Online product reviews of everything from goods (for example, books) to services (for example, plumbing) are popular among consumers. However, people who are paid to post reviews of a particular product can manipulate reviews. These are considered fraudulent or manipulative reviews. Hu et al. examine the landscape of this phenomenon in an effective and useful manner. They present several different ways to detect fraudulent reviews, the types of manipulation that such reviews employ, and the various effects that manipulation has on the consumers reading the reviews. Their findings are the result of careful analysis and statistical examination, which they report in detail. The methods do not necessarily identify fraudulent reviews, but instead find the probability that some reviews are fraudulent.
One of the ways the authors detect the probability of manipulative reviews is to assess writing styles. They assume that if all of the reviews were written by different individuals, the writing style data will have a random distribution. This reveals one weakness of their analysis, because they also assume that most or all of the manipulative reviews of a product written by the same person will occur in a relatively short period of time. If so, more than one review written by the same person will be discovered by this analysis. However, if the fraudulent reviews are spread out over a longer period of time, this method might not detect them.
This paper is easy to read and has valuable information. I recommend it for any organization that uses consumer reviews.