Online reviews can make or break a business. Unfortunately, unscrupulous individuals frequently cheat by submitting fraudulent online reviews that either increase or lessen a business’ reputation, regardless of whether the individuals have actually used the product or service or have an opinion. These reviews are sometimes quite hard to catch and weed out, but researchers at State University of New York, Stony Brook, have found that there’s a way to catch these reviews by using mathematics.
Math on the case
What happens when the U.S. Federal Trade Commission can’t spot the fraud? The organization has fine power that can discourage fraudulent online review activity, but when it can’t tell the difference, the damage continues to propagate. Researchers at State University of New York, Stony Brook believe that the answer lay in science. By looking at how fraudulent online reviews distort the statistical distribution of a retailer’s online review scores, mathematic “forensic analysis” points out not individual fake reviews, but trends that can lead toward eventual detection.
Results and recommendations from the new Stony Brook study were recently presented at the International Conference on Weblogs and Social Media in Dublin, Ireland.
Breaking the J-curve
Head Stony Brook researcher Yejin Choi pointed out that by recognizing when trends occur, statistical analysis can help businesses zero in on the origin area of the fraudulent review
“(The technique) is able to pinpoint where the densities of false reviews are for any given hotel,” said Choi, an assistant professor of computer science.
In general terms, researchers found that online review scores for any product or service, when plotted on a graph, tend to take a letter J shape. According to associate professor of information management systems Dr. Paul Pavou of Temple University, the J-curve distribution is caused by the tendency consumers have to buy things they like, and in turn like what they buy. Often, if a purchase meets expectations, the buyer is not as likely to write a review than if the product or service experience was extremely bad or extremely good.
Fraudulent online reviews distort the J-curve. Stony Brook researchers began to study this by selecting reviewers who had written at least 10 reviews, more than a day or two apart, with an average rating that didn’t appear to stray much from the overall average rating. These ratings were then compared with those of single-time reviewers. When large discrepancies were found to exist, Choi noted that the group would then compare the ratio of positive to negative, with an eye toward sudden bursts of given reviews that could have been placed as a part of a company’s marketing campaign (or competition attack).
Follow the footprints
Ultimately, by checking and cross-checking the data, Choi found that fraudulent online reviews could be caught as much as 72 percent of the time. That’s not perfect, but it is significant, said Pavou.
“It’s really unlikely some random strategy would achieve 72 percent accuracy,” she said.
Choi sees these results as a footprint that makes it much easier for online fraud to be traced to the source.
“(Fake reviewers) might think that it was a perfect crime, but the truth is, they distorted the shape of the review scores of their own hotels, and that leaves a footprint of the deceptive activity, and the more they do it, the stronger it becomes,” said Choi.