Open PeterDeSOM opened 1 year ago
Thanks for your attention. It's possible to conduct graph-level clustering. The general pipeline is as follows. 1) Conduct representation learning on graphs via self-supervised techniques, e.g., contrastive learning. 2) Apply the readout function (pooling function) on the learned node embeddings to obtain the graph embeddings. 3) Conduct clustering on the learned graph embeddings. (Maybe borrow some methods in this repository). I hope these suggestions help you.
I have millions of graph data that cannot be labeled by human efforts. I need it to be classified by unsupervised technique, not semi and not the node level. Is it possible?