Comprehensive Out-of-context Misinformation Detection via Global Information Enhancement

False information that spreads rapidly on the Internet may have a negative impact on individuals, organizations, and society.Cheapfake, which refers to out-of-context misinformation, involves reusing the pairing of images and captions to support false descriptions.Out-of-context misinformation detection requires consideration of the global semantic understanding of images and intertwined contextual information.Currently, the baseline method don't extract enough global information.This paper uses visual-semantic reasoning to extract the global features of images and texts, and proposes a new detection framework that employs fake statement checking and multimodal natural language inference to determine whether the input triples are cheapfake through multiple steps.The accuracy of our method is 3.4% higher than that of the baseline method, and it also has its own advantages compared with other improved methods.