Online reviews have become an increasingly important part of how we pick everything from hotels to electronics, but what's to stop an ambitious company from gaming the system with fake, glowing reviews? Math, at least for now. Researchers have come up with a statistical technique for finding false reviews, noting how fake ratings skew the natural J-shaped curve that real reviews produce, reports Technology Review.
The J shape comes because most reviews tend to cluster around excellent or terrible—plenty of one-star reviews, then few twos, threes, and fours, but a lot of fives. The researchers compared hotel reviews made by people who had made more than 10 reviews with those who had made just one. If the one-time reviews clustered differently than the multiple reviewers, they were labeled suspicious. Sudden clusters of reviews were flagged, too, as possible marketing campaigns. By comparing their results with earlier studies, researchers said they found fraudulent reviews 72% of the time.