This video shows how Neo4j can be used together with Google Cloud Vertex AI generative AI to automatically generate knowledge graphs and then query them. You can check out the source code for the demo here: github.com/neo...
This is great! I have also been building a KG for Chinese Medicine Knowledge with GPT4, and also use GPT4 to turn natural language queries to cypher queries! To me it performs so much better than vectorstore retrieval!
I find straight dictionaries (df) easier to work with during GPT prompt + responses but it adds another schema view style. In case you missed it, langchain has some cypher module support now.
We're still trying to fix out the right mix of fine tuning and prompting for the best results. Regardless, we're trying to feed the model information about both the data model in the knowledge graph and the Neo4j Cypher query language (or some subset of it). The notebook has more detail. And we're always here to help too...
The notebook is designed to run in a Google Cloud Vertex AI managed notebook. It would probably run in an unmanaged notebook. The managed notebook has the advantage of automatically updating itself. It also stops when you're not using it which minimizes bills. Getting the notebook to run in environments outside of Google Cloud such as Colab or on a laptop might require changes but should be possible.