In a different world, I'd suggest it's the political line. I.e. that a dedicated ML space is in some ways a greater threat than a left-unity space. But there are three problems with that idea.

One, hexbear seems to be fairly consistently on the same page as lemmygrad e.g. LGBTQ+ rights, Ukraine, Palestine, and the need to abolish capitalism.

Two, the authors of the paper in question haven't displayed any indication that they understand or could understand the difference.

Three, it's a link aggregator. We've had some successful hauntings but otherwise we aren't much of a threat at all. As said in another comment, we don't even like to discuss 'local' issues to avoid doxing, nevermind organising on here. It shows the liberal mentality, to see the mere existence of contrary views as such a threat. Our simple presence, here, stating that Nazis, genocide, and transphobia are bad is considered to be as extreme as the far right organising in plain sight to run over protesters protesting the things we agreed are bad.

I'll accept the former

I think it might just be because Hexbear is "difficult to understand". If you pick random high score comments from there without context and show it to a data-driven nerd, I doubt they'd even be able to describe what's going on.

But since federation, at some point any (naïve) analysis of either forum will inevitably include a lot of posts from the other. This could be an actually interesting area for social network research (social dynamics between "native" and "foreign" users in an instance), but since there's no ready-made model for this in scikit-learn the Data-Driven bros will never risk actually doing research.