Integrasi Graph-Based Retrieval Augmented Generation (GraphRAG) pada Large Language Model (LLM) dalam Pengembangan Chatbot Biomedis Multimodal

Abstract

The availability of massive and multimodal biomedical data such as electronic health records, laboratory results, and prescriptions poses significant challenges in fast and accurate information extraction. This study aims to develop and evaluate a biomedical chatbot to address these challenges. The proposed method integrates Graph-Based Retrieval Augmented Generation (GraphRAG) with a Large Language Model (LLM) to enable contextual understanding and presentation of information from complex databases. The system is built using the MIMIC-IV dataset to structure biomedical data into a knowledge graph using Neo4j. The chatbot architecture is then developed using the LangChain framework to orchestrate workflows between the LLM and the graph database. GraphRAG techniques are implemented to perform superior information retrieval on graphically structured data, ensuring the relevance and accuracy of the context provided to the LLM. The system was evaluated using the RAGAs framework to measure performance across various aspects. Individual metrics in RAGAS recorded scores of 80,5% for answer_correctness, 97% for faithfulness, 76,9% for answer_relevancy, 85,7% for context_precision, and 81,7% for context_recall. These results demonstrate that the GraphRAG approach can enhance the LLM's ability to provide answers that are not only relevant and factual but also consistent with complex source data. The developed chatbot shows great potential as a reliable tool for both medical professionals and patients in navigating complex health information.