Awesome GraphRAG

This repository contains a curated list of resources on graph-based RAG methods.

Continuously updating, stay tuned!

Introduction

Large language models (LLMs) have demonstrated remarkable capabilities across diverse tasks, achieving significant advancements in text comprehension, question answering, and content generation. However, their performance in knowledge-intensive tasks, particularly those requiring domain expertise, remains suboptimal due to the folllowing limitations: their pretrained knowledge, though broad, lacks depth in specialized areas due to reliance on general-domain data, resulting in inconsistencies with current domain standards; they struggle with precise, multi-step reasoning required by domain-specific rules, often failing to maintain logical consistency and accuracy in complex reasoning chains; and they exhibit limited context sensitivity, frequently misinterpreting or oversimplifying domain-specific terms and concepts that vary meaning based on context.

Retrieval-Augmented generation (RAG) offers a promising solution to customize LLMs for specific domains. Rather than retraining LLMs to incorporate updates, RAG enhances these models by leveraging external knowledge from text corpora without modifying their architecture or parameters. This approach enables LLMs to generate responses by leveraging not only their pre-trained knowledge but also real-time retrieved domain-specific information, thereby providing more accurate and reliable answers. However, the practical effectiveness of RAG systems in real-world applications is hindered by limitations in complex query comprehension, difficulties in synthesizing distributed domain knowledge, inherent constraints of LLMs, and issues with system efficiency and scalability[1].

To address these limitations, graph retrieval-augmented generation (GraphRAG) has recently emerged as a new paradigm to customize LLMs with well-organized background knowledge and improved contextual reasoning. GraphRAG, formally defined as a specialized subclass of the RAG framework, utilizes graph structures to systematically organize and retrieve domain-specific knowledge. Its workflow encompasses two key stages: offline indexing and online retrieval. The overall pipeline is as follows. We collect the recent influential papers and benchmarks about GraphRAG. The following contents are listed in chronological order of publication.

📑 Research Paper

Date	Venue	Title	Code	Notes
2025-03-18	Arxiv	KG-IRAG: A Knowledge Graph-Based Iterative Retrieval-Augmented Generation Framework for Temporal Reasoning	No	Graphs for Knowledge Indexing & Graphs as Knowledge Carrier(KG construction from corpus)
2025-03-14	ICML 2025 Spotlight	Taming Knowledge Conflicts in Language Models	Yes	Mechanisms Controlling LLMs' Behavioral Preferences
2025-03-13	Arxiv	Retrieval-Augmented Generation with Hierarchical Knowledge	Yes	Graphs for Knowledge Indexing & Graphs as Knowledge Carrier(KG construction from corpus)
2025-03-06	Arxiv	In-depth Analysis of Graph-based RAG in a Unified Framework	Yes	modularizing and decoupling some graph-based RAG methods to unveil the mystery behind them and share fun and valuable insights!
2025-02-20	Arxiv	(HippoRAG 2)From RAG to Memory: Non-Parametric Continual Learning for Large Language Models	Yes	Graphs for Knowledge Indexing & Graphs as Knowledge Carrier(KG construction from corpus)
2025-02-14	Arxiv	ArchRAG: Attributed Community-based Hierarchical Retrieval-Augmented Generation	No	Graphs for Knowledge Indexing & Graphs as Knowledge Carrier(KG construction from corpus)
2025-02-08	NAACL 2025	(KG2RAG)Knowledge Graph-Guided Retrieval Augmented Generation	Yes	Graphs for Knowledge Indexing & Graphs as Knowledge Carrier(KG construction from corpus)
2025-02-06	The ACM Web Conference 2025	MedRAG: Enhancing Retrieval-augmented Generation with Knowledge Graph-Elicited Reasoning for Healthcare Copilot	Yes	Graphs as Knowledge Carrier(KG construction from corpus)
2024-12-17	Arxiv	SimGRAG: Leveraging Similar Subgraphs for Knowledge Graphs Driven Retrieval-Augmented Generation	Yes	Graphs as Knowledge Carrier(KG construction from corpus & with existing KGs)
2024-10-28	ICLR 2025	Simple Is Effective: The Roles of Graphs and Large Language Models in Knowledge-Graph-Based Retrieval-Augmented Generation	Yes	Graphs as Knowledge Carrier(KG construction from corpus)
2024-10-08	Arxiv	LightRAG: Simple and Fast Retrieval-Augmented Generation	Yes	Graphs as Knowledge Carrier(KG construction from corpus)
2024-05-23	NeurIPS 2024	HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models	Yes	Graphs as Knowledge Carrier(KG construction from corpus)
2024-04-24	Arxiv	From Local to Global: A Graph RAG Approach to Query-Focused Summarization	Yes	Graphs as Knowledge Carrier(Knowledge Graph Construction from Corpus)
2024-02-28	ACL 2024 Findings	Cutting Off the Head Ends the Conflict: A Mechanism for Interpreting and Mitigating Knowledge Conflicts in Language Models	No	Mechanisms Controlling LLMs' Behavioral Preferences
2024-01-22	ACL 2024 Long Papers	Blinded by Generated Contexts: How Language Models Merge Generated and Retrieved Contexts When Knowledge Conflicts?	Yes	LLMs' Behavioral Preferences Under Knowledge Conflicts
2023-10-24	EMNLP 2023	Characterizing Mechanisms for Factual Recall in Language Models	No	Mechanisms Controlling LLMs' Behavioral Preferences
2023-09-29	ACL 2024 Long Papers	Intuitive or Dependent? Investigating LLMs' Behavior Style to Conflicting Prompts	No	LLMs' Behavioral Preferences Under Knowledge Conflicts
2023-05-22	ICLR 2024 Spotlight	Adaptive Chameleon or Stubborn Sloth: Revealing the Behavior of Large Language Models in Knowledge Conflicts	Yes	LLMs' Behavioral Preferences Under Knowledge Conflicts
2022-10-25	EMNLP 2022	Rich Knowledge Sources Bring Complex Knowledge Conflicts: Recalibrating Models to Reflect Conflicting Evidence	No	LLMs' Behavioral Preferences Under Knowledge Conflicts
2021-09-10	EMNLP 2021	Entity-Based Knowledge Conflicts in Question Answering	Yes	LLMs' Behavioral Preferences Under Knowledge Conflicts

🔥 Hot Methods

Date	Method	Graph Type	Index Component	Retrieval Primitive	Retrieval Granularity	Code
2024-10-31	fast-graphrag	Textual Knowledge Graph	Entity	Entities in question	Entity, Relationship, Chunk	Yes
2024-10-08	LightRAG_Local	Rich Knowledge Graph	Entity, Relationship	Low-level keywords in question	Entity, Relationship, Chunk	Yes
2024-10-08	LightRAG_Global	Rich Knowledge Graph	Entity, Relationship	High-level keywords in question	Entity, Relationship, Chunk	Yes
2024-10-08	LightRAG_Hybrid	Rich Knowledge Graph	Entity, Relationship	Both high- and low-level keywords	Entity, Relationship, Chunk	Yes
2024-05-23	HippoRAG	Knowledge Graph	Entity	Entities in question	Chunk	Yes
2024-04-24	GraphRAG_Local	Textual Knowledge Graph	Entity, Community	Question vector	Entity, Relationship, Chunk, Community	Yes
2024-04-24	GraphRAG_Global	Textual Knowledge Graph	Community	Question vector	Community	Yes
2024-01-31	RAPTOR	Tree	Tree node	Question vector	Tree node	Yes

Note: This repo summarizes the graph constructed from the corpus into the following categories according to [2]:

1. Passage Graph: each chunk represents a node. If two chunks contain a number of the same entities larger than a threshold, link an edge for these two nodes.
2. Tree: a tree is built progressively by clustering leaf nodes at each layer and using an LLM to generate higher-level summary nodes for clusters with multiple children.
3. Knowledge Graph: extracting entities and relationships from each chunk
4. Textual Knowledge Graph: a specialized KG , with the key difference being that in a TKG, each entity and relationship is assigned a brief textual description.
5. Rich Knowledge Graph: an extended version of TKG, containing more information, including textual descriptions for entities and relationships, as well as keywords for relationships.

🧐 Survey Papers

Date	Venue	Title	Repo
2025-01-21	Arxiv	A Survey of Graph Retrieval-Augmented Generation for Customized Large Language Models	Yes
2024-12-31	Arxiv	Retrieval-Augmented Generation with Graphs (GraphRAG))	Yes
2024-08-15	Arxiv	Graph Retrieval-Augmented Generation: A Survey	Yes
2024-05-10	KDD 2024	A Survey on RAG Meeting LLMs: Towards Retrieval-Augmented Large Language Models	No
2024-03-13	Arxiv	Knowledge Conflicts for LLMs: A Survey	Yes

🥇 Benchmark

Dataset	Task	Info	Metrix
NaturalQuestions	Simple QA	A general-domain simple QA dataset based on the Wikipedia dump knowledge base, testing the model’s ability to extract answers from encyclopedia content.	EM, F1 Score
PopQA	Simple QA	PopQA is a large-scale open-domain question answering (QA) dataset, consisting of 14k entity-centric QA pairs. Each question is created by converting a knowledge tuple retrieved from Wikidata using a template. Each question come with the original subject_entitiey, object_entityand relationship_type annotation, as well as Wikipedia monthly page views.	Accuracy, F1 Score
SimpleQuestion	Simple QA	SimpleQuestions is a large-scale factoid question answering dataset. It consists of 108,442 natural language questions, each paired with a corresponding fact from Freebase knowledge base. Each fact is a triple (subject, relation, object) and the answer to the question is always the object.	EM, F1 Score
WebQ	Simple QA	This dataset consists of 6,642 question/answer pairs. The questions are supposed to be answerable by Freebase, a large knowledge graph. The questions are mostly centered around a single named entity.	Precision, Recall
WebQSP	Simple QA	A general simple QA dataset finishing semantic parsing and QA tasks based on Freebase.	EM
MuSiQue	Multi-hop QA	A general multi-hop QA dataset with implicit in-data knowledge, testing multi-step reasoning ability.	answer_f1, support_f1
2WikiMultihopQA	Multi-hop QA	A general multi-hop QA dataset with in-data implicit knowledge, emphasizing cross-Wikipedia paragraph reasoning.	EM, F1 Score
HotpotQA	Multi-hop QA	A question answering dataset featuring natural, multi-hop questions, with strong supervision for supporting facts to enable more explainable question answering systems based on Wikipedia dump.	EM, F1 Score, Accuracy, Recall
CWQ	Multi-hop QA	A dataset for answering complex questions that require reasoning over multiple web snippets via Freebase.	Accuracy
MultiHop-RAG	Multi-hop QA	A Dataset for Evaluating Retrieval-Augmented Generation Across Documents.	Hits@10, Hits@4, MAP@10, MRR@10, Accuracy, Recall
MetaQA	Multi-hop QA	A movie-domain multi-hop QA dataset relying on the in-data movie knowledge base for reasoning.	Accuracy
Mintaka	Complex QA	Mintaka is a complex, natural, and multilingual question answering (QA) dataset composed of 20,000 question-answer pairs elicited from MTurk workers and annotated with Wikidata question and answer entities.	EM, F1 Score, Hits@1
GrailQA	Complex QA	A a new large-scale, high-quality dataset for question answering on knowledge bases (KBQA) on Freebase with 64,331 questions annotated with both answers and corresponding logical forms in different synta	Accuracy
UltraDomain	Complex QA	An 18-domain complex QA dataset testing cross-domain complex question-handling ability.	LLM as judger, like GraphRAG & LightRAG
TriviaQA	Complex QA	TriviaqQA is a reading comprehension dataset containing over 650K question-answer-evidence triples. TriviaqQA includes 95K question-answer pairs authored by trivia enthusiasts and independently gathered evidence documents, six per question on average, that provide high quality distant supervision for answering the questions.	Accuracy
LC-QuAD v2	Complex QA	LC-QuAD 2.0 is a Large Question Answering dataset with 30,000 pairs of question and its corresponding SPARQL query. The target knowledge base is Wikidata and DBpedia, specifically the 2018 version.	Accuracy
KQAPro	Large-scale Complex QA	A large-scale dataset of complex question answering over Wikidata. The questions are very diverse and challenging, requiring multiple reasoning capabilities including compositional reasoning, multi-hop reasoning, quantitative comparison, set operations, and etc.	Accuracy
FACTKG	Fact Verification	Fact Verification via Reasoning on Knowledge Graphs. It consists of 108k natural language claims with five types of reasoning: One-hop, Conjunction, Existence, Multi-hop, and Negation based on DBpedia.	Accuracy
DDXPlus	Medical QA & Diagnostic support	A new large-scale dataset for Automatic Symptom Detection (ASD) and Automatic Diagnosis (AD) systems in the medical domain.	IL (Interaction Length), GTPA (Ground Truth Probability Above Threshold), DDR/DDP/DDF1 (Differential Diagnosis Recall/Precision/F1), DSP/DSR/DSF1 (Severity Precision/Recall/F1)
NarrativeQA	QA & Discourse Understanding	An English-lanaguage dataset of stories and corresponding questions designed to test reading comprehension, especially on long documents.	BLEU, MET, ROU

📊 Existing Knowledge Graphs

Date	Venue	Title	Homepage	Domain
2023-08-23	SIGIR 2024	YAGO 4.5: A Large and Clean Knowledge Base with a Rich Taxonomy	Yes	General
2023-02-09	Bioinformatics	The scalable precision medicine open knowledge engine (spoke): a massive knowledge graph of biomedical information	Yes	Biomedical
2018-11-22	Nucleic acids research	String v11: protein–protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets	Yes	Protein-protein interaction prediction
2018-05-12	LREC workshop	Lynx: building the legal knowledge graph for smart compliance services in multilingual europe	Yes	Legal

✅ Citation

• [1] A Survey of Graph Retrieval-Augmented Generation for Customized Large Language Models.
• [2] In-depth Analysis of Graph-based RAG in a Unified Framework.