Community-Centric Graph Unlearning

TL;DR

Graph unlearning is essential for privacy in GNN-based systems. The authors introduce the Graph Structure Mapping Unlearning (GSMU) paradigm and its practical instantiation, the Community-centric Graph Eraser (CGE), which maps subgraphs to mapped nodes to form a reduced, connected graph and enables deterministic, node-level unlearning with far fewer parameters and training data. Across four real-world datasets and three GNN backbones, CGE outperforms BP-SM-TA baselines on Macro F1 while delivering substantially faster unlearning and strong resistance to membership inference attacks, evidencing effective structure preservation through community-aware mapping. This work delivers a scalable, privacy-preserving framework for graph learning by integrating mapped graph representations with modular community detection and node-level update mechanisms, offering a practical path toward deployment in sensitive domains.

Abstract

Graph unlearning technology has become increasingly important since the advent of the `right to be forgotten' and the growing concerns about the privacy and security of artificial intelligence. Graph unlearning aims to quickly eliminate the effects of specific data on graph neural networks (GNNs). However, most existing deterministic graph unlearning frameworks follow a balanced partition-submodel training-aggregation paradigm, resulting in a lack of structural information between subgraph neighborhoods and redundant unlearning parameter calculations. To address this issue, we propose a novel Graph Structure Mapping Unlearning paradigm (GSMU) and a novel method based on it named Community-centric Graph Eraser (CGE). CGE maps community subgraphs to nodes, thereby enabling the reconstruction of a node-level unlearning operation within a reduced mapped graph. CGE makes the exponential reduction of both the amount of training data and the number of unlearning parameters. Extensive experiments conducted on five real-world datasets and three widely used GNN backbones have verified the high performance and efficiency of our CGE method, highlighting its potential in the field of graph unlearning.

Figures (5)

• Figure 1: Community-centric Graph Eraser (CGE) framework. CGE maps the communities in original graph $\mathcal{G}$ to obtain graph $\mathcal{\widetilde{G}}$ by graph structure mapping. Only the related mapped nodes need to be updated when nodes require unlearning.
• Figure 2: Quality and utility evaluation of different graph partitioning schemes.
• Figure 3: Time consumption of each method under different unlearning node ratios.
• Figure 4: Conductance evaluation of different graph partitioning schemes.
• Figure 5: The comparison of Macro F1 scores of different community detection methods on four datasets and three GNN backbones.

Theorems & Definitions (1)

• Definition 1