Multi-level graph embeddings

[alzaia]

Hello guys, I have a use case where I want to see if combining 2 different levels of my graphs could help for my task. I have my whole graph of nodes/edges where I'm doing node classification. These nodes are also part of subgraph ids (which I do not use for now). What I wanted is, in parallel of my node embeddings operation, to do a subgraph embedding operation that would learn subgraph features that could then be used for my node classification task. For example: node i has its node features Fi, but is also part of subgraph m that has subgraph features Fm. Are there any layers/models in PyG that would be worth checking out? Or just relevant similar examples to my case? Thanks a lot!

[rusty1s]

This sounds super similar to one of my research topics, see here. Happy to help with any implementation questions.

[alzaia]

Thanks @rusty1s for the always fast and helpful replies! I checked the links, do you have this implementation already in PyG, or at least something very similar to that concept of hierarchical embeddings in PyG? I wanted to do a quick first experiment first before going full into it.

[rusty1s]

There is a PyG implementation here, but I haven't run it in a long time :(