Is it possible to apply your GCN on directed graphs ?

[pinkfloyd06]

Hi @tkipf ,

I'm wondering if it's possible to apply GCN on directed graphs !

If no, is that related to the fact that :

1. Laplacian is not invariant to rotation. how come ?
2. Laplacian matrix is not circulant ? If yes, why it is important to be circulant ?

Thank you a lot Thomas

[zachlefevre]

A CGN operates on a non-symmetric adjacency matrix, and therefore is already over a directed graph. Somebody correct me if I'm wrong.

[tkipf]

Yes, you could simply take a non-symmetric adjacency matrix and normalize with D^(-1)*A instead of D^(-1/2)*A*D^(-1/2). Have a look at this paper for more details: https://arxiv.org/abs/1703.06103

…

On Mon, Mar 4, 2019 at 1:41 AM zachlefevre ***@***.***> wrote: A CGN operates on a non-symmetric adjacency matrix, and therefore is already over a directed graph. Somebody correct me if I'm wrong. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#91 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHAcYI0Uk30zUuVMGpcvsCqHoM--cJO3ks5vTGvWgaJpZM4bXwtP> .

[fansariadeh]

Yes, you could simply take a non-symmetric adjacency matrix and normalize with D^(-1)A instead of D^(-1/2)AD^(-1/2). Have a look at this paper for more details: https://arxiv.org/abs/1703.06103
…
On Mon, Mar 4, 2019 at 1:41 AM zachlefevre @.**> wrote: A CGN operates on a non-symmetric adjacency matrix, and therefore is already over a directed graph. Somebody correct me if I'm wrong. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#91 (comment)>, or mute the thread https://github.com/notifications/unsubscribe-auth/AHAcYI0Uk30zUuVMGpcvsCqHoM--cJO3ks5vTGvWgaJpZM4bXwtP .

Dear Dr Kipf,

In the code using Cora dataset, you have used a directed graph, which leads to a asymmetric adjacency matrix, but you made this matrix symmetric.
Could you please let me know why it should symmeteric matix?
And is it the same as making the graph directed and then find the adjacency matrix?

Thank you million for sharing your knowledge.

[fansariadeh]

A CGN operates on a non-symmetric adjacency matrix, and therefore is already over a directed graph. Somebody correct me if I'm wrong.

Do you have any idea why in the code the adjancecy matrix has become symmetric while the graph is directed?

[hlzuo]

Yes, you could simply take a non-symmetric adjacency matrix and normalize with D^(-1)A instead of D^(-1/2)AD^(-1/2). Have a look at this paper for more details: https://arxiv.org/abs/1703.06103
…
On Mon, Mar 4, 2019 at 1:41 AM zachlefevre @.**> wrote: A CGN operates on a non-symmetric adjacency matrix, and therefore is already over a directed graph. Somebody correct me if I'm wrong. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#91 (comment)>, or mute the thread https://github.com/notifications/unsubscribe-auth/AHAcYI0Uk30zUuVMGpcvsCqHoM--cJO3ks5vTGvWgaJpZM4bXwtP .

I have a similar question regarding citation relationships. While citation relationships are typically considered to be directed, in the paper, the links between nodes are treated as non-directed. I'm curious if there is a specific reason for this approach, or if it was simply done for the sake of simplicity.

I would greatly appreciate it if someone could provide a more detailed explanation on this matter.