In classifying news as fake, the intention of its creator/distributor is peripheral; only its truth value and the presumed effect it has on the addressee are relevant. Therefore, FAKEROM defines fake news as “a (sub)genre of journalistic/informative discourse that conveys false information presented as true, with the purpose of eliciting a certain type of action in a certain community”.
True news and fake news are not separated by a mere difference of degree (= the gradual fallacy), but by deep structural differences. At the same time, news should not be evaluated as “true” or “false” exclusively (= the binary fallacy), since there is also a third type of category bridging these two extremes, that of “imaginary” news. For example, satirical news is not fake news, but imaginary news.
Based on its previous findings, the FAKEROM project advances a classification of news into three main categories (true news – fake news – imaginary news), which can be divided, in turn, into six subcategories (real news – authentic news; propaganda news – fabricated news; satirical news – fictional news).
Within fake news, it is important to dissociate between fabricated news (= misleading by addition/substitution) and propaganda news (= misleading by omission). However, the difference between them is only formal, since from a functional point of view, propaganda news can be as harmful as fabricated news.
Our analyzes show that, in order to be more convincing, fake news imitates not only the content of true news, but also its argumentative structure, amplifying the rhetorical features that typically characterize genuine news. Compared to true news describing similar phenomena, fake news is generally longer, syntactically more complex, more rhetorical (featuring a greater number of special punctuation marks), more detailed (with more vivid adverbs and adjectives), and more “nuanced” (with more modal verbs and adverbs). Therefore, paradoxically, the excess of sophistication should be a warning sign to readers about the increased probability that a particular news story is fake, and not a guarantee of its authenticity.
By training the BERT models on the FAKEROM dataset, our automated analyzes managed to achieve an accuracy of over 80% in identifying fake news, as well as its subcategories (fabricated news and propaganda news).
In general, fake news spreads virally on social media through the process of sharing. Our position in this respect is that not only the creators, but also the simple distributors of fake news have a moral responsibility regarding the act of possibly misinforming others.
