NU_DeepHealthNLP at #SMM4H-HeaRD 2026: Entity-Conditioned Generation and a Four-Stage Pipeline for Automated SOAP Note Generation

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Science & Research — Health & Medical Research, Mathematics & Computational Sciences · Depth: Expert, medium

Summary

NU_DeepHealthNLP submitted two systems to Task 4 of the SMM4H-HeaRD 2026 Shared Task for automated SOAP note generation from doctor–patient dialogues. The first system is an entity-conditioned generation model, Mistral-7B-Instruct-v0.1, fine-tuned with QLoRA on 8,529 MedSynth training dialogues. This model's prompts for both training and inference include clinical entities extracted and grouped by SOAP section. The second submission is a four-stage modular pipeline, which extends the first by adding a hybrid retrieval stage and a rule-based verification stage. A key finding is that integrating structured clinical domain knowledge, specifically NER entities grouped by SOAP section, into the generation prompt consistently improves performance over dialogue-only generation. The four-stage pipeline achieved an average score of 0.54 on the official test set, securing the first rank on the shared task leaderboard.

Key takeaway

For NLP Engineers developing medical text generation systems, this work demonstrates that explicitly structuring clinical domain knowledge within prompts significantly boosts performance. You should consider integrating Named Entity Recognition (NER) outputs, grouped by relevant sections like SOAP, directly into your LLM inputs. This approach, exemplified by the top-ranked four-stage pipeline, can yield more consistent and reliable automated clinical documentation, potentially reducing post-generation editing.

Key insights

Incorporating structured clinical entities into generation prompts significantly enhances automated SOAP note accuracy.

Principles

• Structured domain knowledge improves text generation.
• Modular pipelines can enhance complex NLP tasks.

Method

A four-stage pipeline for SOAP note generation involves entity-conditioned generation using a fine-tuned Mistral-7B-Instruct-v0.1, augmented with hybrid retrieval and rule-based verification stages.

In practice

• Fine-tune Mistral-7B-Instruct-v0.1 with QLoRA for medical text.
• Embed NER entities grouped by section into LLM prompts.

Topics

• SOAP Note Generation
• Clinical NLP
• Mistral-7B-Instruct-v0.1
• QLoRA Fine-tuning
• Named Entity Recognition
• Medical Dialogue Systems

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• 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.
• How to Leverage NER and Advanced NLP Techniques for Life Sciences — NER and advanced NLP are essential for structuring and interpreting the vast unstructured text data in Life Sciences.
• Analysing Lightweight Large Language Models for Biomedical Named Entity Recognition on Diverse Ouput Formats — Lightweight LLMs offer competitive biomedical NER performance, with specific output formats proving more effective than multi-format tuning.
• KliniskVestBERT: BERT Model Specialised to Norwegian Clinical Texts — Domain-specific pre-training significantly enhances BERT models for Norwegian clinical NLP tasks.
• Concept Extraction and Webb’s Depth of Knowledge: Comparing LLM Question Generation Pipelines for Educational Assessment — Comparing LLM pipelines for educational question generation using concept extraction and Webb's Depth of Knowledge.
• Beyond Literal Summarization: Redefining Hallucination for Medical SOAP Note Evaluation — Lexical faithfulness in LLM evaluation misclassifies valid clinical reasoning as hallucination, inflating error rates.

Open in AIssential →

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