FAIR GraphRAG Improves LLM Accuracy for Semantic Data Analysis

This paper presents FAIR GraphRAG, an innovative framework designed to enhance Retrieval-Augmented Generation (RAG) systems by incorporating FAIR (Findability, Accessibility, Interoperability, and Reusability) principles. Traditional RAG approaches often fall short in complex, domain-specific contexts like medicine due to a lack of structured FAIRification in their underlying knowledge resources. FAIR GraphRAG addresses this by using FAIR Digital Objects (FDOs) as the foundational units within a graph-based retrieval system. Each node in the knowledge graph represents an FDO, encapsulating data, metadata, persistent identifiers, and semantic links. The framework leverages LLMs for schema construction and automated data extraction. Co-designed with medical professionals, its application to a gastroenterology biomedical dataset demonstrated significant improvements in question-answering accuracy, coverage, and explainability, particularly for complex queries involving metadata and ontology links. This work highlights the potential of combining FAIR data practices with advanced graph-based retrieval for specialized fields.