Semi-supervised document clustering of research papers on arXiv. I struggled heavily in the first year of my PhD just learning the meta-game to staying on top of my field (computer security + computer architecture).

It turns out semi-supervised document recommender systems aren't easy to bootstrap with zero user data.