GNN Link Prediction (GraphSAGE/PyG) - Validation AUC Consistently Below 0.5 Despite Overfitting Control


Would you please share yoru neg sampling function?

I am having a similar issue where my AUC hovers around 0.6. My graph is so low in density (among all possible links, only ~10% actual exists), so I feel the neg sampling would be the bottleneck. I believe your solution can inspire me a lot. Thanks in advance!

Posted by Hugh LIU (1) at 2025-07-21T06:40:11Z

Caught for

Contains ?
Low Length
No Code Block
Contains Salutation - Thanks in Advance
Low Rep