[Question]: Could you explain how GraphRAG is currently implemented?

infiniflow / ragflow

RAGFlow is an open-source RAG (Retrieval-Augmented Generation) engine based on deep document understanding.

https://ragflow.io

Apache License 2.0

17.47k stars 1.78k forks source link

[Question]: Could you explain how GraphRAG is currently implemented? #2471

Open Defozo opened 1 day ago

Defozo commented 1 day ago

Describe your problem

What is being done during file parsing and where is it stored (the knowledge graph)? What is being done during inferencing?

KevinHuSh commented 1 day ago

During parsing, use LLM to extract entities, replations, comunities, then, every entity/relation/community store in ES. During inference, use keywords to retrieve related entity/relation/community, then, extract the answer by LLM.

The code is in folder graphrag/.

Defozo commented 10 hours ago

Does it currently work with any agent workflow? Eg. with a "General-purpose chatbot"?