How to evaluate link prediction model

I followed Stochastic Training of GNN for Link Prediction tutorial to build my link prediction model, however, this tutorial evaluates model performance on node classification task. I am wondering how to evaluate my model performance on link prediction task.

There are generally two approaches depending on your use cases: one that cares of AUC and another that cares of ranking. Either way, for each positive edge you will have a set of negative edges to compare against.

Taking AUC as an example, typically you will have a set of test positive edges and a set of test negative edges, and you evaluate it like a normal binary classification model.

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.