How to Perform Negative Sampling for Edge Classification?

I’m working on a task involving multiclass classification for edge classification. How can I train my model by incorporating negative sampling?

Negative sampling is typically performed in link prediction, where node pairs with non-existent edges are sampled to help the model learn presence or absence of edges. Multi-class edge classification generally doesn’t require negative sampling.

You should look into dgl.graphbolt, especially the link prediction examples.