data_loader DBLP about dgfraud

[YingtongDou]

The preprocessed DBLP data has already added self-loops for each subgraph.

In our code, we process the adjacency matrix using our function which can be found in utils.py.

So we delete the existing self-loops to unify the data i/o format.

[bobo18801737494]

Thanks very much!
I have another question about train_test_split of load_data_dblp, split the adjacency matrix, may be lead to nodes and realationship broken?

[YingtongDou]

The train_test_split only selects the nodes whose labels are used in the training phase or testing phase. It does not split the adj matrix.

During training, we leverage the relations and neighbors of the nodes in the training set to learn its embedding and predict its class.