Research map / Adversarial robustness and security / Attacks on text and graphs
BOCLOAK: Optimal Transport-Guided Adversarial Attacks on Graph Neural Network-Based Bot Detection
Kunal Mukherjee, Zulfikar Alom, Tran Gia Bao Ngo, Cuneyt Gurcan Akcora, Murat Kantarcioglu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| enforce social-domain constraints (no human follow-back, directionality, temporal rules) on generic graph attacks | Nettack/FGA/PR-BCD/GOttack · Cresci-15, GAT, budget 1 | misclassification rate (%): 95.45 -> 8.17 (Nettack) | Hurt |
| use OT-guided cloak templates (BOCLOAK) instead of generic attacks under constraints | bot detectors · TwiBot-22, BotRGCN, constrained, budget 1 | misclassification rate (%): 9.33 -> 86.67 (vs best Nettack) | Improved |
| add OT-margin BCE loss to geometry training (ablation) | BOCLOAK · Cresci-17, BotRGCN, budget 5 | misclassification rate (%): 38.21 -> 99.67 | Improved |
| add plausibility loss (ablation) | BOCLOAK · Cresci-17, BotRGCN, budget 5 | misclassification rate (%): 57.99 -> 99.67 | Improved |
| add degree-based neighbor weighting (ablation) | BOCLOAK · Cresci-17, BotRGCN, budget 5 | misclassification rate (%): 73.23 -> 99.67 | Improved |
| add age-based neighbor weighting (ablation) | BOCLOAK · Cresci-17, BotRGCN, budget 5 | misclassification rate (%): 87.23 -> 99.67 | Improved |
| add sparsity loss (ablation) | BOCLOAK · Cresci-17, BotRGCN, budget 5 | misclassification rate (%): 96.88 -> 99.67 | Improved |
| apply adversarial defenses (GNNGuard, GRAND, RobustGCN) | BOCLOAK attack · Cresci-15 and TwiBot-22 | misclassification rate (%): Cresci-15 BotRGCN: 99.34 vanilla vs 98.91 GNNGuard | No clear effect |
| apply GNNGuard defense | BOCLOAK attack · BotSim-24, BotRGCN, budget 1 | misclassification rate (%): 58.22 -> 52.68 | Hurt |
| add NoisyGNN hidden-state noise defense | constrained generic attacks vs BOCLOAK · TwiBot-22, BotRGCN | misclassification rate (%): generic attacks below 6%; BOCLOAK 85.12% | Mixed |
| increase edge budget from 1 to 5 | constrained SOTA attacks · Cresci-15 | misclassification rate (%): 3-9% -> 12-23% | Improved |
| very small entropic regularization epsilon (0.01) (ablation) | BOCLOAK · sensitivity analysis | flip rate (%): about 83-85% | Hurt |
Closest papers by meaning
- Value at Adversarial Risk: A Graph Defense Strategy against Cost-Aware Attacks (AAAI 2024 · similarity 0.85)
- Highly Imperceptible Black-Box Graph Injection Attacks with Reinforcement Learning (AAAI 2025 · similarity 0.85)
- Exploring and Mitigating Adversarial Manipulation of Voting-Based Leaderboards (ICML 2025 · similarity 0.84)
- Enhancing Robustness of Graph Neural Networks on Social Media with Explainable Inverse Reinforcement Learning (NeurIPS 2024 · similarity 0.84)
- Unveiling the Threat of Fraud Gangs to Graph Neural Networks: Multi-Target Graph Injection Attacks Against GNN-Based Fraud Detectors (AAAI 2025 · similarity 0.84)
- Grimm: A Plug-and-Play Perturbation Rectifier for Graph Neural Networks Defending Against Poisoning Attacks (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
