In addition to a sibling comment about median, it could be that or that other common method (I forget, bayesion average? I think), but also consider - 176K reviews, 76 days since the _expose_, if it were 100 reviews a day on average, that's still 7.6K reviews. If it were 10K in the first few days, then 7.6K after, it's still ~10% of total reviews, where they likely had a ~4.6-8 score before hand. A drop in the bucket.
And it's possible (though imo unlikely) that some reviews were removed, perhaps initially at least, due to suspected botting.
Median has little difference to mean when it is a bounded variable .
Rating can only vary 1-5, so mean and median will always be pretty close . For variables like say wealth that skews the average with outliers is where median is more useful than mean
And it's possible (though imo unlikely) that some reviews were removed, perhaps initially at least, due to suspected botting.