Composition-driven symptom phrase recognition for Chinese medical consultation corpora

Area of Science:

• Natural Language Processing
• Medical Informatics
• Computational Linguistics

Background:

• Symptom phrase recognition is crucial for automated question answering systems using medical consultation corpora.
• Existing methods often require extensive manual annotation or comprehensive symptom dictionaries, which are impractical for real-world scenarios due to data scarcity and linguistic diversity.

Purpose of the Study:

• To develop a composition-driven method for recognizing symptom phrases from Chinese medical consultation corpora without manual annotations.
• To address the limitations of data scarcity and linguistic variation in symptom phrase extraction.

Main Methods:

• Propose a method that learns models capturing the composition of symptom components (semantic units of words).
• Introduce an automatic annotation strategy for standard symptom phrases collected from multiple sources.
• Utilize position information and interaction scores between symptom components to characterize symptom phrases.

Main Results:

• Achieved strong positive results in symptom phrase recognition tasks without any manual annotations.
• Demonstrated the method's robustness in extracting previously unseen symptom phrases.
• Showcased significant potential with access to large corpora.

Conclusions:

• Compositionality provides a viable solution for information extraction from unstructured text with limited labeled data.
• The proposed method effectively handles the scarcity of annotated resources in medical corpora.