This is very cool. It maps to my existing understanding of how knowledge actually works.

You often don't need to see the whole hyperbolic disc, only some region in the center, and there, the text would largely still be readable.

The arrows are drawn in hyperbolic space, but the text is not; it really should be. Then there will never be an overlap problem.

Alternatively, the center of the text (or generally the anchor) of the text box could still be oriented to the screen to seed the render orientation, just like it already is, but allow the rest to be drawn following the rules of geodesics in the hyperbolic space. though I don't know if that would work as well.