RAGbased-Medical-Chatbot

This project presents a comprehensive system for a Retrieval-Augmented Generation (RAG) based medical chatbot specifically designed for the Chinese language. It integrates advanced techniques to enhance the accuracy and efficiency of medical information retrieval and generation.

Introduction

This system innovatively combines knowledge graph with the Chroma vector database to reduce the computational complexity of searching relevant documents. We leverage a knowledge graph to structure medical information and utilize Large Language Model-driven Cypher query generation to automatically identify the correct subject area for a given query. This approach allows us to then efficiently search for documents within a specific, pre-defined subject database, rather than performing a computationally intensive search across the entire corpus. This focused retrieval strategy greatly improves the speed and precision of information delivery.

System Architecture

The system is composed of two distinct parts:

• Frontend: Built with Electron and Vue3, providing a responsive and intuitive user interface for interaction. I used pixiv-live2d for digital human. The system allows you to input by texting and speaking.
• Backend: Handles the core logic, including knowledge graph interactions, LLM integrations, voice data processing, and vector database management. I built Socket and Http services.

Pre-requisites and Setup

To run this system, please follow these preliminary steps:

1. Dataset Construction: Transform the "面向家庭常见疾病知识图谱" (Knowledge Graph for Common Family Diseases) dataset from Open-KG into a sequence labeling dataset.
2. Cloud Storage Setup: Configure a cloud OSS (Object Storage Service) server for temporary storage of voice message data. Ensure your access key and other necessary credentials are updated in Backend/Models/CustomLLM/VoiceDataTransfer.py.
3. QWen API Key: Obtain an API key for Aliyun's QWen model and insert it into Backend/Models/CustomLLM/QWen.py for model invocation.
4. Database Configuration: Set up Redis, MongoDB, and Chroma databases using docker-compose. The necessary .yml file is located in Backend/Docker.
5. Frontend Initialization:
   • Navigate to the Frontend/medchatbot-system directory.
   • Run npm install to install all required packages.
   • Execute npm run start to launch the Electron and Vue3 project.

Forbidden

This repository is intended solely for academic study and research purposes. Commercial use of this project is strictly prohibited.

##Acknowledgements I extend my sincere gratitude to the developer of RAGQnASystem for the valuable contributions and inspiration.