LLMsClinical NLPClassificationEvidence ExtractionShared Task

Insomnia classification and evidence extraction

Built baseline shared-task systems for identifying insomnia cases and extracting supporting evidence spans from clinical notes.

2025-11-011 min read

Overview

Build practical baseline systems for identifying insomnia cases and highlighting the evidence that supports the decision.

System design

  • Collaborated with the shared-task organizers to define baseline modeling and evaluation workflows
  • Evaluated Llama 3 8B and 70B, Qwen2.5-32B-GPTQ, and Qwen3-8B on clinical-note classification
  • Measured both patient-level decisions and supporting evidence span quality
  • Framed the work so predictions stayed explainable instead of stopping at a single label

Results

  • Established shared-task baselines for insomnia classification and evidence extraction
  • Produced comparative results across multiple open-source LLM families
  • Created a foundation for stronger evidence-aware clinical NLP systems