Healthcare Chatbot For Medical Assistance Based On Natural Language Processing

Abstract

Chatbots, sometimes referred to as conversational agents or digital assistants, are artificial intelligence (AI) software programs that take the place of living chat agents by conducting online chat conversations via text or text-to-speech. Chatbots in healthcare can establish a two-way information exchange with patients, lowering healthcare expenses and enhancing public access to medical information while providing effective support for patients and healthcare professionals in therapeutic settings outside of on-site consultations. Researches to build a Vietnamese Healthcare Answering system is more challenging than it is to develop an English system, particularly for a system that uses deep learning techniques since the lack of available Vietnamese resources and technologies. From that, I proposed MedBot, a Vietnamese multifunctional role of chatbots assisting in the medical domain. MedBot is designed as a support tool to conduct communication, provide a potential diagnosis and inform patients of necessary disease materials for their needs. This is accomplished with the application of Natural Language Processing technique and deep learning approaches along with transfer learning for complex text understanding and inference. Additionally, Flask API, Sqlite3 database and basic front-end framework are implemented to create a website interface for user experience. In the intent recognition task for which the project classifier knows the disease and the patient's situation, the PhoBERT model outperforms the neural network model, achieving 0.89 F1 score compare to 0.85. The above result is the result of an experiment on a self-constructed data set obtained from Vietnamese medical websites; it may not be accurate with reality or experts, but it is possible to compare the results of the two models and common experience.