analiza aiqbench
REFLEX introduces a self-refining paradigm that disentangles factual content from deceptive stylistic patterns in LLM-generated explanations, directly addressing the reliability issues of current fact-checking systems. By leveraging self-disagreement signals to construct steering vectors, the method achieves state-of-the-art performance with minimal fine-tuning samples while significantly reducing hallucinations and latency. This approach offers a practical pathway for deploying robust, real-time automated fact-checking tools that provide faithful rationales without relying heavily on external knowledge sources.
REFLEX: Self-Refining Explainable Fact-Checking via Verdict-Anchored Style Control
arXiv:2511.20233v4 Announce Type: replace Abstract: The prevalence of fake news on social media demands automated fact-checking systems to provide accurate verdicts with faithful explanations. However, existing large language model (LLM)-based approaches ignore deceptive misinformation styles in LLM-generated explanations, resulting in unfaithful rationales that can mislead human judgments. They rely heavily on external knowledge sources, introducing hallucinations and even high latency that undermine reliability and responsiveness, which is crucial for real-time use. To address these challenges, we propose REason-guided Fact-checking with Latent EXplanations (REFLEX), a self-refining paradigm that explicitly controls reasoning style anchored on verdict. REFLEX utilizes self-disagreement veracity signals between the backbone model and its fine-tuned variant to construct steering vectors, naturally disentangling fact from style. Experiments on the real-world dataset show REFLEX achieves state-of-the-art performance under LLaMA-series models with only 465 self-refined samples. Moreover, owing to its transferability, REFLEX yields up to a 7.54% gain on in-the-wild data. Our results further demonstrate that our method effectively mitigates faithful hallucination, thereby guiding the model toward more accurate verdicts than previous works in explainable fact-checking.
Abstract:The prevalence of fake news on social media demands automated fact-checking systems to provide accurate verdicts with faithful explanations. However, existing large language model (LLM)-based approaches ignore deceptive misinformation styles in LLM-generated explanations, resulting in unfaithful rationales that can mislead human judgments. They rely heavily on external knowledge sources, introducing hallucinations and even high latency that undermine reliability and responsiveness, which is crucial for real-time use. To address these challenges, we propose REason-guided Fact-checking with Latent EXplanations (REFLEX), a self-refining paradigm that explicitly controls reasoning style anchored on verdict. REFLEX utilizes self-disagreement veracity signals between the backbone model and its fine-tuned variant to construct steering vectors, naturally disentangling fact from style. Experiments on the real-world dataset show REFLEX achieves state-of-the-art performance under LLaMA-series models with only 465 self-refined samples. Moreover, owing to its transferability, REFLEX yields up to a 7.54% gain on in-the-wild data. Our results further demonstrate that our method effectively mitigates faithful hallucination, thereby guiding the model toward more accurate verdicts than previous works in explainable fact-checking.
Computation and Language (cs.CL)
arXiv:2511.20233 [cs.CL]
(or arXiv:2511.20233v5 [cs.CL] for this version)
Text integral sub licența deschisă a publicației · 181 cuvinte