An improved data augmentation approach and its application in medical named entity recognition

Area of Science:

• Natural Language Processing
• Bioinformatics
• Machine Learning

Background:

• Medical named entity recognition (NER) faces challenges like high data acquisition costs, specialized terminology, and limited training resources.
• These challenges hinder achieving high performance in medical NER systems.
• Data augmentation is essential to generate synthetic data, balance distributions, enrich datasets, and improve model generalization.

Purpose of the Study:

• To propose novel data augmentation techniques for medical NER.
• To address data scarcity and imbalance in the medical domain.
• To enhance the performance and accuracy of deep learning-based Chinese NER models.

Main Methods:

• Contextual Random Replacement based on Word2Vec Augmentation (CRR).
• Targeted Entity Random Replacement Augmentation (TER).
• Integration with a deep learning-based Chinese NER model (BERT-BiLSTM-CRF).

Main Results:

• Both CRR and TER methods significantly improve medical NER performance.
• The BERT-BiLSTM-CRF model achieved a notable F1 score of 83.587% with augmentation.
• This represents a 1.49% increase in performance over the baseline model, validating the augmentation strategies.

Conclusions:

• Data augmentation is crucial and effective for improving medical NER, particularly under resource constraints.
• The proposed CRR and TER methods offer practical solutions for enhancing medical NER systems.
• These techniques contribute to more accurate and robust recognition of medical entities.