将Microsoft GraphRAG集成到Neo4j中
Store the MSFT GraphRAG output into Neo4j and implement local and global retrievers with LangChain or LlamaIndex
将MSFT GraphRAG的输出存储到Neo4j中，并使用LangChain或LlamaIndex实现本地和全局检索器
Microsoft’s GraphRAG implementation has gained significant attention lately. In my last blog post, I discussed how the graph is constructed and explored some of the innovative aspects highlighted in the research paper. At a high level, the input to the GraphRAG library are source documents containing various information. The documents are processed using an Large Language Model (LLM) to extract structured information about entities appearing in the documents along with their relationships. This extracted structured information is then used to construct a knowledge graph.
Microsoft的GraphRAG实现最近引起了很大关注。在我的上一篇博客文章中，我讨论了图的构建方式，并探讨了研究论文中强调的一些创新方面。在高层次上，GraphRAG库的输入是包含各种信息的源文档。使用大型语言模型（LLM）处理这些文档，以提取有关文档中出现的实体及其关系的结构化信息。然后，使用这些提取的结构化信息构建知识图。
High-level indexing pipeline as implemented in the GraphRAG paper by Microsoft — Image by author
Microsoft在GraphRAG论文中实现的高级索引流程-作者提供的图像
After the knowledge graph has been constructed, the GraphRAG library uses a combination of graph algorithms, specifically Leiden community detection algorithm, and LLM prompting to generate natural language summaries of communities of entities and relationships found in the knowledge graph.
在构建完知识图谱之后，GraphRAG库使用了一系列图算法，特别是Leiden社区检测算法和LLM提示，来生成关于知识图谱中实体和关系的社区的自然语言摘要。
In this post, we’ll take the output from the GraphRAG library, store it in Neo4j, and then set up retrievers directly from Neo4j using LangChain and LlamaIndex orchestration frameworks.
在本文中，我们将使用GraphRAG库的输出，将其存储在Neo4j中，然后使用LangChain和LlamaIndex编排框架直接从Neo4j设置检索器。
The code and GraphRAG output are accessible on GitHub, allowing you to skip the GraphRAG extraction process.
代码和GraphRAG输出可在GitHub上访问，让您可以跳过GraphRAG提取过程。
Dataset
数据集
The dataset featured in this blog post is “A Christmas Carol” by Charles Dickens, which is freely accessible via the Gutenberg Project.
本博客文章中使用的数据集是查尔斯·狄更斯的《圣诞颂歌》，该数据集可以通过古腾堡计划免费获取。