By Michael Martinez and Lori Cameron
Before Amazon became the juggernaut that made Jeff Bezos a household name, its recommender system was already helping customers find what book to read next.
Since then, that simple system of determining a customer’s next likely purchase has penetrated just about every nook and cranny of our digital lives.
Indeed, the recommendation algorithm is now used for more than just buying books from Amazon, which began as an online bookseller.
Today, the online world can’t watch a movie, buy a laptop, select a dress, or pick a YouTube video without encountering one of the most popular recommendation algorithms today.
Computer experts call it “item-based collaborative filtering,” and when Amazon began using it in the 1990s, it broke from tradition. Up until then, algorithms were “user-based” and recommended the next purchase based on what people with similar interests and purchase patterns were finding.
Instead, Amazon devised an algorithm that began looking at items themselves. It scopes recommendations through the user’s purchased or rated items and pairs them to similar items, using metrics and composing a list of recommendations.
The beauty is that such an algorithm uses far less data space, on a scale of up to three orders of magnitude.
The result has been an algorithm that “scales to hundreds of millions of users and tens of millions of items without sampling or other techniques that can reduce the quality of the recommendations,” according to a new research report by Microsoft data scientist Greg Linden and Amazon recommender system expert Brent Smith.
In fact, Amazon has since grown a hundred-fold and is now the world’s largest retailer, bigger than Wal-Mart, in terms of market cap. Also, Bezos, CEO of Amazon, is just $5 billion away from being the richest man in the world, as of June 2017. Microsoft’s Bill Gates holds that title, at the moment.
In 2013, Amazon even won an Emmy Award for its work on video recommendations.
“Nearly two decades ago, Amazon.com launched recommendations to millions of customers over millions of items, helping people discover what they might not have found on their own. Since then, the original algorithm has spread over most of the Web, been tweaked to help people find videos to watch or news to read, been challenged by other algorithms and other techniques, and been adapted to improve diversity and discovery, recency, time-sensitive or sequential items, and many other problems,” Smith and Linden wrote in their “Two Decades of Recommender Systems at Amazon.com” in the July 2017 issue of IEEE Internet Computing magazine.
“Yet the field remains wide open,” they wrote.
They imagine machine learning providing an interactive service where shopping is as easy as a conversation.
“This moves beyond the current paradigm of typing search keywords in a box and navigating a website. Instead, discovery should be like talking with a friend who knows you, knows what you like, works with you at every step, and anticipates your needs,” they wrote.
“This is a vision where intelligence is everywhere. Every interaction should reflect who you are and what you like, and help you find what other people like you have already discovered. It should feel hollow and pathetic when you see something that’s obviously not you; do you not know me by now?” they further stated.
The future of recommendations and personalization will continue to be computers helping people help other people, they said.