Research map / Adversarial robustness and security / Attacks on text and graphs
Robustness in Text-Attributed Graph Learning: Insights, Trade-offs, and New Defenses
Runlin Lei, Lu Yi, Mingguo He, Pengyu Qiu, Zhewei Wei, Yongchao Liu, Chuntao Hong
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 |
|---|---|---|---|
| text-structure trade-off: structure-oriented models (LLaGA, GCN) under structural vs textual attack | LLaGA/GCN · TAG benchmarks | robust accuracy | Mixed |
| text-oriented models (SFT-neighbor, GraphGPT) under textual poisoning | SFT-neighbor · CiteSeer | accuracy drop: 25% drop vs 5%-10% for most GNNs | Hurt |
| use advanced text encoders with GNNGuard | GNNGuard · structural evasion attacks | robust accuracy | Improved |
| add Guardual adaptive filtering to GNNGuard | GNNGuard · structural evasion | robust accuracy | Improved |
| add structural noise injection (-noise) to GraphLLMs (ablation) | GraphLLMs · structural evasion | robust accuracy | Improved |
| add textual noise injection (-noisetxt) to GraphLLMs (ablation) | GraphLLMs · textual evasion | robust accuracy | Improved |
| combine both noises (-noisefull) in GraphLLMs (ablation) | GraphLLMs · structural+textual evasion | robust accuracy | No clear effect |
| add GNNGuard-style edge similarity filtering (-simf) to GraphLLMs (ablation) | GraphLLMs · structural vs textual attacks | robust accuracy | Mixed |
| modify SFT prompt for adaptive reliance (-simp) (ablation) | GraphLLMs · TAG evasion attacks | robust accuracy | No clear effect |
| add SFT-auto detection and recovery to SFT-neighbor | SFT-neighbor · structural and textual attacks | robust accuracy | Improved |
| TextFooler attack with mismatched embedding (BoW or RoBERTa vs MiniLM) | GNNs · Cora | TextFooler attack effectiveness | Hurt |
| add validation to attack evaluation | GCN · Cora (HeuristicAttack) | accuracy: 61.30 -> 70.33 | Improved |
Closest papers by meaning
- Towards Robust Text-Attributed Federated Graph Learning: Multimodal Threats and Defense (AAAI 2026 · similarity 0.93)
- Intruding with Words: Towards Understanding Graph Injection Attacks at the Text Level (NeurIPS 2024 · similarity 0.91)
- Bounding the Expected Robustness of Graph Neural Networks Subject to Node Feature Attacks (ICLR 2024 · similarity 0.88)
- Generalization Principles for Inference over Text-Attributed Graphs with Large Language Models (ICML 2025 · similarity 0.87)
- Beyond the Golden Teacher: Enhancing Graph Learning through LLM-GNN Co-teaching (NeurIPS 2026 · similarity 0.87)
- SSTAG: Structure-Aware Self-Supervised Learning Method for Text-Attributed Graphs (NeurIPS 2025 · similarity 0.87)
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.
