How do I access the content used for the Neo4j LLM Knowledge Graph Builder demo?

Neo4j Graph Platform
1

Our Seed Community of Care (CoC) is exploring the Neo4j LLM Knowledge Graph Builder to see if it might facilitate focused knowledge sharing across our CoCs. Does anyone know if the content used to support Neo4J’s demo and it’s associated graph schema is publicly available for review? In addition, how about sample content used to test the schemas available in the app?

We‘ve reviewed the demo videos as well as taken an initially look at the GitHub. Before embarking on our own project, we want to walk through a working reference example to ensure we get all our components lined up.

Here are three related questions:
1. Do we have to ensure that all the concepts in our schema are represented in the LLM we select?
2. How do we configure the LLM(s) we want the app to use?
3. Do we have to populate any vector embeds related to the schema or does Neo4J manage that process for us?

Thank you!

Robert

2

Hi Robert,

thanks for your interest in the knowledge graph builder.
Everyone is using their own datasets to explore what's possible.

I use a mixture of wikipedia articles of Google, DeepMind, AlphaFold, Sundar and a few others together with youtube transcripts of the Gemini launch.

But I also used PDFs, Wikipeida articles and youtube videos about specific topics, like the Berlin main station.

As you're the expert on your own data, I suggest to use something where you can also reason about the extracted information and also provide a sensible graph model.

The concepts are not in the LLM, the concepts are in your text, you just use the language skills of the LLM together with some prompting and an optional graph schema (that you can define) to extract entities.

  1. Features documentation - Neo4j Labs
  2. see: Documentation for local deployments - Neo4j Labs
  3. embedding vectors are automatically added for text chunks and entities, those are then used for downstream tasks like Q&A or de-duplication

HTH

Michael

3

Thank you for the clarification, Michael!

Robert Damashek
571-221-3914 Mobile

4

Thanks, Michael. Sorry for the delay. I’ll try it, but if I provide a graph schema, what process does the builder use to map the extracted content semantically to my model?

Robert