Ask for some recommend methods.

[StefanIsSmart]

If I have a graph, that is N nodes and N * N edges (Nodes and Edges all have many features), Is there any graph NN recommends using?
As we all know, the graph is very easy to over-smoothing. (Especially fully connected NetWork.)
I also need a GNN that could catch both the Node and Edge features to get an embedding of them.
Is there some recommended model?

Thanks!

[rusty1s]

For fully-connceted networks, you will need some anisotropic GNN, e.g., using attention, or something like EdgeConv.