On Generalized Degree Fairness in Graph Neural Networks

Conventional graph neural networks (GNNs) are often confronted with fairness issues that may stem from their input, including node attributes and neighbors surrounding a node. While several recent approaches have been proposed to eliminate the bias rooted in sensitive attributes, they ignore other key input of GNNs, namely node, which can introduce since GNNs hinge on neighborhood structures generate representations. In particular, varying across nodes, manifesting themselves drastically different degrees, give rise diverse behaviors nodes biased outcomes. this paper, we first define generalize degree using generalized definition as manifestation quantification multi-hop around nodes. To address context classification, propose novel GNN framework called Generalized Degree Fairness-centric Graph Neural Network (DegFairGNN). Specifically, each layer, employ learnable debiasing function contexts, modulate layer-wise aggregation originating degrees among Extensive experiments three benchmark datasets demonstrate effectiveness our model both accuracy metrics.