Extracting Symptom Names and Disease-Symptom Relationships from Web Texts Using a Multi-Column Convolutional Neural...

Area of Science:

• Natural Language Processing
• Medical Informatics
• Web Archiving

Background:

• Developing accurate medical dictionaries is crucial for diagnostic support systems.
• Existing medical dictionaries may lack comprehensive symptom information or natural language expressions used by patients.
• Large-scale web archives offer a rich source of patient-generated health-related text.

Purpose of the Study:

• To propose a method for creating large-scale Japanese medical dictionaries from web archives.
• To develop a diagnosis support system utilizing natural language inputs from patients.
• To construct dictionaries for symptom names and disease-symptom relationships.

Main Methods:

• Utilized a large web archive containing text from non-medical experts.
• Developed two methods for dictionary construction: symptom name extraction and disease-symptom relationship extraction.
• Employed sentence retrieval using WISDOM X and neural classifiers for information extraction.

Main Results:

• Achieved a 93.8% F1-score for symptom name extraction.
• Achieved an 88.3% F1-score for disease-symptom relationship extraction.
• Demonstrated the effectiveness of the proposed methods for building medical dictionaries.

Conclusions:

• The proposed method successfully creates large-scale Japanese medical dictionaries from web archives.
• The developed dictionaries and extraction methods can support natural language-based diagnosis systems.
• High F1-scores indicate the potential for accurate disease prediction using patient-provided information.