They could bring in some common sense measures like "what is the ratio of complaints to total number of sales" and "did the seller get the chance to rectify the situation"
I’d expect that they are doing that. This story only quotes a few reviews, but we don’t know how many more customers complained through Amazon’s customer service system (customers are probably more likely to do a return than write a public review). Amazon likely has some threshold for when to initiate action against a seller based on this data.