Design and Evaluation of a Medical Chatbot Built on BERT Language Model for Remote Health Assistance

Telemedicine refers to delivering healthcare services, including diagnosis, treatment, and consultations, remotely through digital communication tools such as video calls, phone calls, and online messaging. Access to quality healthcare remains a significant challenge in underserved rural areas worldwide. Due to transportation issues and financial support in remote areas, people can only visit hospitals close to their location. Current telemedicine applications provide video conversations with doctors but need more seamless services due to the huge population increase in cities and the unavailability of internet services for people in rural areas. This article presents a medical chatbot application trained on a customized health assistant corpus to generate fast responses in text format to queries related to common health issues. The solution is built on the BERT model to train the chatbot, which includes MLM and NSP, which is one of the best ways to obtain accurate results. The BERT model trained on customized health assessment corpus consists of multiple health assessment corpora and health assistant queries/responses collected from different sources and achieved 96.5% test accuracy. Since the model is pre-trained, it will generate the response as fast as possible. In addition to that, the application provides 24/7 availability, user-friendly interfaces, remote monitoring, and robust data security.