paper on GATv2 using Edge embedding

[XIANG64Young]

Hi, I would like to know more details about the use of edge embedding for GATv2.
Just as How Attentive are Graph Attention Networks demonstrated the change from GAT to GATv2, Edge-Featured Graph Attention Network demonstrates the effectiveness of GAT using edge embeddings.
However, I found no relevant articles proving the validity of using edge embedding on GATv2. If anyone finds it, please get back to me in time.

[rusty1s]

What do you mean with validity of using edge embeddings? Can you clarify?

[XIANG64Young]

In How Attentive are Graph Attention Networks, the formula is $e(h_i, h_j)= \mathbf{a}^{\top}\mathrm{LeakyReLU}(W [h_i , \Vert , h_i])$.
But in docs, there exist a new formula using edge embeddings: $e(h_i, h_j, e_{ij})= \mathbf{a}^{\top}\mathrm{LeakyReLU}(W [h_i , \Vert , h_i ,\Vert \ e_{ij}])$.
I wondered if there was some paper that demonstrated the validity of the new formula in some experiments or mathematics, just like Edge-Featured Graph Attention Network on GAT.

[rusty1s]

Actually no, this was just added by us based on popular request.

[XIANG64Young]

Well. Thank you for your generous response!