A comprehensive survey of knowledge Graph-Augmented Generation (Graph-RAG) for trustworthy large language models

LLM hallucination: Large language models suffer from hallucination because their reliance on parametric memory can produce false yet plausible statements not grounded in reality.

RAG lacks verifiable reasoning: Retrieval-augmented generation still retrieves semi-unstructured passages and lacks rigorous, verifiable, logically coherent reasoning processes.

Graph-RAG scaling and evaluation challenges: Graph-RAG faces challenges in scaling and in evaluation protocols.

LLM hallucination: Large language models suffer from hallucination because their reliance on parametric memory can produce false yet plausible statements not grounded in reality.

RAG lacks verifiable reasoning: Retrieval-augmented generation still retrieves semi-unstructured passages and lacks rigorous, verifiable, logically coherent reasoning processes.

Graph-RAG survey: The paper reviews recent progress in Knowledge Graph Enhanced Generation as an approach for structured retrieval and rigorous reasoning in LLM generation.

Graph-RAG taxonomy framework: The paper designs a computation framework that classifies Graph-RAG methods into Graph Indexing, Graph Guided Retrieval, and Graph Enhanced Generation.

Causal and actionable graph reasoning agenda: The paper discusses future directions for Graph-RAG focused on causal graph reasoning and actionable graphs beyond simple retrieval.

Graph-RAG survey: The paper reviews recent progress in Knowledge Graph Enhanced Generation as an approach for structured retrieval and rigorous reasoning in LLM generation.

Graph-RAG taxonomy framework: The paper designs a computation framework that classifies Graph-RAG methods into Graph Indexing, Graph Guided Retrieval, and Graph Enhanced Generation.

Comprehensive Graph-RAG taxonomy:

Graph-RAG challenges summarized:

Graph-RAG supports trustworthy AI: