Machine Learning: Netflix Knows What You Want

In our 5-Minute Guide to Machine Learning, we note that machine learning is everywhere. We offer several examples including Google and Facebook. And another very compelling example is Netflix. Every movie recommended to you on the Netflix streaming service comes from an algorithm tuned via machine learning.

What 800 Engineers Can Do

Back in 2013, Wired magazine peered inside the Netflix “suggestion machine.” Now, remember, this is four years ago. At the time, Netflix employed 800 engineers at its Silicon Valley headquarters. That’s a serious commitment to algorithms and machine learning. 

What Netflix learned then continues to be confirmed today: user behavior is much more predictive about viewing preference than user ratings. In other words, what people say they like and want to watch is different from what they actually watch.

And, according to a blog on RTInsights, that realization saves them one billion a year in customer retention.

Machine Learning Makes Cognitive Search Smarter

Apparently, Netflix users have a short attention span when it comes to searching for videos — 60 to 90 seconds. So, it would prefer to hook them with recommendations based on past viewing. But, of course, you can search on Netflix and there’s an algorithm powering Netflix search, which is also refined by machine learning.

In the enterprise scenario, attention span isn’t so much of an issue, but productivity is. Time spent searching is time not spent translating insight into action. So past search behavior is a mother lode of valuable data for search applications just like past viewing behavior is for Netflix.

Context… the X Factor

The Wired article on Netflix also explored the issue of context. Netflix noticed that viewing behavior could differ by day of the week, time of day, device, and location.

Think about that in terms of cognitive search. If you’re looking for information about same store sales on Monday, it could be for a completely different reason that if you were looking on Friday. On Monday, maybe you need general trends for an all-hands meeting. But on Friday, you need very detailed statistics for a report. Or on Monday, you’re almost always in the office. And, on Friday, you’re usually on the road.

Eventually, the algorithms should “learn” this and adjust your search results accordingly. But, context is “squishy.” And, Netflix acknowledged there were “practical challenges” to coding for that. 

Netflix for Knowledge Management?

A recent blog on TechCrunch suggested that what we really need is a Netflix for knowledge management (KM). And there’s no conceptual reason why the Netflix recommendation model couldn’t be applied to KM.

The blog quotes former GE CEO Jack Welch, who said, “An organization’s ability to learn, and translate that learning into action rapidly, is the ultimate competitive advantage.” Increasing that ability would be the focus of a Netflix for knowledge management. And machine learning would play a significant role.

To learn more about how machine learning contributes to Attivio’s Cognitive Search and Insight Platform, download our 5-Minute Guide.