Link Prediction on bipartite graph

[Vidhan]

I have a heterogenous graph with two sets of nodes skills and external_skills and only one edge type is_ext_aligned_to. Each skill is aligned to only one external_skill in the given dataset and there are no edges between skill and skill or external_skill and external_skill. Both have text embeddings as features. I want to train a GNN to predict links between new skill and all available external skills. I have some doubt related to my use case.

1. Is using GNN makes sense in my usecase ?

Lets say its feasible. I was able to train a GNN following this tutorial. My model converged. Now I want to make predictions. Lets say I want to predict links between skill1 and all avaliable external_skills. So for that what should be input graph ?

[mufeili]

I don’t think GNNs will be very useful in your case as a node is connected to at most a node. Without rich structural information, GNNs will perform no better than MLPs.