Projects
All projects
NLPSequence LabelingTransformersRLBiomedical NLP

Biomedical negation detection with Transformers, CRF, and PPO

Fine-tuned BERT-family models for negation cue and scope tagging, then compared CRF, hierarchical, and PPO-based sequence-labeling approaches.

2024-07-01 · 1 min read
Code

Overview

Detect negation cues and their scope in biomedical text with models that can handle imbalanced labels and structured output behavior.

System design

  • Fine-tuned BERT, BioBERT, and ClinicalBERT for cue and scope tagging
  • Compared CRF and hierarchical variants under severe class imbalance
  • Trained PPO-based sequence-labeling policies using rewards derived from human-annotated spans
  • Compared aggregate and label-level results to understand minority-class behavior

Results

  • Established strong supervised baselines for negation cue and scope detection
  • Compared PPO-based sequence labeling with supervised approaches
  • Produced detailed evidence on model behavior under severe label imbalance