Just wondering whether the spectral GNN conv layers can work on homogenous directed multigraphs:
(especially SGConv and GraphConv)
If yes, could you please briefly describe the adjacency matrix and the node feature/embedding matrix for multigraphs?
Thank you!
DGL conv layers: dgl.nn (PyTorch) — DGL 0.8.2post1 documentation
In general, GNN conv layers can only work on undirected graphs. For directed graphs, you will need to either make it bidirectional or add another edge type (and use a heterogeneous GNN) to maintain the directional information. The primary reason is to allow messages to flow backwards along the edges as well.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.