CELL: A Causal Perspective for Fairness-aware Graph Adaptation

This paper studies fairness-aware graph adaptation, aiming to transfer knowledge from a labeled source graph to an unlabeled target graph while addressing fairness. Most prior methods require target-domain attributes to learn invariant graph representations of sensitive attributes, which are often unavailable in practice. To address this limitation, we introduce \underline{C}ausality-attended Repres\underline{e}ntation Dientang\underline{l}ement with Structural A\underline{l}ignment (CELL) for fairness-aware graph adaptation without requiring target sensitive labels. CELL constructs a causal graph to model the graph-generation mechanism and guide fair representation disentanglement. Specifically, CELL uses sensitive and causal encoders to extract sensitive and causal factors, respectively, and promotes disentanglement by minimizing their conditional mutual information. To leverage unlabeled target data, we further generate pseudo-labels for both target task labels and sensitive attributes, and use similarity relations to derive unbiased node representations. Finally, to further mitigate domain shift, we build a fairness-aware bipartite graph that provides additional structural supervision for cross-domain alignment. Experiments on benchmarks show that CELL consistently outperforms strong baselines in both predictive performance and fairness.

Many AI systems make predictions using network-like data, such as social networks, financial relationships, or links between people and organizations. These systems can unintentionally treat different groups unfairly, especially when the data used in a new setting differs from the data used for training. A major practical challenge is that information about sensitive groups, such as age, gender, or race, may be unavailable in the new setting because of privacy concerns or incomplete records. This paper proposes CELL, a method for making fairer predictions on a new network by learning from another network where both outcome labels and sensitive-group information are available. CELL separates information that is useful for the prediction task from information that may reflect sensitive-group bias. It then uses the unlabeled new network to carefully estimate useful training signals and aligns the two networks in a way that discourages biased relationships from being transferred. Experiments on real-world and synthetic datasets show that CELL can improve fairness while maintaining strong prediction accuracy, making it useful for applications such as credit risk assessment, social network analysis, and decision-support systems where fairness matters.