Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evading Community Detection via Counterfactual Neighborhood Search

Andrea Bernini, Fabrizio Silvestri, Gabriele Tolomei

TL;DR

This work tackles privacy risks from community detection by introducing community membership hiding, where a target node strategically rewires its neighborhood to avoid detection by a given algorithm $f(\cdot)$ under a budget. It formulates the problem as a constrained counterfactual graph objective and solves it with a deep reinforcement learning approach (A2C) that operates on a graph neural encoder within an MDP, learning feasible add/del edge actions for the target node. The key contributions include formal problem definition, a DRL-based solution with a GNN encoder, and validation across five real-world datasets showing superior balance between hiding success (SR) and preservation of the original graph structure (NMI), with demonstrated transferability to unseen detection algorithms. The findings have practical implications for privacy-preserving options on social platforms and highlight considerations for ethical use and potential misuse, motivating further work on larger-scale graphs and multi-node hiding scenarios.

Abstract

Community detection techniques are useful for social media platforms to discover tightly connected groups of users who share common interests. However, this functionality often comes at the expense of potentially exposing individuals to privacy breaches by inadvertently revealing their tastes or preferences. Therefore, some users may wish to preserve their anonymity and opt out of community detection for various reasons, such as affiliation with political or religious organizations, without leaving the platform. In this study, we address the challenge of community membership hiding, which involves strategically altering the structural properties of a network graph to prevent one or more nodes from being identified by a given community detection algorithm. We tackle this problem by formulating it as a constrained counterfactual graph objective, and we solve it via deep reinforcement learning. Extensive experiments demonstrate that our method outperforms existing baselines, striking the best balance between accuracy and cost.

Evading Community Detection via Counterfactual Neighborhood Search

TL;DR

This work tackles privacy risks from community detection by introducing community membership hiding, where a target node strategically rewires its neighborhood to avoid detection by a given algorithm under a budget. It formulates the problem as a constrained counterfactual graph objective and solves it with a deep reinforcement learning approach (A2C) that operates on a graph neural encoder within an MDP, learning feasible add/del edge actions for the target node. The key contributions include formal problem definition, a DRL-based solution with a GNN encoder, and validation across five real-world datasets showing superior balance between hiding success (SR) and preservation of the original graph structure (NMI), with demonstrated transferability to unseen detection algorithms. The findings have practical implications for privacy-preserving options on social platforms and highlight considerations for ethical use and potential misuse, motivating further work on larger-scale graphs and multi-node hiding scenarios.

Abstract

Community detection techniques are useful for social media platforms to discover tightly connected groups of users who share common interests. However, this functionality often comes at the expense of potentially exposing individuals to privacy breaches by inadvertently revealing their tastes or preferences. Therefore, some users may wish to preserve their anonymity and opt out of community detection for various reasons, such as affiliation with political or religious organizations, without leaving the platform. In this study, we address the challenge of community membership hiding, which involves strategically altering the structural properties of a network graph to prevent one or more nodes from being identified by a given community detection algorithm. We tackle this problem by formulating it as a constrained counterfactual graph objective, and we solve it via deep reinforcement learning. Extensive experiments demonstrate that our method outperforms existing baselines, striking the best balance between accuracy and cost.
Paper Structure (18 sections, 13 equations, 4 figures, 4 tables)

This paper contains 18 sections, 13 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Background and Preliminaries
  4. Community Membership Hiding
  5. Problem Formulation
  6. Counterfactual Graph Objective
  7. Markov Decision Process
  8. Proposed Method
  9. Experiments
  10. Experimental Setup
  11. Community Membership Hiding Task
  12. Results and Discussion
  13. Parameter Sensitivity
  14. Computational Complexity
  15. Discussion
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Given a graph $\mathcal{G}$, a node $u$ (in this case $u=D$), a budget of actions $\beta$, and the set of communities identified by the community detection algorithm $f(\cdot)$ (including the community $\mathcal{C}_i$ to which the node belongs), community membership hiding consists of adding inter-community edges $\mathcal{E}_{u,i}^+$ (green edges), or removing intra-community edges $\mathcal{E}_{u,i}^-$ (red edge), so that the value returned by the similarity function $sim(\cdot, \cdot)$, between the new community to which the node belongs after rewiring, and the original one, is lower than the $\tau$ constraint.
  • Figure 2: $\mathbf{F1}$ score of SR and NMI in symmetric setting (training: greedy; testing: greedy; $\boldsymbol{\tau=0.5}$; $\boldsymbol{\beta=1\mu}$).
  • Figure 3: $\mathbf{F1}$ score of SR and NMI in asymmetric setting (training: greedy; testing: louvain; $\boldsymbol{\tau=0.5}$; $\boldsymbol{\beta=1\mu}$).
  • Figure 4: $\mathbf{F1}$ score of SR and NMI in asymmetric setting (training: greedy; testing: walktrap; $\boldsymbol{\tau=0.5}$; $\boldsymbol{\beta=1\mu}$).