Super interesting. I wonder if the metamodel of the differences between reliable and unreliable could be related to whether the overall lexicon (and implied ont

Super interesting. I wonder if the metamodel of the differences between reliable and unreliable could be related to whether the overall lexicon (and implied ontology) of a given article is more focused on the endogenous logic of the subject and the events it describes, or relies on "callouts" to exogenous ideas. ( for exmaple, this comment is a callout, instead of a discussion of their methodology) It would be a kind of subtext analysis, where the question is whether the article has more of an internal logical consistency, or if it resembles a form that is a vehicle for an external idea. The concept worth exploring might not be "reliability," but "value." Sorting news into tribes is pretty simple and doesn't really require ML, so their examples of unreliable news sites are useful to prove the apparatus tech for a surveillance scheme, but I'd suggest less scientifically meaningful. Their sample sites were more about social class than reliability (the same thing to them, probably), as there are definitely news sources that are both extremely unreliable and high value. That unreliable/high-value quadrant is probably more useful to study for determining how to discover value than just numbering people for their class and beliefs. Instead of a sentiment analysis, the question is, "is this article about a subject, or is the subject of this article a vehicle for another idea?" Wonder if that's possible. It would be like determining sentiment to separate reporting from criticism.