Research map / Adversarial robustness and security / Attacks on text and graphs
Revisiting Item Promotion in GNN-Based Collaborative Filtering: A Masked Targeted Topological Attack Perspective
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use masked targeted gradient topological attack instead of the strongest baseline, RUFilter | LightGCN · Gowalla Q10, Δ₁₀¹ | PHN@50: 41.4 vs 9.1 | Helped |
| Use masked targeted gradient topological attack instead of the strongest baseline, RUFilter | LightGCN · Gowalla Q10, Δ₁₀² | PHN@50: 117.6 vs 15.6 | Helped |
| Use masked targeted gradient topological attack instead of the strongest baseline, RUFilter | LightGCN · Gowalla Q30, Δ₃₀¹ and Δ₃₀² | PHN@50: 39.1 vs 14.2; 106.3 vs 25.7 | Helped |
| Use masked targeted gradient topological attack instead of the strongest baseline, RUFilter | LightGCN · Gowalla Q50, Δ₅₀¹ and Δ₅₀² | PHN@50: 28.9 vs 17.1; 87.7 vs 29.0 | Helped |
| Use masked targeted gradient topological attack instead of the strongest baseline, RUFilter | LightGCN · Yelp2018 Q10, Δ₁₀¹ and Δ₁₀² | PHN@50: 60.6 vs 7.4; 140.8 vs 19.1 | Helped |
| Use masked targeted gradient topological attack instead of the strongest baseline, RUFilter | LightGCN · Yelp2018 Q30, Δ₃₀¹ and Δ₃₀² | PHN@50: 32.6 vs 4.8; 106.7 vs 10.5 | Helped |
| Use masked targeted gradient topological attack instead of the strongest baseline, RUFilter | LightGCN · Yelp2018 Q50, Δ₅₀¹ and Δ₅₀² | PHN@50: 20.7 vs 7.6; 75.0 vs 16.5 | Helped |
| Add node masking by excluding users below the target-item score threshold (ablation) | Masked targeted topological attack · Gowalla and Yelp2018 Q10 | PHN@50 | Helped |
| Use the proposed attack rather than the strongest baseline in retraining scenes | LightGCN retrained on perturbed graphs · Gowalla and Yelp2018, Q10/Q30/Q50, Δ¹ | PHN@50: about 1.4 × to 4.6 × | Helped |
| Restrict gradient computation to the target item's interaction column | Masked targeted attack · Large user-item graphs | memory consumption: from S (M × N) to S (M) | Helped |
| Apply the proposed perturbations to the graph | LightGCN recommendation performance · Gowalla and Yelp2018 perturbed graphs | NDCG | No clear effect |
Closest papers by meaning
- Trust-GRS: A Trustworthy Training Framework for Graph Neural Network Based Recommender Systems Against Shilling Attacks (AAAI 2025 · similarity 0.87)
- Can Graph Neural Networks Expose Training Data Properties? An Efficient Risk Assessment Approach (NeurIPS 2024 · similarity 0.85)
- 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)
- Highly Imperceptible Black-Box Graph Injection Attacks with Reinforcement Learning (AAAI 2025 · similarity 0.84)
- Value at Adversarial Risk: A Graph Defense Strategy against Cost-Aware Attacks (AAAI 2024 · similarity 0.84)
- Practicable Black-Box Evasion Attacks on Link Prediction in Dynamic Graphs—a Graph Sequential Embedding Method (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.