Sponsored By

Amazon Applies Machine Learning To Improve Reviews

Amazon.com has for several years capitalized on machine learning for its recommendations that come with customer purchases on the retail site. Now, it's extending those efforts to product reviews and star ratings to enhance customer experiences.

Larry Loeb

June 23, 2015

3 Min Read
<p align="left">(Image: <a href="https://pixabay.com/en/users/Simon-3/" target="_blank">Simon Steinberger</a> via <a href="https://pixabay.com/p-447034/?no_redirect" target="_blank">Pixabay</a>)</p>

COBOL Leads Us Back To The Future

COBOL Leads Us Back To The Future


COBOL Leads Us Back To The Future (Click image for larger view and slideshow.)

Amazon is turning to machine learning to improve the customer review experience for US users of its massive e-commerce site. As anyone who has used the service knows, star ratings and customer reviews have been a fixture of the Amazon.com shopping experience for more than 20 years.

Amazon.com has for several years capitalized on machine learning for the company's affinity marketing, or what it calls "Similarities," recommendations that come with a book or other customer purchase on the retail site. It sought to provide its internal teams with machine learning tools that have been put to use multiple times inside the company. Machine learning is used to make 50 billion predictions a week on Amazon.com, according to an April article by InformationWeek's Charles Babcock.

Although the latest changes went into effect on Amazon.com last week, a company spokesperson said they're not yet visible to users.

The practical application of machine learning in this context could hold long-term lessons for any CIO or IT leader looking to improve customer relationships for their business.

[ Does automation really matter to IT? Read Driverless Cars, AI, Robots: Why CIOs Should Care. ]

In the latest machine learning application, Amazon perceived a problem with its reviews system that allowed positive initial reviews (and the "star" ratings that they generate) to obscure any not-so-positive ones that were entered at a later date. Julie Law, an Amazon spokesperson, told InformationWeek in an email: "The enhanced system will use a machine-learned model to give more weight to newer, more helpful reviews from Amazon customers. The system will continue to learn which reviews are most helpful to customers, and improve the experience over time."

According to Law, the machine learning will be factored into two major aspects of how customer reviews appear on the site:

  • Star rating: A product's overall star rating will now factor in the age of a review, "helpful" votes by customers, and whether the reviews are from verified purchasers.

  • Review ranking: Similarly, these factors will be used to determine where a review appears in the overall list.

The machine learning system does not pass judgment on the actual content of the reviews (save for their age). The primary factor that the system pays attention to is whether or not other customers found the reviews useful.

In April, Amazon Web Services launched Machine Learning-as-a-Service for customers of its EC2 cloud.

Amazon Machine Learning is a managed service that analyzes a user's historical data to look for patterns and deploy predictive models. It can examine customer data and find patterns of likely customer turnover or churn. It can find typical issues in customer support. It can isolate and detect the patterns of problem transactions. The Machine Learning API and developer's guide, in the form of wizards, are available as another Amazon service.

About the Author(s)

Larry Loeb

Blogger, Informationweek

Larry Loeb has written for many of the last century's major "dead tree" computer magazines, having been, among other things, a consulting editor for BYTE magazine and senior editor for the launch of WebWeek. He has written a book on the Secure Electronic Transaction Internet protocol. His latest book has the commercially obligatory title of Hack Proofing XML. He's been online since uucp "bang" addressing (where the world existed relative to !decvax), serving as editor of the Macintosh Exchange on BIX and the VARBusiness Exchange. His first Mac had 128 KB of memory, which was a big step up from his first 1130, which had 4 KB, as did his first 1401. You can e-mail him at [email protected].

Never Miss a Beat: Get a snapshot of the issues affecting the IT industry straight to your inbox.

You May Also Like


More Insights