Files
Abstract
As Graph Neural Networks (GNNs) find increasing applications, the fairness of these networks has emerged as a pressing issue. This thesis presents an innovative exploration into the domain of fairness in graph learning, where we introduce a novel problem concerning disparate data distributions in the training and test sets. Our experiments investigate the limitations of existing fairness graph learning methodologies, revealing their potential failure to mitigate bias when confronted with differing data distributions between the training and test sets. To address this challenge, we propose an innovative framework capable of managing such disparities, thus enhancing the fairness of outcomes. The experiments demonstrate that our method outperforms the state of art model regarding fairness metrics and maintains a comparable prediction accuracy.