Personalized Differential Privacy Graph Neural Network

TY - JOUR

T1 - Personalized Differential Privacy Graph Neural Network

AU - Yuan, Yanli

AU - Lei, Dian

AU - Zhang, Chuan

AU - Xiong, Zehui

AU - Li, Chunhai

AU - Zhu, Liehuang

N1 - Publisher Copyright: © 2026 Institute of Electrical and Electronics Engineers Inc.. All rights reserved.

PY - 2026

Y1 - 2026

N2 - Dear Editor, This letter addresses the critical challenge of preserving privacy in graph learning without compromising on data utility. Differential privacy (DP) is emerging as an effective method for privacy-preserving graph learning. However, its application often diminishes data utility, especially for nodes with fewer neighbors in graph neural networks (GNNs). Given that most real-world graph data follow a power-law distribution with a majority of low-degree nodes, we propose personalized differential privacy graph neural network (PDPGNN), a novel GNN training method. The novelty of PDPGNN lies in uniquely offering personalized differential privacy by allocating privacy budgets based on node degrees, effectively improving the data utility for nodes with fewer connections. Additionally, PDPGNN integrates a weighted aggregation mechanism to enhance model accuracy. Theoretical analysis shows that PDPGNN can achieve $\epsilon$-differential privacy for graph data, making a balance between privacy protection and data utility. Experimental results on four real-world graph datasets demonstrate the effectiveness of PDPGNN.

AB - Dear Editor, This letter addresses the critical challenge of preserving privacy in graph learning without compromising on data utility. Differential privacy (DP) is emerging as an effective method for privacy-preserving graph learning. However, its application often diminishes data utility, especially for nodes with fewer neighbors in graph neural networks (GNNs). Given that most real-world graph data follow a power-law distribution with a majority of low-degree nodes, we propose personalized differential privacy graph neural network (PDPGNN), a novel GNN training method. The novelty of PDPGNN lies in uniquely offering personalized differential privacy by allocating privacy budgets based on node degrees, effectively improving the data utility for nodes with fewer connections. Additionally, PDPGNN integrates a weighted aggregation mechanism to enhance model accuracy. Theoretical analysis shows that PDPGNN can achieve $\epsilon$-differential privacy for graph data, making a balance between privacy protection and data utility. Experimental results on four real-world graph datasets demonstrate the effectiveness of PDPGNN.

UR - https://www.scopus.com/pages/publications/105008185479

U2 - 10.1109/JAS.2025.125279

DO - 10.1109/JAS.2025.125279

M3 - Article

AN - SCOPUS:105008185479

SN - 2329-9266

VL - 13

SP - 498

EP - 500

JO - IEEE/CAA Journal of Automatica Sinica

JF - IEEE/CAA Journal of Automatica Sinica

IS - 2

ER -