A cross-site comparison of online review manipulation using Benford’s law

There is a growing concern that online reviews are targets of systematic manipulation, and manipulated reviews serve purposes other than informing consumers. In this article, we report a cross-site comparison of the aggregate-level manipulation using Benford’s law to detect anomalies. Benford’s law states that digits in naturally occurring data follow a logarithmic distribution. Deviation from such distribution is considered as a sign of systematic manipulation. We empirically examine word-count distributions of reviews on a Chinese food delivery service platform (FDS), Dianping, Yelp, and Amazon. Our empirical analysis suggests, in general, word counts of online review contents do not obey Benford’s law, although Benford’s law holds among high-quality reviews. Deviation from Benford’s law is larger in emerging markets compared with mature online marketplaces. Further analyses reveal that positive reviews, especially positive and extreme reviews, exhibit more deviation from Benford’s law.