Rex Ying

Thanks for the interest! For now you can just cite the github repository. @misc{DeepSNAP, author = {Rex Ying, Jiaxuan You, Zecheng Zhang, Xinwei He, Rok Sosic, Jure Leskovec}, title =...

http://snap.stanford.edu/graphsage/ has the info about the datasets. For more on PPI, see #78

That could be true. We are relying on the assumption that there is only a small probability where random sampling negatives will sample a positive example, when the entire graph...

The code for https://arxiv.org/pdf/1806.01973.pdf is not open-source due to corporate constraints. But the changes from GraphSAGE are described in paper. In the case of negative sampling, just need to implement...

So in the training time, we also want to remove a subset of edges. This is not for test/validation. This is just so that the model does not know existence...

Yes usually concatenating the node features with a one hot of whether the node is user or item would work.

Yes it'd be a dynamic model. At training time, you take in consecutive time step t and t+1. The node labels are just the usage at t+1, and inputs are...

The GraphSAGE json files are just numpy arrays of features, adjacency etc. So you don't have to go through the intermediate networkx format. Simple read the text file of SNAP...

Is there anything other than the input format section in README needed? what details and which format you are not sure about?

0 layers of message passing would be feature-only baseline right? E.g. for node classification, you simply need to learn an MLP that takes in node features and outputs the class.