Want CNET to notify you of price drops and the latest stories?

YouTube gets better at watching you

The video site improves its recommendation engine, with hopes of getting us to watch even more.

Ben Fox Rubin Former senior reporter
Ben Fox Rubin was a senior reporter for CNET News in Manhattan, reporting on Amazon, e-commerce and mobile payments. He previously worked as a reporter for The Wall Street Journal and got his start at newspapers in New York, Connecticut and Massachusetts.
Ben Fox Rubin
2 min read

YouTube has figured out a way to get deeper inside your head.

The Internet's biggest video site on Tuesday rolled out a smarter machine learning engine on its iOS and Android mobile apps, allowing it to serve up better recommendations for viewers. The gussied-up recommendation system is based on deep neural network technology -- the same type parent Google uses for search results -- which will help YouTube find patterns and learn more about what a viewer wants with each visit.

"Delivering a personal recommendation engine that shows YouTube really understand you is our goal," Johanna Wright, vice president of YouTube product management, said in an interview. "We're able to do this because Google has some of the best machine learning in the world."

Enlarge Image

The new YouTube app home page will include bigger images and a more powerful machine learning system that can sense which cat videos you want to watch next.


One of the biggest changes involves something YouTube calls "freshness," which allows the recommendation engine to find and serve up videos that were uploaded less than a hour earlier. That feature will also help the system recommend newly uploaded videos from channels a viewer subscribes to. Doing that is no simple feat, since 400 hours of video are uploaded to YouTube every minute, and every day YouTube recommends hundreds of millions of different videos to its legions of users.

With YouTube's strategy focused on mobile first, Wright didn't have anything to share on when the new system would be added to YouTube's desktop website.

Getting the recommendation right is critical for YouTube, since identifying new videos a user will want to click on will keep him watching. That helps the site serve up more ads and make more money from its 1 billion monthly visitors.

YouTube is confident enough in its under-the-hood recommendations improvements that it also on Tuesday cleaned up some clutter on its apps, providing bigger images and fewer clusters of videos on the apps' home pages. That means the apps will show fewer videos at a time, but, hopefully, more useful ones for viewers.

"We've made so many changes to machine learning system," Wright said, "that we can more easily tell what videos people want to watch."

But, don't worry too much if you want YouTube to get out of your head. In the interest of transparency and control for users, Wright said people will still be able to log out of YouTube, see their watch history and delete it if they want.