Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robustness Evaluation of Graph-based News Detection Using Network Structural Information

Xianghua Zeng, Hao Peng, Angsheng Li

TL;DR

This work tackles the robustness of graph-based fake news detectors by introducing SI2AF, a structural-information-guided adversarial framework that uses structural entropy to uncover hierarchical communities and an influence-based multi-agent system to coordinate targeted subgraph attacks. By modeling attacks as a cooperative multi-agent MDP and embedding manipulations back into the network, SI2AF both reveals vulnerabilities and trains detectors to become more robust against structural perturbations. The authors demonstrate that SI2AF substantially improves attack effectiveness (average gains of 16.71%) and robustness (average gains of 41.54%) across multiple detectors and real-world datasets, with detailed analyses on attack strategies, agent ablations, and parameter sensitivities. The framework provides a principled, scalable approach to robustness evaluation and offers practical insights for strengthening graph-based misinformation detection in dynamic social networks.

Abstract

Although Graph Neural Networks (GNNs) have shown promising potential in fake news detection, they remain highly vulnerable to adversarial manipulations within social networks. Existing methods primarily establish connections between malicious accounts and individual target news to investigate the vulnerability of graph-based detectors, while they neglect the structural relationships surrounding targets, limiting their effectiveness in robustness evaluation. In this work, we propose a novel Structural Information principles-guided Adversarial Attack Framework, namely SI2AF, which effectively challenges graph-based detectors and further probes their detection robustness. Specifically, structural entropy is introduced to quantify the dynamic uncertainty in social engagements and identify hierarchical communities that encompass all user accounts and news posts. An influence metric is presented to measure each account's probability of engaging in random interactions, facilitating the design of multiple agents that manage distinct malicious accounts. For each target news, three attack strategies are developed through multi-agent collaboration within the associated subgraph to optimize evasion against black-box detectors. By incorporating the adversarial manipulations generated by SI2AF, we enrich the original network structure and refine graph-based detectors to improve their robustness against adversarial attacks. Extensive evaluations demonstrate that SI2AF significantly outperforms state-of-the-art baselines in attack effectiveness with an average improvement of 16.71%, and enhances GNN-based detection robustness by 41.54% on average.

Robustness Evaluation of Graph-based News Detection Using Network Structural Information

TL;DR

This work tackles the robustness of graph-based fake news detectors by introducing SI2AF, a structural-information-guided adversarial framework that uses structural entropy to uncover hierarchical communities and an influence-based multi-agent system to coordinate targeted subgraph attacks. By modeling attacks as a cooperative multi-agent MDP and embedding manipulations back into the network, SI2AF both reveals vulnerabilities and trains detectors to become more robust against structural perturbations. The authors demonstrate that SI2AF substantially improves attack effectiveness (average gains of 16.71%) and robustness (average gains of 41.54%) across multiple detectors and real-world datasets, with detailed analyses on attack strategies, agent ablations, and parameter sensitivities. The framework provides a principled, scalable approach to robustness evaluation and offers practical insights for strengthening graph-based misinformation detection in dynamic social networks.

Abstract

Although Graph Neural Networks (GNNs) have shown promising potential in fake news detection, they remain highly vulnerable to adversarial manipulations within social networks. Existing methods primarily establish connections between malicious accounts and individual target news to investigate the vulnerability of graph-based detectors, while they neglect the structural relationships surrounding targets, limiting their effectiveness in robustness evaluation. In this work, we propose a novel Structural Information principles-guided Adversarial Attack Framework, namely SI2AF, which effectively challenges graph-based detectors and further probes their detection robustness. Specifically, structural entropy is introduced to quantify the dynamic uncertainty in social engagements and identify hierarchical communities that encompass all user accounts and news posts. An influence metric is presented to measure each account's probability of engaging in random interactions, facilitating the design of multiple agents that manage distinct malicious accounts. For each target news, three attack strategies are developed through multi-agent collaboration within the associated subgraph to optimize evasion against black-box detectors. By incorporating the adversarial manipulations generated by SI2AF, we enrich the original network structure and refine graph-based detectors to improve their robustness against adversarial attacks. Extensive evaluations demonstrate that SI2AF significantly outperforms state-of-the-art baselines in attack effectiveness with an average improvement of 16.71%, and enhances GNN-based detection robustness by 41.54% on average.

Paper Structure

This paper contains 31 sections, 1 theorem, 15 equations, 7 figures, 10 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Adversarial Attack on Graph Neural Networks
  4. Structural Information Principles
  5. Preliminaries
  6. GNN-based Fake News Detection
  7. Attacks against GNN-based News Detectors
  8. Structural Information Principles
  9. Methodology
  10. Hierarchical Structure Identification
  11. Multiple Agent Design
  12. Target Subgraph Attack
  13. Detection Optimization
  14. Experiments
  15. Experimental Settings
  16. ...and 16 more sections

Key Result

theorem 1

Let $x \in [1,\frac{b}{2}]$ be a positive random variable with a probability density function $q_0(x)$. Given the transformation $x^\prime = -\frac{x}{b} \cdot \left(\log_{2} \frac{c}{b} x\right)$, under the condition $0 < c \leq \frac{2}{e}$, the variable $x^\prime$ increases monotonically with the

Figures (7)

  • Figure 1: Comparative illustration between classical methods and our framework. SI2AF minimizes dynamic uncertainty in social engagements and identifies an associated subgraph to strategically establish connections with both target and non-target posts, resulting in significantly enhanced attack effectiveness.
  • Figure 2: Detailed design of our proposed SI2AF framework.
  • Figure 3: Average predictive probabilities of fake and real news before and after adversarial attacks.
  • Figure 4: Success rates of different attack strategies on fake news in the Gossipcop dataset.
  • Figure 5: Attack visualization of SI2AF and MARL against GNN-based fake news detectors.
  • ...and 2 more figures

Theorems & Definitions (1)

  • theorem 1