* https://github.com/microsoft/graphrag
* https://microsoft.github.io/graphrag/
- paper: From Local to Global: A Graph RAG Approach to Query-Focused Summarization (2024/4)
- https://arxiv.org/abs/2404.16130
- https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data (2024/2)
- https://www.microsoft.com/en-us/research/blog/graphrag-new-tool-for-complex-data-discovery-now-on-github/ (2024/7)
- 出了 Azure graphrag-accelerator 版本
- paper: Retrieval-Augmented Generation with Knowledge Graphs for Customer Service Question Answering (2024/4)
- https://arxiv.org/abs/2404.17723
- https://x.com/iheycc/status/1785357774718976116 （2024/5/1)
- linkedin 導入客服問答
- paper: Graph Retrieval-Augmented Generation: A Survey (2024/9)
- https://arxiv.org/abs/2408.08921
* paper: HybridRAG: Integrating Knowledge Graphs and Vector Retrieval Augmented Generation for Efficient Information Extraction (2024/8)
* https://arxiv.org/abs/2408.04948
* https://x.com/omarsar0/status/1822832843455648000 (2024/8/12)
* Advanced RAG 12: Enhancing Global Understanding (2024/6)
* 有和其他方法一起比較
* https://ai.gopubby.com/advanced-rag-12-enhancing-global-understanding-b13dc9a8db39
* Implementing ‘From Local to Global’ GraphRAG with Neo4j and LangChain: Constructing the Graph (2024/6/10)
* https://medium.com/neo4j/implementing-from-local-to-global-graphrag-with-neo4j-and-langchain-constructing-the-graph-73924cc5bab4
* The GraphRAG Manifesto: Adding Knowledge to GenAI
* https://neo4j.com/blog/graphrag-manifesto/ (2024/6/11)
* Knowledge Graph + RAG > Naive RAG (2024/7/11)
* https://div.beehiiv.com/p/knowledge-graph-rag-naive-rag
* From Knowledge Graphs to GraphRAG: Advanced Intelligent Data Retrieval (2024/7/17)
* https://div.beehiiv.com/p/knowledge-graphs-graphrag-advanced-intelligent-data-retrieval
* 網路大大測試 GraphRAG 跑一個電影介紹文章, 並記錄Indexing stage中每個階段用的prompt, 可以觀察它是如何用LLM建立graph:
* https://x.com/Evan_Lin/status/1813016483573710853
* https://steep-swamp-d7f.notion.site/GraphRAG-b24660e87ab7479ea019e39746978325
* 影片 GraphRAG: The Marriage of Knowledge Graphs and RAG: Emil Eifrem (neo4j) (2024/8)
* https://www.youtube.com/watch?v=knDDGYHnnSI
## LazyGraphRAG
取“惰性”的索引方式，僅建立與向量RAG相當的輕量索引，並在查詢時才逐步擴充需要的圖譜信息
* https://www.microsoft.com/en-us/research/blog/lazygraphrag-setting-a-new-standard-for-quality-and-cost/ (2024/11/25)
## 其他實作
* nano-graphrag
* https://github.com/gusye1234/nano-graphrag
* LightRAG
* 基於 nano-graphrag 開發
* https://github.com/HKUDS/LightRAG
* https://www.jiqizhixin.com/articles/2024-10-14-2
* https://medium.com/@samarrana407/lightrag-a-graphrag-alternative-636d644a05cb
* 使用較少的 API 呼叫和輕量級模型
* 允許對圖形進行增量更新，而無需完全重新生成
* 支持雙層檢索（本地和全球），這提高了回應質量
* Fast GraphRAG framework
* https://github.com/circlemind-ai/fast-graphrag
* AutoFlow
* https://github.com/pingcap/autoflow/
* AWS
* https://aws.amazon.com/tw/blogs/database/introducing-the-graphrag-toolkit
* https://github.com/awslabs/graphrag-toolkit
* Amazon Neptune
* https://medium.com/@bechbd/knowledge-graphs-and-generative-ai-graphrag-with-amazon-neptune-and-llamaindex-part-1-39cd7255bac4
* neo4j
* https://neo4j.com/developer-blog/graphrag-ecosystem-tools/
* https://neo4j.com/developer-blog/microsoft-graphrag-neo4j/
* https://neo4j.com/developer-blog/graphrag-agent-neo4j-milvus/
* langchain + neo4j
* https://medium.com/towards-data-science/building-knowledge-graphs-with-llm-graph-transformer-a91045c49b59
* Memgraph https://memgraph.com/
* Nebula https://www.nebula-graph.io/
## Llamaindex
* https://medium.com/@transformergpt/unleashing-the-power-of-knowledge-graphs-in-retrieval-augmented-generation-rag-step-by-step-84c2adc66c1c (2024/5/6)
* https://www.llamaindex.ai/blog/introducing-the-property-graph-index-a-powerful-new-way-to-build-knowledge-graphs-with-llms (2024/5/29)
* https://www.llamaindex.ai/blog/customizing-property-graph-index-in-llamaindex (2024/6/11)
* https://www.youtube.com/watch?v=LDh5MdR-CPQ
* https://www.youtube.com/playlist?list=PLTZkGHtR085ZYstpcTFWqP27D-SPZe6EZ
* https://medium.com/neo4j/entity-linking-and-relationship-extraction-with-relik-in-llamaindex-ca18892c169f (2024/8/12)
* https://x.com/llama_index/status/1827367293376184418 (2024/8/24)
* https://www.llamaindex.ai/blog/constructing-a-knowledge-graph-with-llamaindex-and-memgraph (2024/11/21)
* text2cypher https://www.llamaindex.ai/blog/building-knowledge-graph-agents-with-llamaindex-workflows (2025/1/15)
## 批評
* When (not) to Use GraphRAG (2024/8/25)
* https://pub.towardsai.net/when-not-to-use-graphrag-02a80d77fcbf
* GraphRAG 在處理複雜的多面向查詢時最為有用，這些查詢需要遍歷多個資訊片段
* 對於具有簡單關係的簡單數據集（和單一面向的查詢）或主要處理結構化文本文件時，傳統的 RAG 或先進的搜索方法可能更有效
* GraphRAG 在索引創建和查詢處理中引入了額外的複雜性和計算開銷，這對於簡單的信息查詢任務可能不值得
* GraphRAG 在複雜的、相互連結的數據集和需要深入關聯理解的查詢中，提供了顯著的資訊檢索能力提升。然而，它也帶來了更高的複雜性和資源需求
* GraphRAG Analysis (2024/7/10)
* https://www.jonathanbennion.info/p/graphrag-analysis-part-1-how-indexing
* https://www.jonathanbennion.info/p/graphrag-analysis-part-2-graph-creation
* GraphRAG 增加了忠實度，但並未改善其他 RAGAS 指標 - 知識圖譜的投資回報率可能無法證明其炒作的合理性
* https://x.com/Nils_Reimers/status/1813266482672939363
* 大多數「知識圖譜」實際上只是元數據過濾（例如僅搜索 2024 年創建的文檔或標籤為 X 的文檔）。有時這個術語用於多字段搜索（以及字段的提升）。 在大規模建立實際的知識圖譜是非常困難的
* https://x.com/fuckpoasting/status/1821364149089353957
* 人們發明像「GraphRAG」這樣的詞，實際上只是一次外鍵查詢
* https://x.com/SaZhang_/status/1813121432752431440
* Graph 又慢又贵维护还麻烦
* 增量更新是個大問題
* https://pub.towardsai.net/critical-challenges-encountered-in-the-practical-implementation-of-rag-3974b69eaac1
* 似乎有在 v0.4.0 版本中引入了**增量索引功能**，允許在已有知識圖譜基礎上合併新內容，而不必完全重建索引
* 雖可局部更新，但若新資料引入大量全新實體或使原有社群結構發生劇變，可能仍需考慮定期進行一次全面重建，以獲得最優的圖譜結構（尤其當累積小改動可能導致子最佳的分群時）。目前的增量合併策略相對「**簡單樸素**」
---
* Deep Research (2025/2/12)
* https://chatgpt.com/c/67ab7439-6a60-8008-a25f-1ba3569a27c7