Research interest

Few-label node classification

How far do a few labels go when the graph does most of the work?

Experiments · Aalborg University

Code
Draft. Mostly placeholder: add your project's own question and results.
Labels help most when communities are clear. Label propagation on 3-block SBM graphs with 150 nodes; mean ± s.d. over 20 graphs. With strong communities one label per class gives 0.86 accuracy; with weak communities ten labels per class reach only 0.68.

The question

In many graphs only a handful of nodes carry labels. Semi-supervised methods spread those labels along edges, so how well they work depends on how strongly the graph's structure matches the classes.

What I'm doing

[Your project: methods, datasets and the question you're testing.]