Team Aurevia at CLPsych 2026: Local Healthcare NLP for Schema-Constrained Self-State Modeling

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Clinical Natural Language Processing · Depth: Expert, quick

Summary

Team Aurevia introduced a local open-weight healthcare NLP system for the CLPsych 2026 Shared Task, designed to predict MIND-coded self-state elements, moments of change, summaries, and dynamic signatures from social media timelines. This task is challenging due to the need for diverse longitudinal evidence across privacy-sensitive mental health language, encompassing coarse presence, fine-grained ABCD subelements, and timeline-level changes. The system integrates TF-IDF retrieval, schema-constrained local Qwen2.5 prompting, ordinal calibration, and conservative post-processing. Among official runs, Aurevia achieved notable results, ranking 3rd of 17 for Task 1.2 presence prediction, 5th of 13 overall for Task 3.1, 1st on Task 3.1 consistency, and 2nd of 9 for MIND-coded deterioration signatures. This performance demonstrates the competitiveness of constrained local LLM pipelines in sensitive healthcare NLP, while also reducing dependence on hosted proprietary inference solutions.

Key takeaway

For NLP Engineers developing solutions for sensitive healthcare data, you should consider implementing local, schema-constrained LLM pipelines. This approach, exemplified by Team Aurevia's Qwen2.5 system, allows you to achieve competitive performance in tasks like self-state modeling while significantly reducing reliance on potentially costly or privacy-compromising hosted proprietary inference. Evaluate integrating TF-IDF retrieval and ordinal calibration to enhance your system's accuracy and consistency.

Key insights

Constrained local LLM pipelines can achieve competitive performance in sensitive healthcare NLP tasks.

Principles

• Local LLMs reduce reliance on hosted proprietary inference.
• Schema constraints enhance NLP accuracy for complex data.
• Longitudinal evidence is crucial for self-state modeling.

Method

The system combines TF-IDF retrieval, schema-constrained local Qwen2.5 prompting, ordinal calibration, and conservative post-processing for self-state prediction.

In practice

• Implement local Qwen2.5 with schema constraints for sensitive data.
• Integrate TF-IDF retrieval for initial evidence gathering.
• Apply ordinal calibration to refine model predictions.

Topics

• Healthcare NLP
• Local LLMs
• Schema Constraints
• Qwen2.5
• Self-State Modeling
• CLPsych Shared Task

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Measuring the sensitivity of LLM-based structured extraction to prompt, model, and schema choices in clinical discharge summaries — LLM-based clinical data extraction sensitivity varies significantly with model and schema, especially regarding "absence-versus-silence" distinctions.
• Executable Schema Contracts: From Automatic Ingestion to Multi-Source Retrieval — An automatically induced, executable schema can unify knowledge graph ingestion and multi-source query-time retrieval.
• Team MKC at CLPsych 2026: Capturing and Characterizing Mental Health Changes through Social Media Timeline Dynamics — An LLM-based pipeline analyzes social media timelines for early mental health change detection and continuous monitoring, integrating post and user-level assessments.
• Cross-Lingual Empirical Evaluation of Large Language Models for Arabic Medical Tasks — English-centric LLMs exhibit significant performance gaps and reliability issues in Arabic medical tasks.
• sebis at CRF Filling 2026: A Two-Stage Local LLM Pipeline for Medical CRF Filling — A two-stage local LLM pipeline using MedGemma-27B achieves competitive, privacy-preserving medical CRF filling with strict evidence adherence.
• T2D-Bench: Evidence-Gated Evaluation of LLM Outputs for Type 2 Diabetes Using a Multi-Layer Clinical-Lifestyle Knowledge Graph — LLMs require evidence-gated evaluation against clinical knowledge graphs to ensure medical accuracy and guideline compliance.

Open in AIssential →

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