Machine Learning Makes TV More Watchable Than Ever

Machine Learning Makes TV More Watchable Than Ever

The television industry is being turned on its head by video streaming services that use machine learning (ML) to make entertainment more personalized, accessible, and automated than ever before.

Let’s start with the sheer volume of TV that people are consuming. In 2017, subscribers to one of the largest online television services watched a combined 52 billion hours of content.1 To put that number into perspective, if people could travel at the speed of light, this would be the same amount of time to travel to the Andromeda Galaxy—the closest galaxy to the Milky Way—and back!

And the video streaming industry is only getting bigger. According to analysts, the video streaming market is expected to pull in nearly $20 billion USD in 2018 and grow over the next 5 years to reach $25 billion in 2023.2 That means more people than ever before glued to their favorite shows on home televisions, laptops, tablets, and mobile phones.

Machine Learning Makes TV More Watchable Than Ever
An estimated 966.5 million users will use video streaming by 2023. That means more days and nights, cuddled up on the couch, binge-watching the latest-and-greatest TV series, films, and more.

As the video streaming industry reaches record revenues and users, machine learning is playing a key role. That includes personalized user interfaces, expertly recommended content, and automated video production. In this blog post, we’ll take a look at how machine learning is scaling enjoyable television experiences.

A Personalized TV Experience through Machine Learning

Online streaming companies are creating personalized experiences for each of their users—and they’re taking a rigorous, scientific approach to do so. Like a scientist conducting an experiment, teams of data scientists use extensive testing to understand their users’ behaviors.3

The process starts by understanding each viewer: what they’re watching now, what they watched before, what they watched next, when they watched, how often they “binge” watch, where they click most on the screen, and more. At the same time user behavior is collected, the entire video library is tagged based on genre, length, and a number of other, nuanced features within the content. Finally, machine learning algorithms are run to find out which videos should be recommended, where they should be placed on the screen, which graphics should be used, and more. The challenge is to glean insights from a user’s implicit behavior, which form the majority of useful data.

The process naturally tends to group users into micro-communities of people who tend to watch the same content. On a macro-scale, this means each user is able to discover content that they’re likely to enjoy, but might not have discovered on their own. On a micro-scale, the result is a customized homepage for each user, personalized from the user interface to the video content curated to the title cards.

All these efforts to create a hyper-personalized experience are paying off. For example, over 80 percent of all TV shows watched on a leading video streaming platform are from the platform’s recommendation engine.4 That means more people watch TV shows that are recommended to them than they choose independently. In terms of revenue, one streaming television company even estimates that machine learning saves $1 billion per year in subscription renewals.3 When it comes to video streaming, happy customers are loyal customers.

Automating Live Event Coverage for Broadcast TV

To combat heavy losses of subscribers, traditional cable companies along with satellite TV providers are starting to adopt machine learning. For instance, the largest British public service broadcaster plans to use artificial intelligence (AI) and ML to curate an evening’s worth of video content from their massive archive.5 Doing so could extract value from the metadata of archived content through automated content repurposing. It’s a renewed focus on media archival and digital asset protection.

Outside of archived content, traditional cable companies are also looking into ways to scale their live event coverage. Traditionally, however, this has required a proportional increase in production staff to cover each event. Machine learning might change that. The same British broadcasting service is exploring the use of an automated video editing package that uses AI and ML to select and sequence video crops recorded on high-resolution cameras.5 The benefit would be to automatically frame and cut live event coverage that is close enough in quality to a manually produced piece—while giving human editors control over final creative direction.

MORE: The second largest jazz festival in the world—preserves nearly 50 years of music history using cloud object storage.

Machine learning isn’t just something that broadcasters are committed to today: they also see additional applications in the future. One potential application is location scouting, the intensive process for directors to find the right venues for their video shoots. By feeding a database of photographs and videos to a machine learning algorithm, directors could quickly produce a shortlist of potential shoot locations. In addition, broadcasters are looking at uses of machine learning to select high-quality shots within a large pool of video assets. This would free up time and resources to be spent on other pre-production and post-production tasks.

Machine Learning Goes to the Movies

Not to be outdone by broadcast or streaming television, some movie studios are starting to turn to machine learning to predict box office success. Recently, a major movie studio partnered with a leading search engine to analyze trailers of recent movies.6 Their goal was to predict whether or not people would see the full movie in theaters, after seeing the trailer. The research pair first used machine learning to understand how certain features in movie previews feed into our understanding of a film’s genre. Then, the AI combined movie categorization with historical data of a moviegoer’s preferences to predict their likelihood of buying a ticket for a given follow-up movie.

The experiment worked well. In fact, in one of their test cases, the algorithm was able to correctly predict 11 out of the top 20 follow-up movies that moviegoers watched.6

Closing Thoughts

Machine learning is making TV and film more personalized, accessible, and automated than ever. In addition to customizing each user’s experience, machine learning is informing other steps in the TV show workflow—from coordination of logistics to how certain shows are recorded and produced. The end result has been an increase in the number of people who subscribe to streaming services, as well as the amount of streaming television that they watch.

So the next time that you’re curled up, binge-watching your favorite show, you can thank machine learning for the recommendation.


Read More on Machine Learning in Media & Entertainment


  1. Netflix users collectively watched 1 billion hours of content per week in 2017.
  2. Video Streaming (SVoD) – worldwide.
  3. AI is changing how you watch TV.
  4. This is how Netflix’s top-secret recommendation system works.
  5. AI in Production.
  6. Fox is Using Google’s Machine Learning to Predict What Movies You’ll Like.

Related Posts