Language inference-based learning for Low-Resource Chinese clinical named entity recognition using language model.

Area of Science:

• Natural Language Processing
• Medical Informatics
• Machine Learning

Background:

• Electronic Health Records (EHRs) are increasingly replacing paper records, necessitating efficient information extraction.
• Clinical Named Entity Recognition (CNER) is crucial for extracting valuable data from EHRs.
• Traditional supervised methods for CNER require extensive annotated data, which is scarce in clinical settings.

Purpose of the Study:

• To address the challenge of limited annotated data in clinical Named Entity Recognition (NER).
• To propose a novel language inference-based learning method (LANGIL) for low-resource clinical NER.
• To leverage the comprehension capabilities of Pre-trained Language Models (PLMs) with minimal training samples.

Main Methods:

• Developed a language inference-based learning method (LANGIL) for clinical NER.
• Reformulated the entity recognition task into a language inference problem using prompt learning.
• Avoided training additional network layers from scratch, mitigating the pre-training/downstream task gap.

Main Results:

• LANGIL demonstrated significant improvements in F1-score on four Chinese clinical NER datasets.
• The proposed method effectively utilizes PLMs' comprehension abilities under low-resource conditions.
• Achieved better performance compared to previous methods in scarce data scenarios.

Conclusions:

• LANGIL offers a promising solution for clinical NER in low-resource environments.
• The language inference-based approach enhances information extraction from EHRs with limited annotations.
• This method alleviates the data annotation bottleneck in clinical natural language processing research.