Auto Response Generation in Online Medical Chat Services

Area of Science:

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

Background:

• Telehealth adoption has surged, increasing demand for efficient remote medical services.
• The COVID-19 pandemic accelerated the need for virtual consultations.
• Effective communication in telehealth chat sessions is crucial for quality patient care.

Purpose of the Study:

• To develop an intelligent auto-response generation mechanism for doctor-patient chat sessions.
• To enhance the efficiency and quality of medical conversations in telehealth.
• To assist doctors in managing consultation requests during peak times.

Main Methods:

• Utilized a dataset of over 900,000 anonymous doctor-patient messages.
• Applied clustering algorithms to identify frequent doctor responses and manually labeled data.
• Trained machine learning models, including BERT, for response generation.
• Implemented a two-step process: a filtering model and a response generator suggesting top-3 replies.

Main Results:

• The developed auto-response system effectively filters patient messages and suggests relevant doctor responses.
• The BERT model achieved 85.41% precision@3, demonstrating high accuracy and parameter robustness.
• The system aids doctors in responding more efficiently during busy telehealth sessions.

Conclusions:

• The smart auto-response mechanism significantly improves the efficiency of telehealth consultations.
• AI-driven tools like this are vital for scaling virtual healthcare services.
• This technology supports healthcare professionals in delivering timely and effective remote patient care.