graph-distillation

Proposes STAD, a semi-supervised framework for distilling text-attributed graphs using Wasserstein distance, dual-pathway encoders, and LLM-based text synthesis to achieve a state-of-the-art performance-compression trade-off.