Heidi

Hi,
I am trying to build a KnowledgeGraph from structured Data in Python.
My data is similar to GoogleMaps restaurants.

Example Data:


[
 {
  "name": "Italian Restaurant 1",
  "price": "10EUR",
  "location": "New York",
  "opening": "10am",
  "closing": "10pm",
 },
 {...},
]

I am adding Nodes to the KnowledgeGraph based on this example: https://gpt-index.readthedocs.io/en/v0.6.32/examples/index_structs/knowledge_graph/NebulaGraphKGIndexDemo.html#optional-try-building-the-graph-and-manually-add-triplets
At the section: Building the Graph and Manually adding Triplets

Using this Example I would have Triplets looking similar to this:

[
 ("Italian Restaurant 1", "costs", "10EUR"),
 ("Italian Restaurant 1", "city", "New York"),
 ("Italian Restaurant 1", "opens", "10am"),
 ("Italian Restaurant 1", "closes", "10pm")
]

I would have many of those for my entire Dataset.
Now my general question is how I would create the custom Nodes.

I have been using TextNodes so far, creating a new Node for each Data entry (restaurant).


node = TextNode()
triplets = [...]
for t in triplets:
 index.upsert_triplet_and_node(t, node)

This to me seems wrong, as I don't really have a text for this Node. Using the NodeParsers or from_document options as stated in many other examples is not really applicable to this structured data.

Can someone guide me into the right direction? How should I use Initialize Nodes? What relationships should I use? Should I use a node per Restaurant or per Triplet?

The goal is to query the index for something like "What restaurant is open in Now York at 2pm?"

Find answers from the community

Hi