LLM-Derived Knowledge Graphs

GraphRAG is a novel approach that combines large language models (LLMs) with knowledge graphs to enhance retrieval operations and enable new analytical scenarios. It consists of two main steps: indexing private data to create LLM-derived knowledge graphs, which serve as memory representations, and an orchestration mechanism that utilizes these indices for improved retrieval. GraphRAG’s key differentiators include enhanced search relevancy due to a holistic view of dataset semantics and the ability to handle complex queries that require understanding of dataset-wide trends and relationships.