Amazon puts a learning AI in charge of reviews

Amazon has quietly tweaked its reviews algorithm, using machine learning to not only push fresher feedback to customers, but learn which reviews are most useful. The changes, which went into action on Amazon's US site late last week, could well make a noticeable difference to which customer reviews show up top of the list and which are buried at the bottom; meanwhile, it could also have a big impact on how many stars each product scores.

"The enhanced system will use a machine-learned model to give more weight to newer, more helpful reviews from Amazon customers," a company spokesperson told GeekWire. "The system will continue to learn which reviews are most helpful to customers and improve the experience over time."

That will initially be visible in two ways. First, Amazon's machine learning system will adjust the order by which reviews are shown according to its ongoing assessment of which are most important.

Currently, which review is seen first is generally shaped by the number of votes it has been given by other shoppers.

Secondly, Amazon will take into account not only the average rating given by each reviewer to figure out the overall star rating, but a number of other criteria. For instance, the age of the review might be taken into consideration, along with whether it was left by a verified purchaser.

The change should address criticisms that products, although perhaps successful initially, can prove less capable over time, but that those changes aren't visible in reviews because earlier, more positive feedback remains the first seen.

Exactly how Amazon will calculate the weightings is likely to remain a closely-guarded secret, so as to avoid unscrupulous vendors attempting to game the system.