We’re still only beginning to understand the vast capabilities of machine learning. Systems are starting to search through data to look for patterns, detect those patterns, and adjust actions accordingly without being explicitly programmed to do so.
Machine learning’s greatest use cases thus far include data security, personal security and fraud detection, financial trading, and healthcare, to name a few. But some of the most significant advances it brings to the world of marketing are personalization marketing and the ability to recommend. Consider, for example, the suggested products that appear after you’ve made a purchase on Amazon, or the list of shows and movies recommended for you based on your watch history on Netflix.
Another example of this that’s particularly important to video marketers is YouTube’s use of machine learning technology.
In April 2016, YouTube made what seemed at the time like a minor announcement via its Official Blog about a new mobile design of its Home screen for YouTube’s iPhone and Android apps. But today, it’s fairly clear that this wasn’t just a new coat of paint on the same old Home screen. YouTube had coupled a fresh design with the significant application of machine learning to radically change the world of creating videos.
Here’s what YouTube did, why very few content creators realized what had hit them a year ago, and how you should optimize your videos to take advantage of the way that YouTube is using machine learning.
Acelerate your career with the Post Graduate Program in AI and Machine Learning with Purdue University collaborated with IBM.
YouTube’s Use of Machine Learning
Beyond the fresh design, last year’s announcement was about making “more relevant personalized recommendations” to help users discover videos that they would be excited to watch on the YouTube app. Product Manager Brian Marquardt said at the time, “Every day, we recommend hundreds of millions of different videos on Home, billions of times, in 76 languages...one of the biggest improvements is how the system suggests more recent videos and those from the creators you love. People who have tried the new system have spent more time watching fresh videos and content from their subscriptions.”
So, if you’re publishing fresh videos frequently on channels that already have thousands of subscribers, keep doing what you’re doing. But, if you aren’t a YouTube star yet, or you’re the video marketing manager for the typical brand, then you need to radically change what you’re doing. Why?
We discovered recently that 70 percent of YouTube viewing happening on mobile devices. And what we’ve known since July 2015, is that more than 400 hours of video are uploaded to YouTube every minute, (although I’m starting to suspect that YouTube is dragging its feet when it comes to announcing it’s updated numbers).
<Callout> “70 percent of YouTube viewing happening on mobile devices...and more than 400 hours of video are uploaded to Youtube every minute.” - Greg Jarboe
So, the odds of getting new videos discovered on YouTube channels with very few subscribers are now greater than winning the lottery.
How YouTube Buried the Lead
How could so many content creators have missed the implications of YouTube’s announcement?
Yes, it was hidden under a headline that touted a new mobile design of a user’s ‘Home’ screen. But YouTube also used a second technique to hide the news in plain sight: geek speak.
In his post, Marquardt said, “The new recommendation system is based on deep neural network technology, which means it can find patterns automatically and keep learning and improving as it goes.” His statement even included a link to a video entitled, “How Does Your Phone Know This Is A Dog?”
Now, I don’t know about you, but I’m not familiar with deep neural network technology. And I wasn’t aware that anyone would use a phone to figure out that this is a photo of a dog.
So, I had to watch 16 seconds of the cryptically-titled video to discover that deep neural network technology uses “machine learning,” a type of artificial intelligence (AI) that provides computers with the ability to learn without being explicitly programmed.
Machine learning focuses on the development of computer programs that can change when exposed to new data. And I had to watch until the 3:26 mark to learn that “deep learning” is a specific machine learning technique. Most deep learning methods involve artificial neural networks, modeling how our brains work. Deep learning forms the basis for most of the incredible advances in machine learning (and in turn, AI).
<callout> “Most deep learning methods involve artificial neural networks, modeling how our brains work. Deep learning forms the basis for most of the incredible advances in machine learning (and in turn, AI).” - Greg Jarboe
If you’ve ever finished a YouTube video and then enjoyed watching another (and another) thanks to the related videos that appear at the end of the video or on the sidebar, you’ve already benefited from an enhanced prediction engine that uses machine learning, deep learning, artificial neural networks, and deep neural network technology.
How to Optimize Videos
So how should you optimize your videos to take advantage of the way that YouTube is using machine learning? Marquardt gave us a big hint when he revealed, “People who have tried the new system have spent more time watching fresh videos and content from their subscriptions.”
In other words, YouTube considers the new system successful because people spent more time watching fresh videos and content from their subscriptions. But the key in both cases is that YouTube is rewarding “watch time.”
Watch time is the amount of time that a viewer has watched a video. It can give you a sense of what content viewers actually watch (as opposed to videos that they click on and then abandon).
YouTube uses watch time as a metric in its algorithm for suggesting videos. The algorithm prioritizes videos that lead to longer overall watch time or viewing sessions, rather than videos that get more clicks. If viewers watch your videos beyond the first click, those videos are likely to be suggested more often.
<callout> “YouTube uses watch time as a metric in its algorithm for suggesting videos. The algorithm prioritizes videos that lead to longer overall watch time or viewing sessions, rather than videos that get more clicks.” - Greg Jarboe
The idea behind the algorithm is that viewers can see more enjoyable content suggested to them, and creators can cultivate more engaged audiences. You can use the tips below to optimize your videos for higher watch time.
Enhance your skill set and give a boost to your career with the Post Graduate Program in AI and Machine Learning.
Use reports to see what's working
You can use YouTube Analytics to see what videos are successful at keeping viewers watching:
- Watch time report: Find out which of your videos has the greatest watch times and view-through rates
- Audience retention report: See which of your videos has low watch times and view-through rates. Recurring dips or decreases in audience retention graphs may have a common reason why viewers abandon the videos
- Audience engagement reports: See which of your videos drive community actions like comments, favorites, and likes
Attract viewers with title and thumbnail
Your video thumbnail and title are the first things a viewer sees when your video is suggested to them. Use these tips to make engaging content:
- Create descriptive thumbnails that show a quick snapshot of your video
- Use compelling titles for your videos that accurately represent the content (don’t use misleading thumbnails or titles)
- Translate your titles, descriptions, and captions by buying translations or having your community add subtitles
Techniques to keep viewers watching
- Be an effective editor: Create a compelling opening to your videos and then use programming, branding, and packaging techniques to maintain and build interest throughout the video
- Build your subscriber base: Subscribers are your most loyal fans and will be notified of new videos and playlists to watch
- Engage your audience: Involve your audience in your videos and encourage comments and interact with your viewers as part of the content
Organize and program your content
- Build long watch-time sessions for your content by organizing and featuring content on your channel, including using series playlists
- Create a regular release schedule for your videos when uploading to encourage viewers to watch sets of videos over single videos. You can even schedule video publish time
That’s how machine learning is changing the world of creating videos. Yes, relevance is still a factor, so you should optimize your metadata—your video’s title, tags and description—to ensure that YouTube indexes your video correctly. But watch time is now a much more important factor, so, if you want your videos discovered today, you need to figure out what YouTube is using machine learning to accomplish and make sure that you use the watch time optimization tips above.
Learn more about what Machine Learning is in this introduction to machine learning video
You can also take-up the AI and Machine Learning courses in partnership with Purdue University collaborated with IBM. This program gives you an in-depth knowledge of Python, Deep Learning with the Tensor flow, Natural Language Processing, Speech Recognition, Computer Vision, and Reinforcement Learning.