Signed Graph Unlearning

TL;DR

SGU addresses unlearning in signed graphs where edge polarity and balance complicate partition-based methods. It introduces a bottom-up signed-graph reconstruction via quadratic optimization to extract $k$-OCGs using $FOCG$, followed by sign-aware agglomerative clustering that selects partitions by balancing edge structure and polarity. The method is evaluated on four real-world signed graphs with three SGNN backbones, outperforming unsigned-graph unlearning baselines in model utility (Macro F1) and achieving strong unlearning efficiency, as shown by faster partitioning and reduced membership-inference risk ($MIA$). This work demonstrates practical impact by enabling privacy-preserving updates in signed networks while maintaining predictive performance.

Abstract

The proliferation of signed networks in contemporary social media platforms necessitates robust privacy-preserving mechanisms. Graph unlearning, which aims to eliminate the influence of specific data points from trained models without full retraining, becomes particularly critical in these scenarios where user interactions are sensitive and dynamic. Existing graph unlearning methodologies are exclusively designed for unsigned networks and fail to account for the unique structural properties of signed graphs. Their naive application to signed networks neglects edge sign information, leading to structural imbalance across subgraphs and consequently degrading both model performance and unlearning efficiency. This paper proposes SGU (Signed Graph Unlearning), a graph unlearning framework specifically for signed networks. SGU incorporates a new graph unlearning partition paradigm and a novel signed network partition algorithm that preserve edge sign information during partitioning and ensure structural balance across partitions. Compared with baselines, SGU achieves state-of-the-art results in both model performance and unlearning efficiency.

Figures (3)

• Figure 1: Partitioning results on the Bitcoin-OTC dataset using GraphEraser. The top figure shows the ratio of positive and negative edges within each partition, while the bottom figure presents the number of nodes and edges in each partition.
• Figure 2: The four undirected signed triangle types according to balance theory. The two on the left, with an even number of negative edges, are balanced; the two on the right, with an odd number, are unbalanced.
• Figure 3: The Overall Framework of SGU. The Signed Structure Extraction module uses a quadratic optimization to identify balanced signed structures, ensuring the preservation of sign information and structural coherence. The Signed-Aware Balanced Clustering module generates edge-balanced and structurally complete partitions which serve as disjoint shards for training.