Multimodal Disinformation Detection with Joint Propagation Structure

Abstract

The spread of disinformation can easily bring serious consequences for society. This means the detection of disinformation cannot be ignored. Current research on multimodal disinformation detection tends to ignore the influence of propagation structure. Therefore, this paper proposes a multimodal disinformation detection method based on text, images and the propagation structure, which learns new text structure features from a heterogeneous graphical model based on global-local relationships and then splices the new text structure features with traditional text auxiliary features. Finally, the picture features are combined with the text features obtained after the splicing to obtain the multimodal joint representation. The experimental results show that our model has higher accuracy and stronger generalization ability compared with the related multimodal models on the microblogging dataset.