Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification

In three linesNew approach for node classification in partially labeled graphs. Authors propose Transductive Sharpening (TS), a loss-level modification that minimizes prediction entropy on unlabeled nodes while counterbalancing effects on labeled nodes. Consistent improvements across multiple benchmarks without architectural changes.

Summary generated by Claude — human-verified