A Survey on Graph-Based Retrieval-Augmented Generation: Architectures, Methods, and Applications

Flat-text RAG limitations: Traditional RAG based on flat text corpora and similarity retrieval struggles with intricate relations, multi-hop dependencies, and structured semantics.

Vector-token retrieval weaknesses: Vector-based and token-based retrieval frameworks have weaknesses that motivate graph-aware retrieval architectures.

Graph-RAG open challenges: Graph-RAG still faces open issues in scalability, dynamic graph updates, interpretability, and evaluation.

Flat-text RAG limitations: Traditional RAG based on flat text corpora and similarity retrieval struggles with intricate relations, multi-hop dependencies, and structured semantics.

Vector-token retrieval weaknesses: Vector-based and token-based retrieval frameworks have weaknesses that motivate graph-aware retrieval architectures.

Graph-RAG survey: The paper surveys graph-based Retrieval-Augmented Generation, including architectures, methods, and applications.

Graph-aware RAG architecture: Graph-RAG adds systematic knowledge representations into retrieval and generation by organizing knowledge into graphs.

Graph-RAG method taxonomy: The survey classifies Graph-RAG methods by architectural design, retrieval schemes, learning methods, and reasoning methods.

Hybrid graph reasoning methods: The surveyed methods include graph neural networks, hybrid graph-vector retrieval, and multi-hop inference.

Graph-RAG survey: The paper surveys graph-based Retrieval-Augmented Generation, including architectures, methods, and applications.

Relational reasoning support:

Retrieval weakness mitigation:

Graph-RAG method classification: