Like Article, Like Audience: Enforcing Multimodal Correlations for Disinformation Detection

User-generated content (e.g., tweets and profile descriptions) and shared\ncontent between users (e.g., news articles) reflect a user's online identity.\nThis paper investigates whether correlations between user-generated and\nuser-shared content can be leveraged for detecting disinformation in online\nnews articles. We develop a multimodal learning algorithm for disinformation\ndetection. The latent representations of news articles and user-generated\ncontent allow that during training the model is guided by the profile of users\nwho prefer content similar to the news article that is evaluated, and this\neffect is reinforced if that content is shared among different users. By only\nleveraging user information during model optimization, the model does not rely\non user profiling when predicting an article's veracity. The algorithm is\nsuccessfully applied to three widely used neural classifiers, and results are\nobtained on different datasets. Visualization techniques show that the proposed\nmodel learns feature representations of unseen news articles that better\ndiscriminate between fake and real news texts.\n