To train the algorithm, we focus on topic-agnostic linguistic & stylistic features. Our focus is on the so-called "language of #FakeNews". We build on extant theoretical & conceptual work by Alyt Damstra & colleagues to identify the potentially relevant features. 🙏

Our findings confirm that there actually is such a thing as the #LanguageOfDisinformation. Moreover, the study shows that the model is (largely) robust with regard to specific feature-groups but performs best when trained on a combination of the different feature-groups.