Unifying Large Language Models and Knowledge Graphs: A Roadmap

This influential survey categorizes approaches into three paradigms: KG-Enhanced LLMs (improving LLM training and inference), LLM-Augmented KGs (using LLMs to construct and reason over KGs), and Synergized LLM+KG (bidirectional enhancement). Cited over 800 times, it provides the theoretical framework for understanding GraphRAG in the broader LLM+KG landscape.