SINAI at #SMM4H–HeaRD 2026: Multilingual Clinical NER with MrBERT-biomed and Optuna Hyperparameter Optimization

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Health & Medical Research · Depth: Expert, medium

Summary

The SINAI system, submitted to the MultiClinAI shared task at the 11th SMM4H-HeaRD Workshop (ACL 2026), addresses multilingual clinical Named Entity Recognition (NER) for Disease, Procedure, and Symptom entities in Spanish clinical texts. This approach fine-tunes MrBERT-biomed, a domain-adapted ModernBERT model pre-trained on biomedical corpora, utilizing multilingual clinical data from seven European languages. The system trains independent entity-specific models, each optimized through Bayesian hyperparameter search with Optuna, and incorporates a deterministic post-processing step to align predicted spans to word boundaries. On the official test set, the system achieved strict micro-F1 scores of 0.7453 for Disease, 0.7107 for Procedure, and 0.6603 for Symptom.

Key takeaway

For NLP Engineers developing clinical Named Entity Recognition systems, consider adopting a strategy that combines domain-adapted language models with targeted hyperparameter optimization. Fine-tuning models like MrBERT-biomed on multilingual clinical data, coupled with Bayesian search tools such as Optuna, can significantly improve performance on specific entity types. This approach, demonstrated by the SINAI system's F1 scores, offers a robust pathway to enhance accuracy in complex clinical text analysis.

Key insights

The SINAI system leverages domain-adapted BERT and Bayesian optimization for robust multilingual clinical Named Entity Recognition.

Principles

• Domain-specific BERT models enhance clinical NER.
• Independent models improve entity-specific performance.
• Bayesian optimization refines model hyperparameters.

Method

Fine-tune MrBERT-biomed on multilingual clinical data, train separate entity-specific models optimized via Optuna's Bayesian search, then apply deterministic post-processing for span alignment.

In practice

• Utilize MrBERT-biomed for biomedical text tasks.
• Employ Optuna for efficient hyperparameter tuning.
• Consider entity-specific model training.

Topics

• Named Entity Recognition
• Clinical NLP
• MrBERT-biomed
• Hyperparameter Optimization
• Optuna
• Multilingual Models
• Biomedical NLP

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• KliniskVestBERT: BERT Model Specialised to Norwegian Clinical Texts — Domain-specific pre-training significantly enhances BERT models for Norwegian clinical NLP tasks.
• Predicting relations between SOAP note sections: The value of incorporating a clinical information model — Fine-tuning a RoBERTa-base model on manually annotated clinical data improves NER for SOAP note sections.
• Nested Named Entity Recognition in Plasma Physics Research Articles — A lightweight BERT-CRF model with entity-specific specialization and Bayesian Optimization effectively extracts nested entities from plasma physics texts.
• EDEN: A Large-Scale Corpus of Clinical Notes for Italian — EDEN provides the largest Italian clinical note corpus and a CRF-filling benchmark for LLM-based information extraction.
• Intro to NLP with spaCy (4): Detecting programming languages — Training a custom spaCy Named Entity Recognition (NER) model automates programming language detection by learning from labeled data.
• MentalMARBERT: Domain-Adaptive Pre-training and Two-Stage Fine-Tuning for Arabic Mental Health Disorders Detection — Domain-adaptive pre-training and hierarchical classification significantly improve Arabic mental health disorder detection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.