How can I include edge attributes in graph learning

Questions
#1

Hi there,

I am learning to use the Graph neutral network. I have read GCN and GAT; however, I realize that these two methods seem to only take node attributes as input/feature, but do not support edge attributes.

Are there ways to incorporate edge attributes to GCN and GAT? Or are there other methods that are more appliable to learning with both edge and node attributes? Thanks

#2

You may find this discussion helpful. Roughly speaking, you can separately propagate node features and edge features and then combine the aggregated messages from node features and edge features.