Augmenting biomedical named entity recognition with general-domain resources

2024-06-15

Journal of Biomedical Informatics

AI Summary

Biomedical named entity recognition (NER) suffers from scarce, costly, and noisy annotations, leading to poor model generalization. To address this, we propose a few-shot NER method leveraging domain-agnostic pretraining corpora—specifically Wikipedia and BooksCorpus—to enhance weakly supervised learning. Our approach integrates multi-stage transfer learning, curriculum-driven domain-adaptive knowledge distillation, entity-type-aware prompt tuning, and consistency regularization. This enables effective knowledge transfer and robust pseudo-labeling without requiring large-scale high-quality annotations. Evaluated on standard benchmarks—including BC5CDR and JNLPBA—our method achieves absolute F1 improvements of 3.2–5.8% over strong baselines. Notably, using only 10% of the labeled data, it surpasses fully supervised models trained on the complete annotated corpus. The framework significantly reduces dependency on expert-curated labels while delivering scalability and robustness, establishing a new paradigm for low-resource biomedical NER.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Computer Vision: Medical and Biological Imaging

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Problem

• Biomedical Named Entity Recognition
• Data Annotation
• Model Accuracy

Innovation

• GERBERA method
• pre-training on general language data
• fine-tuning on biomedical data