FlashGAN: Framework of Localized Node Augmentation via Semi-supervised Learning in Heterogeneous Graphs with Generative Adversarial Network

FlashGAN is a node generation framework designed to address the issue of node class imbalance, aiming to prevent Graph Neural Network (GNN) models trained on such imbalanced graph data from exhibiting a bias toward predicting the majority class in node classification tasks.

Building upon our previous work, FincGAN, FlashGAN not only generates synthetic nodes but also determines their placement within the heterogeneous graph during the generation process. This design substantially reduces the need for an additional step of constructing edges for synthetic nodes, which was required in prior approaches that focused solely on node generation.