Amazon has declared it’s utilizing increasingly sophisticated AI tools to combat counterfeit reviews and non-authentic comments on its marketplace.
The world’s leading eCommerce marketplace has wrestled with fake review brokers for many years, recently culminating in legal action against NiceRebate.com, a fake review broker targeting UK customers.
Dharmesh Mehta, head of Amazon’s customer trust team, stated, “We are aggressively fighting review brokers,” and revealed that Amazon had taken legal action against 94 such review fraudsters globally, including in the US, China, and Europe.
In 2022, Fakespot, a tool built to detect fake reviews, analyzed 720 million reviews and estimated that some 42% were counterfeit in some way.
Almost anyone who’s studied Amazon reviews has likely come across suspicious reviews that feel intuitively fake. Many reviews are paid endorsements aimed at boosting a seller’s ratings or sabotaging a rival company.
Amazon is investing in machine learning (ML) models capable of analyzing thousands of data points to help identify fraudulent review activity. The AI model evaluates various factors to determine the likelihood of a review being fake, such as the author’s relationship with other online accounts, sign-in activity, review history, and unusual behavior.
Amazon is also pursuing social media groups set up to trade reviews. In 2022, Amazon identified over 23,000 social media groups, consisting of more than 46 million members involved in facilitating fake reviews.
Numerous research projects have been aimed at busting fake reviews using natural language processing (NLP) AI techniques.
For example, one study in 2022 used GPT-2 to train a fake review classifier that outperformed humans.
Mehta stated, “We use machine learning to look for suspicious accounts, to track the relationships between a purchasing account that’s leaving a review and someone selling that product. Through a combination of meticulous vetting, advanced machine learning, and artificial intelligence, we can prevent fake reviews from reaching the customer.”
How AI detects fake reviews
If you’ve ever read a review and thought, “That looks fake,” then that more-or-less proves AI can do the same. There are signals and cues you pick up on, and AI can learn these too.
AIs can use NLP to look at factors such as excessive punctuation, poor grammar, and overly negative or positive tones, also called sentiment analysis. However, there’s more to detecting fake reviews than that, and if Amazon doesn’t define strict criteria, they risk removing genuine reviews.
Other methods include analyzing what types of products receive fake reviews, the timing of when fake reviews are left, and whether generic 5-star reviews come from the same account.
Of course, there’s no one better placed to analyze that data than Amazon, which begs the question, why is the problem as big (or bigger) than ever?
As much as Amazon wants to foster trust by removing fake reviews, maintaining large volumes of all reviews is in their best interests.
As Saoud Khalifah of Fakespot says, “If Amazon deleted 100 percent of fake reviews, they would lose hundreds of billions of shareholder value.”
