In the company’s first ever transparency report, the review platform revealed how it weeds out fraudulent or malicious reviews.
Consumer review site Trustpilot has revealed in its first ever transparency report that 2.2m fake reviews were removed from its platform in 2020.
The Copenhagen-based company is reportedly preparing for an IPO this year and is on its way to reach a $1bn valuation, which would make it the latest European tech company to achieve unicorn status.
In its transparency report, Trustpilot said the types of reviews it considered to be fake included businesses leaving reviews on their own profiles, a review that has been paid for, a review that contains harmful or illegal content and a review that is not based on a genuine experience.
“Fake reviews are identified through a number of sources; through our automated software, through our flagging tools, and by our dedicated fraud analysts,” the report said.
Trustpilot’s automated tools removed the majority of the 2.2m fake reviews in 2020, while content reviewers manually removed almost 30pc.
How fake reviews are detected
According to Trustpilot, its fraud analysts use a variety of tools to detect fake reviews, including an anomaly detection model.
“This model analyses behavioural patterns using data points such as how reviews are submitted and collected by businesses over time and response rates by consumers. Any changes in the behavioural patterns we recognise are highlighted, and then used to carry out further investigations,” the report said.
“This technology is being developed into an early warning system, flagging suspected breaches of our guidelines to help our fraud analysts and investigators take proactive and swift action.”
Trustpilot said it also uses a dedicated data analysis tool to identify clusters of suspected fake reviews. This uses natural language processing, machine learning classifiers and cluster detection techniques to analyse large numbers of reviews and their associated behaviours across the platform and the internet more broadly.
The company also outlined the steps it takes to stop users posting fake reviews, including issuing warnings and formal notices and ending its relationship with businesses for breaching guidelines.
Complaints about Trustpilot
While the platform hosts reviews for other companies, products and services, Trustpilot itself is also reviewed on its site. The company addressed two key themes that have cropped up in its low reviews regarding how it treats flagged reviews and its response times.
“Previously, when a review was flagged by a business, our policy was always to temporarily move the review offline while we investigated it,” the report said. “Many consumers have a perception that this process censors their voice or, worse still, that Trustpilot is deleting their comments unfairly.”
The company said that as of December 2020, all flagged reviews will remain online while they are being investigated, except for content flagged as ‘harmful or illegal’.
Trustpilot also said in terms of response times, it was able to meet its year-end target of responding to all reported reviews within 48 hours of when they were first flagged.
Online consumer reviews and star ratings have become a common feature across a number of industries. But the practice has come under fire in recent years, particularly in the case of Uber where user ratings can affect drivers.
French ride-hailing app Heetch recently removed driver ratings from its service, with its chief executive Teddy Pellerin saying that the company doesn’t want to reduce people to a grade. “The drivers and passengers cannot be reduced to a number of stars.”