This research introduces LRQ-FACT, a novel framework that uses LLMs to generate relevant fact-checking questions, significantly improving the accuracy of automated verification processes.

• LLMs can effectively formulate targeted fact-checking questions (FCQs) when properly prompted
• The framework boosts fact-checking performance by 10.6% compared to methods without FCQs
• Combining textual and visual analysis through multimodal processing yields superior results to single-modality approaches
• Human evaluations confirm LLM-generated questions are comparable to human-crafted ones

For security applications, this research represents a crucial advancement in scalable detection of misinformation across multiple modalities, reducing the dependency on human fact-checkers while maintaining high verification standards.

Can LLMs Improve Multimodal Fact-Checking by Asking Relevant Questions?