"DIFRM: A Domain-Independent Feature Reconstruction Method for Cross-Domain Multimodal Fake News Detection"

"MCFEND[25] is a comprehensive cross-domain benchmark dataset specif-ically designed for the detection of Chinese fake news. It contains 23,789 news articles from diverse sources, including social media platforms, messaging ap-plications, and traditional online news outlets. The dataset is divided into three groups, each representing different domain types. Group 1 consists of 7,959 news articles from 9 Chinese fact-checking organizations. Group 2 includes 6,702 articles verified by 4 English-speaking fact-checking organiza-tions, such as PolitiFact and FakeNewsCorpus. Group 3 encompasses 9,128 news articles sourced from the Weibo-21 dataset, collected from the Weibo platform.  These datasets differ in data sources, content styles, and annota-tion protocols, enabling comprehensive evaluation of the model\u2019s generaliza-tion capability under domain shifts."