How platforms can deal with fraudulent reviews
The power of online reviews over the commercial success of a business is well known. Indeed, all the way back in 2012, research from the University of California, Berkeley, showed that a half-point improvement in ratings across a 5-point scale was enough for a restaurant to be much more likely to sell out their tables during peak hours.
Later research, from Aalto University, reveals that around 40% of our decision making is based upon the views of others, and they believe that a good review is capable of boosting sales by up to 30%. It's perhaps no surprise, therefore, that the market for writing fake reviews is big business. Indeed, according to research from BrightLocal, fake reviews are so prevalent, that 82% of consumers have read one in the last year, with this proportion even higher among 18-34 year olds.
Given both the huge importance of online reviews to the success of online platforms, and the huge incentive to manipulate the results by criminals, there is understandably much being done to try and tackle the problem. For instance, researchers from the University of Chicago worked on a machine-learning based system that had been trained on millions of real reviews to enable it to create its own fake reviews.
The system started strongly, but after a while, the reviews were easy to spot as fake by the volunteer assessors.
The team from Aalto went a bit further with their own neural machine translation system, as their approach helped the system understand the context in which the review was made. It used a text sequence to dissect a review, and was thus able to create more believable reviews than the Chicago team. In true white hat style, they also deployed the same approach to better spot fake reviews.
The importance of these projects was underlined by a recent study from Carnegie Mellon, which looked at how consumers typically respond to any fake reviews they encounter online. One would imagine that the presence of fake reviews would undermine the credibility of the platform, but that wasn’t what the researchers found.
The researchers found that up to 30% of online reviews are fraudulent, but there remains no real consensus across the industry as to quite what to do with them. Some platforms attempt to delete them and target the firms that post them, while others try to flag the reviews as fake.
The researchers developed an experimental restaurant review portal with a reservation system and a behavior tracking system that was able to monitor time on site, clicks, and the pages visited by each person, while also tracking the restaurants ultimately chosen by the potential diner.
The results suggest that users actually found the platform most trustworthy when the fraudulent reviews were left in alongside the authentic reviews, with trust actually falling when attempts were made to remove them. The researchers note that the buying behavior of consumers was affected by original uncertainty about the evaluation of the restaurant’s quality.
In other words, when consumers were unsure about a restaurant, they tended to treat even the fraudulent reviews as a useful source of additional information.
Despite this, the study found that the consumers didn’t appear to be unduly influenced by the actual content of the fake reviews themselves. However, they did appear somewhat ill-equipped to distinguish between various types of fraudulent reviews, whether that was designed to boost up a restaurant's reputation or pull it down. The authors believe this indicates that platforms would be best served by deploying an approach whereby the motivational differences between positive and negative fake reviews are taken into account when helping consumers make the right decision.
One possible reason for this might be because we tend to apportion more weight to those reviews from influential people than we do for random strangers on the internet. While this orthodoxy is broadly true, research from Indiana University reminds us that even here, things are seldom straightforward.
The study found that influential reviewers were powerful for new products without much of a track record, but were far less influential for more established brands. What's more, influential reviewers tended to generate more in the way of word of mouth than they did actual sales, with the style of their reviews playing a part in this.
“Our analysis demonstrates that top-ranked reviewers write reviews that are longer, more formal and with more punctuation but that are less social and less effective,” the authors write.
Of course, these longer reviews are also harder for fraudsters to fake, either because the humans they use to write them lack the expertise to do so, or the technology they use to automate the task is not up to the job. One useful strategy against the fake reviews, therefore, is to encourage, and amplify, the submission of detailed reviews from authority figures.
The importance of online reviews seems unlikely to diminish in the coming years, and as such, it’s equally likely that attempts to manipulate them will continue to have a robust trade. While it may seem logical to try and remove all of the fake reviews encountered on a platform, the reality seems somewhat more nuanced. It’s a battle that will rumble on for some time to come.