Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

IGA-LWP: An Iterative Gradient-based Adversarial Attack for Link Weight Prediction

Cunlai Pu, Xingyu Gao, Jinbi Liang, Jianhui Guo, Xiangbo Shu, Yongxiang Xia, Rajput Ramiz Sharafat

TL;DR

IGA-LWP targets robustness of link weight prediction in weighted graphs by an iterative gradient-based perturbation framework guided by a self-attention graph auto-encoder (SEA) surrogate. It formalizes the attack as maximizing prediction error on a chosen set of target links under a perturbation budget and demonstrates strong effectiveness and cross-model transferability on four real networks. The results show substantial degradation in prediction accuracy and reveal fundamental vulnerabilities in weighted-network inference, motivating the development of robust and privacy-preserving link weight prediction methods. Overall, the work provides a principled framework for evaluating and potentially defending against adversarial perturbations in weighted graph analysis.

Abstract

Link weight prediction extends classical link prediction by estimating the strength of interactions rather than merely their existence, and it underpins a wide range of applications such as traffic engineering, social recommendation, and scientific collaboration analysis. However, the robustness of link weight prediction against adversarial perturbations remains largely unexplored.In this paper, we formalize the link weight prediction attack problem as an optimization task that aims to maximize the prediction error on a set of target links by adversarially manipulating the weight values of a limited number of links. Based on this formulation, we propose an iterative gradient-based attack framework for link weight prediction, termed IGA-LWP. By employing a self-attention-enhanced graph autoencoder as a surrogate predictor, IGA-LWP leverages backpropagated gradients to iteratively identify and perturb a small subset of links. Extensive experiments on four real-world weighted networks demonstrate that IGA-LWP significantly degrades prediction accuracy on target links compared with baseline methods. Moreover, the adversarial networks generated by IGA-LWP exhibit strong transferability across several representative link weight prediction models. These findings expose a fundamental vulnerability in weighted network inference and highlight the need for developing robust link weight prediction methods.

IGA-LWP: An Iterative Gradient-based Adversarial Attack for Link Weight Prediction

TL;DR

IGA-LWP targets robustness of link weight prediction in weighted graphs by an iterative gradient-based perturbation framework guided by a self-attention graph auto-encoder (SEA) surrogate. It formalizes the attack as maximizing prediction error on a chosen set of target links under a perturbation budget and demonstrates strong effectiveness and cross-model transferability on four real networks. The results show substantial degradation in prediction accuracy and reveal fundamental vulnerabilities in weighted-network inference, motivating the development of robust and privacy-preserving link weight prediction methods. Overall, the work provides a principled framework for evaluating and potentially defending against adversarial perturbations in weighted graph analysis.

Abstract

Link weight prediction extends classical link prediction by estimating the strength of interactions rather than merely their existence, and it underpins a wide range of applications such as traffic engineering, social recommendation, and scientific collaboration analysis. However, the robustness of link weight prediction against adversarial perturbations remains largely unexplored.In this paper, we formalize the link weight prediction attack problem as an optimization task that aims to maximize the prediction error on a set of target links by adversarially manipulating the weight values of a limited number of links. Based on this formulation, we propose an iterative gradient-based attack framework for link weight prediction, termed IGA-LWP. By employing a self-attention-enhanced graph autoencoder as a surrogate predictor, IGA-LWP leverages backpropagated gradients to iteratively identify and perturb a small subset of links. Extensive experiments on four real-world weighted networks demonstrate that IGA-LWP significantly degrades prediction accuracy on target links compared with baseline methods. Moreover, the adversarial networks generated by IGA-LWP exhibit strong transferability across several representative link weight prediction models. These findings expose a fundamental vulnerability in weighted network inference and highlight the need for developing robust link weight prediction methods.
Paper Structure (19 sections, 12 equations, 3 figures, 3 tables, 2 algorithms)

This paper contains 19 sections, 12 equations, 3 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Problem formulation
  3. Method
  4. Link weight prediction model
  5. Gradient extraction for target links
  6. Iterative generation of adversarial graphs
  7. Global and local attacks
  8. Transferability of adversarial attacks
  9. Performance evaluation
  10. Datasets
  11. Evaluation metrics
  12. Pearson Correlation Coefficient (PCC)
  13. Root Mean Squared Error (RMSE)
  14. Baseline attack methods
  15. Experimental setup
  16. ...and 4 more sections

Figures (3)

  • Figure 1: Framework of the IGA-LWP model for adversarial link weight prediction. This model includes SEA-based gradient derivation and gradient sorting to select key links for perturbations. Perturbations are then superimposed to construct an adversarial network, which leads to erroneous link weight predictions.
  • Figure 2: RMSE vs. perturbation proportion for different attack methods on various datasets.
  • Figure 3: RMSE of the link weight prediction methods (Deepwalk, Node2vec, and GCN) on adversarial graphs generated by different attack methods.