Research map / Adversarial robustness and security / Attacks on text and graphs
Synthetic Disinformation Attacks on Automated Fact Verification Systems
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 |
|---|---|---|---|
| Add Grover-generated evidence documents | FEVER claim verifiers · FEVER dev REF+NEI | accuracy: 73.05 -> 28.59; 70.76 -> 29.02; 75.92 -> 51.86 | Hurt |
| Limit contamination to one retrieved adversarial sentence (ablation) | FEVER claim verifiers · FEVER dev REF+NEI | accuracy: 73.05 -> 34.80; 70.76 -> 34.08; 75.92 -> 60.93 | Hurt |
| Use only generated documents as evidence (ablation) | FEVER claim verifiers · FEVER dev REF+NEI | accuracy: 73.05 -> 21.18; 70.76 -> 23.43; 75.92 -> 29.05 | Hurt |
| Add Grover-generated evidence documents | VeriSci, SciKGAT, and ParagraphJoint · SciFact dev | abstract label: 51.03 -> 25.57; 58.04 -> 37.46; 55.52 -> 41.31 | Hurt |
| Add Grover-generated evidence documents | CovidFact baseline · CovidFact test REF | label accuracy: 83.8% -> 69.4% | Hurt |
| Insert claim paraphrases into existing evidence documents | FEVER claim verifiers · FEVER dev REF+NEI | accuracy: 73.05 -> 32.62; 70.76 -> 37.22; 75.92 -> 52.72 | Hurt |
| Replace claim keywords, entities, or numbers in injected text instead of paraphrasing (ablation) | FEVER claim verifiers · FEVER dev REF+NEI | accuracy: 73.05 -> 53.83; 70.76 -> 42.82; 75.92 -> 60.93 | Hurt |
| Use paraphrased disinformation rather than key-replacement perturbations (ablation) | FEVER claim verifiers · FEVER dev REF+NEI | accuracy: CorefBERT 53.83 -> 32.62; KGAT 42.82 -> 37.22; MLA 60.93 -> 52.72 | Mixed |
| Replace poisoned evidence with random FEVER sentences (ablation) | KGAT · FEVER dev REF claims | accuracy drop: 30.05% to 6.63% | No clear effect |
| Replace poisoned evidence with random FEVER sentences (ablation) | KGAT · FEVER dev NEI claims | accuracy drop: 53.42% to 7.73% | No clear effect |
| Increase evidence sentences from 5 to 10 with one adversarial sentence (ablation) | FEVER REF claims · FEVER dev REF | performance drop: 23.87% ... to 23.69% | No clear effect |
| Increase evidence sentences from 5 to 10 with one adversarial sentence (ablation) | FEVER NEI claims · FEVER dev NEI | performance drop: 49.49% ... to 50.81% | Mixed |
Closest papers by meaning
- DECEIVE-AFC: Adversarial Claim Attacks against Search-Enabled LLM-based Fact-Checking Systems (NeurIPS 2026 · similarity 0.88)
- BiDeV: Bilateral Defusing Verification for Complex Claim Fact-Checking (AAAI 2025 · similarity 0.82)
- Contradicted in Reliable, Replicated in Unreliable: Dual-Source Reference for Fake News Early Detection (AAAI 2025 · similarity 0.82)
- Can LLM-Generated Misinformation Be Detected? (ICLR 2024 · similarity 0.82)
- DEFAME: Dynamic Evidence-based FAct-checking with Multimodal Experts (ICML 2025 · similarity 0.82)
- Beyond Detection: Exploring Evidence-based Multi-Agent Debate for Misinformation Intervention and Persuasion (AAAI 2026 · similarity 0.82)
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.