GCN vs R-GCN

[Professoroo]

Hello Dr kipf,

I am wondering about the relation between GCN and R-GCN. In the former one, it is based on spectral approach of first order approximation of a localised filter. The second one offers weights instead of adjacency.
Using the constant (c) in R-GCN as D^-1, does this explicitly converge to the spectral GCN? In other words, are we able to write it as D^-1AXW? (D=degree matrix, A=adjacency matrix, X=feature matrix, W=trainable weights).

R-GCN looks more like a combination of linear layers corresponding to the adjacent nodes.

Thank you in advance.