GraphSage unsup implementation in ./examples

[OliverRock]

I am trying to implement an unsupervised GraphSage model but I am slightly confused by this example. I think the loss function here is not the same as in the original paper because the code does not use random walk to find the nodes and it is passing the labels so it is not unsupervised. Are these observations correct or have I misunderstood something?

Here is the part of the paper which defines the loss function:

[rusty1s]

Yes, the GraphSAGE unsupervised example doesn't use fixed-length random walks as negative samples but randomly sampled one. It is still unsupervised (not sure what you mean by it is using the label); it uses positive and negative links to compute the constrastive loss.