Fake news detectors improve with multi-style worst-case training
MANTIS: Multi-Axis News Transformations for Instance-Wise Stress Testing and Worst-Case Training in Fake News Detection
Computation and Language
Summary
Fake news on social media can be tricky to detect because people rewrite it in many ways to avoid being caught. The paper introduces a new way to test fake news detectors by changing news stories across several styles, emotions, and framing methods. The authors also created a system called MANTIS that adapts training specifically to each news article, helping detectors focus on the facts rather than superficial writing styles. Their tests show this method makes detectors stronger against different rewriting tricks while still accurately spotting real news.
What this means in practice
- •For social media platforms: Enhance fake news filters with adaptive training that resists diverse rewriting tactics used by misinformation spreaders.
- •For online content moderation teams: Use multi-style manipulated news examples to test and improve the robustness of automated fake news detection tools.
Authors
Dexin Liu, Fan Yang, Zimu Wang, Xunfei Zhu, Bolin Ren, Bingyan Zhang, Wei Wang, Qi Chen
Abstract
Automated fake news detection is vital on social media, where fake news is easily rewritten to fool detectors. However, existing approaches typically rely on superficial writing cues or single-style text rewrites, resulting in poor robustness under diverse cosmetic modifications. We construct MANTIS-Bench, a multi-axis presentation-attack benchmark that pairs original news with rewritten views across writing style, emotional expression, and dissemination framing. We further propose MANTIS, an instance-adaptive adversarial framework enabling dynamic worst-case training tailored to each news article with zero inference-time overhead. This reframes fake news defense from heuristic format smoothing into targeted, fact-invariant credibility assessment. Experiments on MANTIS-Bench show that MANTIS reduces the Union Attack Success Rate (UASR) by 3.55 percentage points relative to SheepDog on GossipCop and consistently outperforms prior single-axis defenses on Macro-F1 while maintaining competitive clean accuracy, demonstrating the effectiveness of multi-axis and instance-adaptive worst-case training.