nanyuu / tua Goto Github PK

Hi Nanyuu!
I am interested in detailed points in your experiments using the benchmark dataset.
In your paper, you define the feature matrix as X \in {0, 1}^{N \times d}; however, the PubMed dataset seems to have continuous-valued features.
When experimenting with PubMed, in Algorithm 1, do the elements of the original X remain continuous values, and only the elements of X_{Fake} are {0, 1}?
If not, are you binarizing the PubMed features somewhere in the code?

hi Nanyuu, this work is very exciting. I've read your paper on it, but I am still confused about the node feature of the adversarial nodes. In ur paper algo.1 (page 4), a certain node feature of the selected node is set to 1 while you are performing experiments on citation datasets, in which the node feature should be the degree of this node. So is the feature of adversarial nodes still 1s and 0s while appending to the benign graphs or they're either scaled or appended to normalized graphs?