Leveraging LLM-Generated Explanations for Detecting Emotionally Rewritten Fake News
Quick Answer
This study introduces a Gated Cross Attention (GCA) framework for detecting emotionally rewritten fake news, enhancing robustness against emotional variations.
Quick Take
Experiments on PolitiFact and LUN show significant improvements, while maintaining competitive performance on GossipCop, indicating the effectiveness of explanation guidance in detection models.
Key Points
- GCA framework integrates emotionally rewritten news with stable explanations for better detection.
- Significant performance improvements observed on PolitiFact and LUN under various emotional conditions.
- Competitive results maintained on GossipCop, showcasing model robustness.
- Study emphasizes the impact of emotional reframing on fake news detection.
- Code and data are publicly available for further research.
DeepSignal Analysis
What happened
A new framework called Gated Cross Attention (GCA) has been developed to detect emotionally rewritten fake news. This method focuses on integrating explanations from original articles to enhance detection robustness against emotional variations. Experiments conducted on datasets like PolitiFact and LUN show notable improvements in detection performance.
Key evidence
- The GCA framework integrates emotionally rewritten news with corresponding explanations to improve detection accuracy.
- Experiments on the PolitiFact and LUN datasets demonstrated significant improvements in detection under various emotional conditions.
- The GCA method maintained competitive performance on the GossipCop dataset, indicating its effectiveness across different benchmarks.
Why it matters
The ability to detect fake news that has been emotionally rewritten is crucial for maintaining information integrity. As fake news evolves, traditional detection methods may falter. This study's approach, which leverages explanations, could provide a more robust solution, potentially reducing the spread of misinformation and its societal impacts.
What to watch
Paper Resources
📖 Reader Mode
~2 min readAbstract:The spread of fake news may cause severe social consequences. Existing fake news detection methods mainly focus on stylistic variations or incorporate external information such as explanations. However, news articles are often rewritten under different emotional backgrounds while preserving their underlying factual claims, which may affect the robustness of detection models. In this work, we investigate fake news detec- tion under fact-preserving emotional variations. To study this problem, we construct emotion-rewritten test sets and generate explanations from the original news articles as stable background knowledge. We then propose a Gated Cross Attention (GCA) framework that adaptively integrates emotionally rewritten news with the corresponding explanations, enabling the model to focus on informative explanation content while reducing potential mismatches caused by emotional reframing. Experiments on PolitiFact, GossipCop, and LUN demonstrate that the proposed method achieves notable improvements under multiple emotional conditions on PolitiFact and LUN, while maintaining competitive performance on GossipCop. We further analyze the effects of explanation guidance and gating mechanisms under different emotional conditions. Our code and data are available at: this https URL gca .
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.08835 [cs.CL] |
| (or arXiv:2610.08835v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.08835 arXiv-issued DOI via DataCite |
Submission history
From: Zekun Yang [view email]
[v1]
Tue, 29 Sep 2026 11:50:30 UTC (852 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →TriAgent: Divergence-Aware Committees for Cost-Efficient Financial Sentiment Analysis
TriAgent introduces a cost-efficient multi-agent system for financial sentiment analysis, combining VADER, FinBERT, and Qwen2.5. It achieves an F1 score of ~0.87 with significant savings of $9.3M/year at a 10M-user scale compared to GPT-4o-mini, while also detecting hallucinations with an AUC of 0.90.