Correct way of importing an RDF Knowledge Graph into PyG

[igrangel]

Hi!

I have an RDF/OWL based KG which have some classes, object properties, etc. I want to use it in PyG.

1. I have read it with rdflib from its turtle serialization

rg = RDFGraph()
rg.parse("data.ttl", format='turtle')
print("rdflib Graph loaded successfully with {} triples".format(len(rg)))

2. Then I have transformed it into the NetworkX library

G = rdflib_to_networkx_graph(rg)
print("networkx Graph loaded successfully with length {}".format(len(G)))

3. And finally, I used the method from_network to bring it into the PyG data.

from torch_geometric.utils import from_networkx
pG = from_networkx(G)

Is this the correct way to bring these types of KGs into PyG?

[rusty1s]

In theory, yes. Although you probably want to group all your different attributes into single feature matrices afterwards. Let me know if you need any further help :)

[igrangel]

Thanks a lot for your reply. Would be great to have some example, do you happen to have something like this?

What do you mean by

Although you probably want to group all your different attributes into single feature matrices afterwards

If you can be more specific, I would really appreciate it ;).

[rusty1s]

I mean that, depending on how your input graph looks like, you will have many different feature vectors in the resulting data object. You can simply print it out to confirm. When working with GNNs, you will probably want to group them into a single feature matrix of shape [num_nodes, num_total_features].