文献类型：期刊文献

英文题名：Fairness-oriented vertical federated GNNs with incomplete sensitive attributes

作者：Wang, Ziqi[1];Yan, Xueming[2,3];Jin, Yaochu[4]

机构：[1]East China Univ Sci & Technol, Sch Informat Sci & Engn, Shanghai 200237, Peoples R China;[2]Guangdong Univ Foreign Studies, Sch Informat Sci & Technol, Guangzhou 510006, Peoples R China;[3]Guangdong Engn Res Ctr Data Secur Governance & Pr, Guangzhou 510006, Peoples R China;[4]Westlake Univ, Sch Engn, Hangzhou 310030, Peoples R China

年份：2026

卷号：127

外文期刊名：INFORMATION FUSION

收录：;EI(收录号：20254519465425);WOS:【SCI-EXPANDED(收录号:WOS:001607027300001)】；

基金：This work was supported in part by the National Natural Science Foundation of China (62576116, 62136003) , the International Collaboration Fund for Creative Research Teams (ICFCRT) of NSFC (W2441019) , the Major Program of the Natural Science Foundation of Zhejiang Province (D25F020001) , the Young Researcher Startup Program of Guangdong-Hong Kong-Macau Center for Applied Mathematics (2025A1515060002) and the Guangdong Philosophy and Social Science Foundation Regular Project (GD25YGG26) .

语种：英文

外文关键词：Fairness; Graph neural network; Vertical federated learning; Incomplete sensitive attributes

摘要：Ensuring fairness in federated learning is crucial, especially in scenarios where sensitive attributes are incomplete or partially missing, as their absence can result in biased predictions. In vertical federated learning, graph neural networks are increasingly used to model distributed graph-structured data while preserving privacy. However, the fairness of graph neural networks in vertical federated learning is challenged by disjoint feature distributions and incomplete sensitive features, which can exacerbate bias in predictions. This paper proposes a fairness-oriented vertical federated graph neural network framework that utilizes sensitive attribute completion to achieve group fairness while adhering to privacy constraints. On the client side, we design a completion-driven adversarial fusion model that infers missing sensitive attributes, incorporates heterogeneous neighbor features, and extracts fair local feature representations through adversarial learning. On the server side, we develop a weighted aggregation method that balances accuracy and fairness to generate global features. Specifically, the server aggregates global similarity metrics and incorporates similarity loss at the client level to improve inter-client consistency. Experiments on public datasets demonstrate that the proposed method achieves an optimal balance between fairness and accuracy across various missing rates while effectively preserving user privacy.