Relation embedding must be part of model for link prediction task?

Is it possible only use subject emb and object emb to do the link prediction task?

If possible, how?

If not possible, why?

Thank you.

1 Like

What do you mean by object embedding and subject embedding?

From https://docs.dgl.ai/tutorials/models/1_gnn/4_rgcn.html

“A knowledge graph is made up by a collection of triples of the form (subject, relation, object)”

Yes. You can see our GCMC example.https://github.com/dmlc/dgl/tree/v0.4.0/examples/mxnet/gcmc

Thank you very much.