GraphRAG is a technology that integrates knowledge graphs and large-scale language models (LLMs) to improve inference accuracy and contextual relevance. However, it lacks a modular workflow analysis, a systematic solution framework, and insightful empirical research. LEGO-GraphRAG is a modular framework proposed to overcome these limitations. It enables fine-grained decomposition of GraphRAG workflows, systematic classification of existing techniques and implemented GraphRAG instances, and creation of new GraphRAG instances. Through comprehensive empirical studies on large-scale real-world graphs and diverse query sets, it analyzes the tradeoffs between inference quality, execution efficiency, and token or GPU costs, providing insights necessary for building advanced GraphRAG systems.
