinical Text Structuring with QLoRA-based Generation and Partial-Label TNM Classification</title>
</titleInfo>
<name type="personal">
<namePart type="given">Eric</namePart>
<namePart type="family">Hsiao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Min-Hsuan</namePart>
<namePart type="family">Ku</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hsuan-Lei</namePart>
<namePart type="family">Shao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 11th Social Media Mining for Health Research and Applications (SMM4H-HeaRD 2026) Workshop and Shared Tasks</title>
</titleInfo>
<name type="personal">
<namePart type="given">Guillermo</namePart>
<namePart type="family">Lopez-Garcia</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Graciela</namePart>
<namePart type="family">Gonzalez-Hernandez</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-432-3</identifier>
</relatedItem>
<abstract>We describe the LLATMU systems submitted to the #SMM4H-HeaRD 2026 shared tasks, covering two related clinical text structuring problems: dialogue-to-SOAP note generation (Task 4) and TNM staging classification from pathology reports (Task 6). Although the two tasks differ in modeling paradigm (text generation versus supervised classification), both require transforming unstructured clinical narratives into structured representations.For Task 4, we instruction-tuned LLMs with parameter-efficient adaptation and submitted a QLoRA-based Ministral-3B system, achieving an official blind test average score of 0.53 and outperforming the task-wide mean and median. For Task 6, we formulate TNM prediction as a three-head classification problem using BioClinical-ModernBERT-large with long-context encoding, class-weighted loss, and normalized partial-label training. The model achieves a validation average macro-F1 of 0.9196 and continues to outperform the official baseline on the more challenging tie-break test set.Across both tasks, our results suggest that robust data handling, stable fine-tuning, and task-appropriate supervision are important for practical clinical NLP under constrained and imperfect shared-task settings.</abstract>
<identifier type="citekey">hsiao-etal-2026-llatmu</identifier>
<location>
<url>https://aclanthology.org/2026.smm4h-1.10/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>56</start>
<end>60</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T LLATMU at #SMM4H-HeaRD 2026: Clinical Text Structuring with QLoRA-based Generation and Partial-Label TNM Classification
%A Hsiao, Eric
%A Ku, Min-Hsuan
%A Shao, Hsuan-Lei
%Y Lopez-Garcia, Guillermo
%Y Gonzalez-Hernandez, Graciela
%S Proceedings of the 11th Social Media Mining for Health Research and Applications (SMM4H-HeaRD 2026) Workshop and Shared Tasks
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, United States
%@ 979-8-89176-432-3
%F hsiao-etal-2026-llatmu
%X We describe the LLATMU systems submitted to the #SMM4H-HeaRD 2026 shared tasks, covering two related clinical text structuring problems: dialogue-to-SOAP note generation (Task 4) and TNM staging classification from pathology reports (Task 6). Although the two tasks differ in modeling paradigm (text generation versus supervised classification), both require transforming unstructured clinical narratives into structured representations.For Task 4, we instruction-tuned LLMs with parameter-efficient adaptation and submitted a QLoRA-based Ministral-3B system, achieving an official blind test average score of 0.53 and outperforming the task-wide mean and median. For Task 6, we formulate TNM prediction as a three-head classification problem using BioClinical-ModernBERT-large with long-context encoding, class-weighted loss, and normalized partial-label training. The model achieves a validation average macro-F1 of 0.9196 and continues to outperform the official baseline on the more challenging tie-break test set.Across both tasks, our results suggest that robust data handling, stable fine-tuning, and task-appropriate supervision are important for practical clinical NLP under constrained and imperfect shared-task settings.
%U https://aclanthology.org/2026.smm4h-1.10/
%P 56-60
Markdown (Informal)
[LLATMU at #SMM4H-HeaRD 2026: Clinical Text Structuring with QLoRA-based Generation and Partial-Label TNM Classification](https://aclanthology.org/2026.smm4h-1.10/) (Hsiao et al., SMM4H 2026)
ACL