Since the obtained node embedding of Deepwalk is not deterministic ( different outputs even with the same graph), I wonder if there is a generic pretrained GCN model that can be used for feature extractions ( like the VGG model or Resnet model for images). The graph tends to be featureless, meaning only the adjacent matrix exits.
If not, can someone tell me that if there is a method working like Deepwalk, but yielding the fixed embeddings for the same two graphs? I think I have to resort to traditional matrix transformations, right?
Thanks in advance!