GRAG: Graph Retrieval-Augmented Generation

The authors tackle two issues in generation using RAGS:

1. For Retrieval: How to efficiently retrieve relevant textual subgraph
2. For Generation: How to deliver textual subgraph joint textual and topological information into LLMS. To do generate the information they use subgraph retrieval, as a heirarchichal description that goes into an embedder that fuses graph tokens and text tokens.