Scholar Feed

Retrieval-augmented generation

GraphRAG

A Survey of Graph Retrieval-Augmented Generation for Customized Large Language Models

Heavily superseded#2 of 1,179 most-superseded · first seen Jan 21, 2025

Heavily superseded — a standard baseline that newer methods routinely beat

38 papers critique it · 31 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites GraphRAG as a baseline.

GraphRAG follows with an average score of 0.50, limited by its dependence on entities and relationships extracted from the knowledge graph, which may be incomplete or sparse.
— Benchmarking Vector, Graph and Hybrid Retrieval Augmented Generation (RAG) Pipelines for Open Radio Access Networks (ORAN)
“GraphRAG edge2024local, adopt a tightly coupled entity-event homogeneous structure, hindering the integration of original context and summary information into the graph.”
— NodeRAG: Structuring Graph-based RAG with Heterogeneous Nodes
“existing GraphRAG methods rely on large or fixed structures that introduce redundant or task-irrelevant information, highlighting the challenge of constructing task-relevant graphs.”
— Retrieving Minimal and Sufficient Reasoning Subgraphs with Graph Foundation Models for Path-aware GraphRAG
“Early work~raptor, graphrag emphasize hierarchical summarization and global information integration, but they insufficiently leveraged the fine-grained structural information.”
— GraphSearch: An Agentic Deep Searching Workflow for Graph Retrieval-Augmented Generation
“These methods provide rich global organization, but the retrieval path can remain tied to precomputed graph structure.”
— EfficientGraph-RAG: Structured Retrieval-State Management for Cross-Task Retrieval-Augmented Generation
“its global summaries remain aggregations of local information and do not recover fine-grained cross-chunk relations”
— RAGA: Reading-And-Graph-building-Agent for Autonomous Knowledge Graph Construction and Retrieval-Augmented Generation
“GraphRAG alone underperforms, especially for smaller models like DS-Qwen-7B and LLaMA3.1-8B, likely due to its focus on entity-level information.”
— RAG+: Enhancing Retrieval-Augmented Generation with Application-Aware Reasoning
“GraphRAG~edge2024local applies uniform community detection regardless of query relevance”
— Query-Aware Flow Diffusion for Graph-Based RAG with Retrieval Guarantees
“Compared to standard RAG, GraphRAG methods convert natural language knowledge into graph structures using LLMs, which results in high cost and often causes semantic loss relative to the original content”
— Graph-R1: Towards Agentic GraphRAG Framework via End-to-end Reinforcement Learning
“GraphRAG still lacks mechanisms to explain how individual graph components contribute to the generated responses.”
— Explainable Knowledge Graph Retrieval-Augmented Generation (KG-RAG) with KG-SMILE
“graph-based RAG methods... are fundamentally limited by binary representations that decompose higher-order dependencies into pairwise edges, often leading to information loss.”
— Knowledge Is Not Static: Order-Aware Hypergraph RAG for Language Models
“GraphRAG method uses all the information from the nodes and edges within certain communities. Similarly, LightRAG retrieves the immediate neighbors of query-related nodes to generate answers. The redundant information retrieved in these methods may act as noise, and negatively affecting subsequent generation.”
— PathRAG: Pruning Graph-based Retrieval Augmented Generation with Relational Paths

Beaten on benchmarks

Head-to-head results where a newer method reports beating GraphRAG. Values are copied from the source paper's tables — verify against the cited paper.

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.