Skip to main content

Emerging Fake Review Strategies and Information Cascades


The linked article describes the efforts of five researchers at the University of Chicago to identify and examine the characteristics of an emerging type of fake review generation rooted in the use of neural nets. Compared to more conventional crowd sourced fake reviews which rely on paying a group of people to leave reviews that hype a certain company’s products or services, neural net generated reviews have the potential to accomplish the same effect while being harder to spot and significantly cheaper to operate. The paper found that fake reviews generated by a neural net are often overlooked by “state-of-the-art statistical detectors” and that people often mistake these fabricated reviews for real reviews. The researchers even state that the fake reviews scored high on “usefulness” when shown to human users. The researchers then discuss the efficacy of several methods of counteracting these neural net reviews including a conventional method involving the language analysis that their neural net often fools. However, it is important to note that in many cases websites can avoid having to figure out which reviews are fake by requiring a “captcha” like verification in order to post in the first place, which potentially explains why crowd-sourced fake review production is the current method of choice rather than neural nets, as the crowd-sourced fake reviews are able to overcome these preventative measures.

The findings presented in the paper have many profound implications. Foremost, it points out that the abatement of fake reviews online will likely continue to be an arms race of sorts where improving technology and techniques on both sides will keep the fight against fake reviews an ongoing issue as methods of generating fake reviews become cheaper and more accessible to online sellers. However, it may not be just doom and gloom on the horizon for online shoppers, as allowing some fake reviews may actually be beneficial. As mentioned in the blog post here online reviews for items on websites like amazon are often self-perpetuating once they clear a certain threshold number of reviews required to earn the trust of an average customer. Introducing fake reviews into the mix could make finding popular products harder for the user, potentially resulting in the user locating a better product than he would have purchased without fake reviews, ultimately helping to prevent information cascades from popularizing low quality products. However, this relies on several assumptions, biggest of which are that the product information presented in an item’s description is accurate, that the popular product isn’t the best product for at least some customers, and that the additional time the customer has to spend searching for a product doesn’t detract from the net value of the product. In light of these heavy-handed assumptions, it is likely that a fake-free review system is the best system currently available despite its tendency for creating information cascades.



Leave a Reply

Blogging Calendar

November 2017