Hello,
Great work with the LLM Knowledge Graph builder. I'm wondering two things more to really consider using it for our project :
The graph builder looks quite nice for ad'hoc analysis and demos. In our usecase, we already have a Neo4J instance up. We would like to enable a link between what is created by the tool, and our current graph.
First, we would like to analyse documents that would be contained in our graph. For example, we have some nodes pointing to URL. Specifying our graph, and the nodes in question to get the URL and the text found there would be a killer feature. Of course we can also build this ourselves, but it would save us quite some features development for something you already do and that we could potentially use out of the box.
Second, and it would be reaaaally great if we could have a process to insert back to our graph what has been created by the Knowledge Graph builder to further enrich it. Even if we are not yet able to connect our current Aura Database to the LLM Knowledge Graph builder, we could definitely have a process to upload some files into S3 and launch the tool, so that it can then insert what has been found back into the graph.
Any thoughts / useful links / on-going development on these two topics ?
Thanks and regards,
Alexis
I didn't get what you are trying to achieve?
It looks like you are trying to get an entity (document), split it (chunking) and create a graph with those embeddings that can be referenced?
That sounds like RAG.
Hey Josh,
I don't know how to express it differently. Let me try with other words :
1st usecase is : use LLM Knowledge Graph builder capacities to analyse documents that referenced in some nodes inside the graph, having an attribute "url" which value could be used to get the files automatically. What you suggest here with "It looks like you are trying to get an entity (document), split it (chunking) and create a graph with those embeddings that can be referenced?" goes in this direction I guess
Another way would be to be able to call programmatically the LLM Knowledge Graph builder ourselve, from a script that would give as input the URL.
2nd usecase is : how to merge the created graph by LLM Knowledge Graph builder with the existing graph. Is that possible today ?
They still sound like a RAG to me, with some user interface.
The URL attribute will need to be interpreted by some code somewhere which is the one orchestrating the document ingestion -> LLM chunking -> Neo4J RAG.
Alright I will deep dive into Neo4J RAG, I am not so familiar with it and might find some answers there !
Though, my questions are related to the features of the LLM Knowledge Graph builder "product". It seems, based on your answer, that it can not solve our two usecases "out of the box".
I would be really interested in knowing if the graph generated by the LLM Knowledge Graph builder could be inserted in an existing Neo4J Aura graph later on. And if it is already (or there is a "not so complicated way" to do so, please let me know ! It would look like
(based on
Neo4j LLM Knowledge Graph Builder - Extract Nodes and Relationships from Unstructured Text - Neo4j Labs)
It depends on where you run everything - you can go straight from LLM to Neo4J or have some intermediate embedding generator that then is uploaded.
It is all in the architecture.
