Novel Approach to Personalized Physician Recommendations Using Semantic Features and Response Metrics: Model...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Natural Language Processing

Background:

• The increasing use of online medical services necessitates improved patient triage systems.
• Traditional methods struggle to accurately match patient queries with physician expertise.
• Developing advanced physician recommendation algorithms is crucial for efficient healthcare access.

Purpose of the Study:

• To develop and validate a patient-physician hybrid recommendation (PPHR) model.
• To enhance triage performance using semantic features and physician response metrics.

Main Methods:

• Collected 646,383 web-based medical consultation records.
• Developed semantic features for patient questions and physician specialties.
• Incorporated physician response rates to refine candidate rankings.
• Evaluated model performance using metrics and physician feedback, comparing Sentence-BERT and Doc2Vec.

Main Results:

• The PPHR model achieved optimal performance with 14 recommended physicians (F1-score: 76.25%).
• Excluding physician characteristics and response rates significantly reduced performance metrics.
• Sentence-BERT demonstrated a higher average hit ratio (88.6%) compared to Doc2Vec (53.4%).

Conclusions:

• The PPHR model effectively utilizes semantic analysis and response metrics.
• Enables patients to identify suitable physicians more accurately through enhanced triage.