Scholarlink

<p>Fake news increasingly spans multiple languages and modalities, especially on short-form video platforms, making detection more complex. This paper introduces the first study addressing multilingual fake news detection for content that incorporates multiple languages within a single news item. This study utilizes the Design Science Research (DSR) paradigm and presents an automated large language model (LLM)-based fact-checking approach for multilingual fake news detection that requires no extensive task-specific training data. Our method includes: (1) LLM-based cross-modal consistency filtering to identify discrepancies across modalities; and (2) LLM-based fact verification implemented through a multi-agent system. We evaluated our approach using an existing multilingual text fake news dataset and a novel multilingual short-form video fake news dataset (MSVFD). Experimental results show our approach outperforms leading supervised and LLM-based prompting methods. This research addresses a critical gap in multilingual fake news literature and contributes to DSR on LLM-based artifacts, highlighting their potential for timely misinformation detection across diverse formats and languages.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An LLM-based Fact-Checking Approach for Multilingual Fake News Detection in Short-Form Video News

  • 期刊论文

  • 发表日期：

    2026年7月22日

查看全文
  • Xu Lian,
  • Jinsoo Park,
  • Jihae Suh

摘要

Fake news increasingly spans multiple languages and modalities, especially on short-form video platforms, making detection more complex. This paper introduces the first study addressing multilingual fake news detection for content that incorporates multiple languages within a single news item. This study utilizes the Design Science Research (DSR) paradigm and presents an automated large language model (LLM)-based fact-checking approach for multilingual fake news detection that requires no extensive task-specific training data. Our method includes: (1) LLM-based cross-modal consistency filtering to identify discrepancies across modalities; and (2) LLM-based fact verification implemented through a multi-agent system. We evaluated our approach using an existing multilingual text fake news dataset and a novel multilingual short-form video fake news dataset (MSVFD). Experimental results show our approach outperforms leading supervised and LLM-based prompting methods. This research addresses a critical gap in multilingual fake news literature and contributes to DSR on LLM-based artifacts, highlighting their potential for timely misinformation detection across diverse formats and languages.