Relational Graph Attention Network is the extension of the original GAT, it is proposed in paper: Graph Attention Network
Why use Relational Graph Attention Network?
As to original GAT, we can compute the coefficient between parent node and its child nodes.
Computing the Coefficient Between Parent Node and Child Nodes
However, the relation between each child node and parent node may be different. For example, words in a sentence have different part-of-speech. The original GAT does not process this kind of relation.
In dependency tree, same words in a graph may have different dependency relations, we take it in consideration in R-GAT.
R-GAT is defined as:
Here \(r_{i,j}\) is the relation between node i and j.
We can find it is different from GAT. R-GAT amis to determine the different weight of each relation, it can tell you which relation is most important.
How to use R-GAT?
We usually incorporate it with GAT, here is an example:
The final output is: