The tone of the introduction is a bit preachy and the findings as stated are underwhelming and beyond obvious [1].
But when reading this type of papers -- and academic literature in general -- don't consider the application and research questions that were chosen by an academic and/or young graduate student with minimal resources. Look at the actual innovation. In this case, the methods. Then ask what else you can do with the increment of innovation.
It is now possible to methodologically analyze rhetorical patterns in open source communications on a shoestring budget. Because producers are also consumers, you can begin to understand at a very granular level which sequences of words elicit the desired effect in subsets of a population, and how. Work like this used to be laborious, expensive, and required a huge amount of socio-cultural training/expertise/judgement. By comparison, all of the technical work described in the paper has a relatively low barrier to entry.
Also: the obvious next step is to treat this as an optimization problem instead of a categorization problem.
--
[1] "othering intensifies during crises" is framed as a finding in this paper, but that's like using nukes to fish for trout. It's a fact we already know and have known for thousands of years. Therefore, the fact that the method reached this conclusion is best understood as validating a proposed method for automating intelligence collection and analysis in the context of open source war propaganda.