A semi-supervised approach for extracting TCM clinical terms based on feature words

Area of Science:

• Natural Language Processing (NLP)
• Medical Informatics
• Traditional Chinese Medicine (TCM)

Background:

• Extracting clinical terms from Traditional Chinese Medicine (TCM) is crucial for medical informatics.
• Manual annotation for Natural Language Processing (NLP) tasks is costly and time-consuming.

Purpose of the Study:

• To develop a semi-supervised model for efficient extraction of TCM clinical terms.
• To reduce the reliance on extensive manual annotation in TCM NLP.

Main Methods:

• The proposed model utilizes a BiLSTM-CRF architecture.
• It integrates semi-supervised learning with a feature word set to enhance extraction.
• This approach leverages existing extraction results to improve performance.

Main Results:

• The model demonstrated improved extraction for five key TCM clinical term types: traditional Chinese medicine, symptoms, patterns, diseases, and formulas.
• The best F1-score achieved on the test dataset was 78.70%.

Conclusions:

• The developed semi-supervised model significantly reduces manual labeling costs for TCM clinical term extraction.
• This method enhances the accuracy and efficiency of Named Entity Recognition (NER) in TCM research.