My knee-jerk reaction is that they are detecting anomalies and low-quality reviews for some definition of quality. As you pointed out, without a consistent gro

My knee-jerk reaction is that they are detecting anomalies and low-quality reviews for some definition of quality. As you pointed out, without a consistent ground truth target how could it really separate "fake" from "legit"? Perhaps they partnered with a fake review business and used their records?