We stream a lot of video content, right? Back in 2020 we collectively managed to spend 480 billion hours in front of our TV’s, phones, and laptops streaming TV shows and movies.
But we don’t stream enough, really. And getting us to watch more quality content is the prime objective for streaming services and OTT platforms like Netflix.
I am here to argue that Netflix can do a little more with AI to provide a better viewing experience to their subscribers. Let me tell you how.
AI is the name of the game
There are two ways in which an OTT platform can deliver a superb viewing experience to its audience.
First thing is the content. A good strategy usually involves acquiring popular titles as well as producing exclusive content. Both mean that fans can watch certain shows and movies only on certain platforms, which guarantees an increase in the number of subscriptions.
Just think about that: for some time, the most viewed show on Netflix was “Friends”. And now Netflix original programming often beats the competition in terms of popularity.
But we are not here for that. There are more educated people out there better suited for content production.
So, let’s talk about the second factor — the tech behind the content. The kind of tech that automates production processes to spare the editors valuable time.
We will discuss the ways in which Netflix uses AI to facilitate some of its content production, and how it can improve on that.
Say you’re Netflix. And you got a bunch of shows that you need a lot of people to watch.
How do you get people to subscribe to the platform? That’s right — promotion! People see a show clip or a trailer on social media; they give it a try; they stay on the platform. Simple.
Now, say you got thousands of TV series, movies, and documentaries. Having a person cut the trailers for all of that content will cost a lot of time, won’t it?
Enter Artificial Intelligence — the tool that seemingly can automate the process of trailer generation.
Now, let’s see that bad boy in action.
We start with the data that AI uses to do its magic. Basically, each title on the streaming platform has a set of data, tags, and summaries attached to it. AI guys at Netflix call that data “embeddings”
Those embeddings usually include:
- Text summaries;
- Content type.
Now, to the technology that relies on that data.
Basically, it is a standard machine learning model that relies on source tasks to analyze the target tasks.
The engineers train a source task using many historical titles and embeddings tied to them. Then they apply the model to target tasks to extract the data they need.
That data is then used to find and tag the most prominent characters and scenes that would make up a trailer or a best moment compilation.
Then those best moments compilations, clips from comedy specials, and trailers can be used on social media to draw in potential subscribers.
But there are a couple of problems with that approach.
The problems with AI in media and entertainment
Let’s hang on to the embeddings for a minute. Why do you need additional data sources if in theory AI should do all of the heavy lifting by itself? It has object recognition, right?
Well, that is the major problem with AI when it comes to generating video content. It’s just not accurate enough to deliver consistent results where understanding the context is essential.
At the end of the day, the AI model is as good as the source task is, which leaves a lot of room for mistakes when working with incomplete sets of data. Moreover, it is unable to pick up the context of the video content just from the visuals alone.
And that is why the engineers use several data sources like tags, content genre and type information, and script extracts.
The other issue one might encounter using AI for content marketing is the time it takes to actually apply it.
No task is the same in automatically generating trailers for a variety of shows and movies. Unfortunately, that variety subtracts the automation aspect out of the equation.
Get this: you can have an algorithm to create an automated trailer of an action movie. Now, let’s switch to a comedy special.
How do you create a trailer for that? Right, you get back to the “source task and target tasks” thingy. That retraining will take a good chunk of the time you hoped to save by automating the process.
What can we do to address those issues?
Cognitive computing picks up the slack
We have the technology that can work with incomplete sets of data, does not require additional sources of information, and needs no training.
Cognitive Computing tech provides something that the AI can not possibly master — human-like intelligence and decision-making.
Let’s see how it solves AI problems.
Powered by Cognitive Computing, CognitiveMill™ processes hours of video and delivers the content in just minutes. It quickly adapts to different genres and types of content, so your staff won’t have to re-train it every time the task changes.
It saves you time in other ways as well.
CognitiveMill™ does not just give the probability scores or json files filled with metadata. It goes further with it and cuts the content for you completely automatically.
Say you need a trailer to promote a movie you are about to host on a platform. Instead of having your editors cut it for hours, you feed a movie to the tool and get a complete automated trailer 50 times faster.
Unlike Artificial Intelligence, Cognitive Computing tech uses an abundance of tools to actually understand the content it analyzes. Just like the context of the task helps humans make accurate decisions.
We are trying to imitate human thought processes, remember?
CognitiveMill™ relies on digital image processing along with cognitive and traditional computer vision to recognize what is happening in the footage.
Make no mistake: it doesn’t just pick up on cheers, or sounds, or certain objects. It makes sense of everything that it sees and makes a decision based on that info.
So rest assured: you won’t have to scramble scripts or seat your editors in front of hours of video content tagging everything.
Sounds like an overall good tech package to me.
Enhancing OTT content production with CognitiveMill™
Technological advancements are great and all, but what about its practical applications?
Well, a more precise and thorough analysis can help platforms such as Netflix to create better short-form content for promotional purposes.
In addition to that, the technology helps do that on a larger scale.
We already talked about automating the trailer generation process, and there are a lot more that CognitiveMill™ allows you to do.
The solution forms an entire ecosystem of products that help you create. manage, and share video content. Here are a couple examples of what you can do:
- CognitiveSkip™. Remember that nifty “skip intro” button? What if you could also skip the closing credits? From the platform side, CognitiveSkip™ analyzes a TV episode or a movie and tags the credits automatically. The viewers can be sure that they won’t miss anything important story-wise: the tool can work around those post-credit scenes that Marvel likes.
- CognitiveCrop™. The best way to share a clip on social media is to make it comfortable to view on there. You guessed it — that means cropping the clip to vertical format. And I just happen to know a solution that can do it automatically with a large volume of video content.
- CognitiveCast™. Cast members' information also improves viewing experience for subscribers. Push a pause button, get names of everybody on the screen, or filter the shows based on which actors star in them.
Those and other products enable marketers to cut the resources and time spent on supporting their content library with features and promotions. You say I can make my shows more comfortable to find and watch, and promote them more effectively? Count me in!
If you want to automate production processes on your OTT platform — let’s talk! You can reach us at firstname.lastname@example.org or by filling in the form below.