Tag
This paper introduces Transductive Sharpening (TS), a loss-level modification for semi-supervised node classification that minimizes prediction entropy on unlabeled nodes while counterbalancing on labeled nodes, achieving consistent performance improvements without architectural changes.