Retrieval-Augmented Generation Based Large Language Model Chatbot for Improving Diagnosis for Physical and Mental Health

Abstract

Chatbot integration in the medical domain results in improved accessibility to healthcare information and services with enhanced patient query communication and patient education. The proposed work is a novel approach to Health Care chatbot named MEDGPT. The model is developed using the Retrieval-Augmented Generation (RAG) framework integrated with external data sources such as PDF documents, CSV files, and PubMed documents related to Health Care. It combines a retriever component of the architecture to fetch relevant information from external sources and a generator component to craft contextually appropriate responses using Large Language Model (LLM's). The architecture employed tools and agents in order to generate response from multiple external sources. Tools are specialized for each data source and are designed to extract relevant information based on user queries. Agents are components within the chatbot architecture that handle different aspects of logical reasoning and decision making. When a user query, requires information from an external data source, the corresponding agent invokes the appropriate retrieval tool. Once the data is processed, it is integrated into the chatbot's knowledge base using the RAG framework. The RAG framework combines the retrieved data with the chatbot's language generation capabilities to craft contextually appropriate responses to user queries. Through extensive testing and evaluation, the chatbot achieved significant improvements in user satisfaction, response accuracy, and engagement, showcasing the potential of the RAG framework in leveraging external data sources for intelligent conversational agents in Health Care. The proposed framework enhanced the RAG based LLM chatbot's capabilities to provide relevant and accurate responses to user queries related to Health Care with respected Physical and Mental health.

Topics & Concepts