Doing so is Fakespot's raison d'etre, but an unimportant side gig for Amazon. I did my MSc thesis on classifying fake reviews (https://douglas-fraser.com/FakeReviews/index.html). It focused on purely text based features, but I am familiar with all of the literature and research at this point.

Non-text based features are more useful (classification based just on text is hard because believable fake reviews are easy to write) but the amount of non-text / behavioral data Amazon gets or would need to process would be a Facebook level of engineering effort - and is it really that cost-effective? (versus letting buyers do the work - humans can't tell if a review is fake from the text very well, but other things like an excessive number of 5* reviews is easy for a human to see)

The removal of these 20,000 reviews (the FT article) is based on a straightforward obvious heuristic and so is cost-effective. But trying to track all the IPs from China leaving reviews? hmmm. But they should have got the distance and banned all the companies associated with the fake reviews; that would definitely send a signal.

Amazon has tried things like Vine etc to deal with the problem; Yelp and TripAdvisor have their own methodologies. But bad actors always seem to find a way around it. Automating the monitoring of communities where spammer groups congregate is possible but that is another major engineering effort - you just can't have humans doing it due to the scale (and Bezos doesn't like paying essential warehouse workers a decent wage either... so why bother with something like this?)