Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploiting Meta-Learning-based Poisoning Attacks for Graph Link Prediction

Mingchen Li, Di Zhuang, Keyu Chen, Dumindu Samaraweera, Morris Chang

TL;DR

This paper addresses the vulnerability of graph link prediction to adversarial attacks by proposing an unweighted graph poisoning strategy that leverages meta-learning with weighted scheme guidance. The approach trains a surrogate model with meta-gradients to identify impactful edge modifications, applying them within a budget to distort learned representations and degrade performance. Empirical results on Cora, Twitch, and MovieLens show the method outperforms baselines, with linear-based weighting delivering the strongest attacks while maintaining covert graph changes. The work highlights significant practical implications for the robustness of link-prediction in GNNs and provides publicly available code for reproducibility.

Abstract

Link prediction in graph data uses various algorithms and Graph Nerual Network (GNN) models to predict potential relationships between graph nodes. These techniques have found widespread use in numerous real-world applications, including recommendation systems, community/social networks, and biological structures. However, recent research has highlighted the vulnerability of GNN models to adversarial attacks, such as poisoning and evasion attacks. Addressing the vulnerability of GNN models is crucial to ensure stable and robust performance in GNN applications. Although many works have focused on enhancing the robustness of node classification on GNN models, the robustness of link prediction has received less attention. To bridge this gap, this article introduces an unweighted graph poisoning attack that leverages meta-learning with weighted scheme strategies to degrade the link prediction performance of GNNs. We conducted comprehensive experiments on diverse datasets across multiple link prediction applications to evaluate the proposed method and its parameters, comparing it with existing approaches under similar conditions. Our results demonstrate that our approach significantly reduces link prediction performance and consistently outperforms other state-of-the-art baselines.

Exploiting Meta-Learning-based Poisoning Attacks for Graph Link Prediction

TL;DR

This paper addresses the vulnerability of graph link prediction to adversarial attacks by proposing an unweighted graph poisoning strategy that leverages meta-learning with weighted scheme guidance. The approach trains a surrogate model with meta-gradients to identify impactful edge modifications, applying them within a budget to distort learned representations and degrade performance. Empirical results on Cora, Twitch, and MovieLens show the method outperforms baselines, with linear-based weighting delivering the strongest attacks while maintaining covert graph changes. The work highlights significant practical implications for the robustness of link-prediction in GNNs and provides publicly available code for reproducibility.

Abstract

Link prediction in graph data uses various algorithms and Graph Nerual Network (GNN) models to predict potential relationships between graph nodes. These techniques have found widespread use in numerous real-world applications, including recommendation systems, community/social networks, and biological structures. However, recent research has highlighted the vulnerability of GNN models to adversarial attacks, such as poisoning and evasion attacks. Addressing the vulnerability of GNN models is crucial to ensure stable and robust performance in GNN applications. Although many works have focused on enhancing the robustness of node classification on GNN models, the robustness of link prediction has received less attention. To bridge this gap, this article introduces an unweighted graph poisoning attack that leverages meta-learning with weighted scheme strategies to degrade the link prediction performance of GNNs. We conducted comprehensive experiments on diverse datasets across multiple link prediction applications to evaluate the proposed method and its parameters, comparing it with existing approaches under similar conditions. Our results demonstrate that our approach significantly reduces link prediction performance and consistently outperforms other state-of-the-art baselines.

Paper Structure

This paper contains 33 sections, 5 equations, 5 figures, 4 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. Gradient-Based Attacks
  4. Heuristic-Based Attacks
  5. Reinforcement Learning-Based Attacks
  6. Background
  7. Graph Adversarial Attack
  8. Graph Neural Networks
  9. Variational Graph Auto-Encoders (VGAE)
  10. Graph Attention Network (GAT)
  11. LightGCN
  12. Adversarial Attack with Meta-Learning
  13. Methodology
  14. Objective
  15. Attack Setting and Threat Model
  16. ...and 18 more sections

Figures (5)

  • Figure 1: Adversarial attacks based on the target stage of the ML pipeline.
  • Figure 2: Framework of the proposed approach.
  • Figure 3: Comparsion using ROC-AUC and NDCG@20 metrics
  • Figure 4: Comparsion using AP and Recall metrics
  • Figure 5: Visualization Examples of original graph and poisoned graph resulting from the proposed attack.