A Survey of Graph Retrieval-Augmented Generation (Graph RAG) for Customized Large Language Models
Overview
- Paper: A Survey of Graph Retrieval-Augmented Generation for Customized Large Language Models
- arXiv: https://arxiv.org/abs/2501.13958
- Date: January 2025
- Type: Survey
- Authors: Qinggang Zhang, Shengyuan Chen, Yuanchen Bei, Zheng Yuan, Huachi Zhou, Zijin Hong, et al. (12 authors total)
- Graph RAG augments conventional retrieval-augmented generation by leveraging knowledge graphs and graph-structured data, improving multi-hop reasoning, factual grounding, and domain customization of LLMs.
- The survey provides a unified taxonomy that decomposes Graph RAG systems into core components: graph-based representation learning, retrievers, rerankers, planners, generators, and feedback mechanisms.
- It contrasts major retrieval paradigms: dense retrieval (high recall, low latency), late interaction (e.g., ColBERT, higher precision but larger indices), generative IR (directly generating document identifiers), and agentic search (search as sequential decision-making with reflection).
- Evaluation typically relies on benchmarks such as MS MARCO, BEIR, and Natural Questions, with metrics including nDCG@10, MRR, Recall@k, task success rate, latency, and token cost; baselines include BM25, dense retrievers, cross-encoder rerankers, and retrieval-free LLMs.
- 2019–2021: BERT-based reranking and DPR establish neural retrieval foundations.
- 2022–2023: RAG and FreshLLM drive retrieval-generation integration.
- 2024: Conversational/agentic search and generative recommender systems (Gen-RecSys) accelerate.
- 2025–2026: RL-trained search agents, Deep Research, and GraphRAG emerge as major growth directions.
- Insufficient unified benchmarks; private data limits reproducibility.
- Evaluation bias in LLM-as-judge; offline metrics diverge from real user satisfaction.
- Latency, cost, safety, and hallucination constraints in industrial deployment.
- Future work: fine-grained process supervision, retrieval-reasoning co-training, enterprise metadata governance, multimodal and cross-lingual consistency, and deeper fusion with knowledge graphs and structured databases.
- A Survey on Retrieval-Augmented Text Generation for Large Language Models (arXiv:2404.10981)
- Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG (arXiv:2501.09136)
- RAG vs. GraphRAG: A Systematic Evaluation and Key Insights (arXiv:2502.11371)
- RAFT: Adapting Language Model to Domain Specific RAG (July 2024)
Key points
Historical trajectory
Open challenges and future directions
Practical implications
1. Architecture: Cascade retrieval + rerank + generation remains mainstream, but agentic paradigms make the number and strategy of retrieval steps itself learnable. 2. Data: High-quality instruction data and click/session logs matter as much as model design; synthetic data risks leakage and distribution shift. 3. Engineering checklist: Consider PII handling and embedding versioning, p99 latency budgets with cascades and early stopping, interleaved online experiments, retrieval poisoning defenses, and per-query token/GPU cost (routing, distillation, hybrid sparse-dense retrieval).