Big data needs a big brain to process it and so Netflix is taking a page from other areas of data analysis and is bringing Deep Learning to the video entertainment recommendation arena. While the recent blog post on the topic at Netflix talks more about infrastructure and server architecture, I thought it would be more interesting to talk about the video discovery aspect of it all.
Artificial Neural Networks are, in a nutshell, what Netflix is looking to bend to their will and help you get the most out of the mammoth content catalog at the service. If you’re super interested in the intricacies of utilizing GPUs and distributed computational models in order to train an artificial neural network, then the Netflix blog post is for you. As for the rest of you, let’s look at more practical uses and how it could eventually impact our daily lives in the online video industry.
In a nutshell, training an artificial neural network could be the beginning of making the computer think more like you do. That in turn means it could give you recommendations more along the lines of your own thinking, therefore getting you to watch more video that you enjoy and make your entire experience all that much better, raising your loyalty level, increasing the likelihood you will recommend them to others and ultimately, generating more subscribers for Netflix.
Deep Learning in Video Discovery
Even though Netflix, smartly, took an approach of having the best of documentaries on every possible topic, there’s still a massive pile of them on the site and sorting them into lists that people would be interested in is a monumental task. To sort through all that content and pull out the select few things you might be interested in is nigh impossible simply by sorting things based on the tiny bits of information you give to Netflix when you build your queues and rate things. Essentially, they want to have an artificial neural network that will model your decision making processes and then be able to output exactly what your own brain would have.
Deep Learning in Content Licensing
When they can make the artificial neural net act very similarly to how you choose your viewing content, they can then not only target specific content to you but they can aggregate the ‘thinking’ of all of their subscribers and better focus their content creation and licensing practices more toward content that will actually be watched by a large number of subscribers. This will help them optimize the money they spend acquiring content because they will be able to get the content that they know you want to watch. Granted, it will never be a perfect system, nor will it ever take every single user into account, but if they can get it to even 60-75% effectiveness it could save them, literally, millions in licensing each year because they could only buy content that they know the majority of the subscribers want, maybe even before the subscribers know they want it.
It might sound like some Minority Report, pre-cog mumbo jumbo, but it could work and could give Netflix a better outlook for its licensing forecasting. Plus, think about the content creation side of things.
Instead of putting money into projects they think people will watch. They will be able to input the parameters of the show into an artificial neural network and then it will decide if it’s worth investing in. I just hope they don’t let it run the whole of the Kree Empire…no? Supreme Intelligence… Marvel comics reference? Fine, forget it.
The artificial neural network could also be used to model viewing patterns so that they can maximize bandwidth availability during certain periods. This could not only help them with their own infrastructure but also help them in terms of their Amazon Web Services planning and purchasing.
By bending these technologies to their will, Netflix could truly begin to optimize multiple facets of its business from the technology side all the way through to the content acquisition side. They could also turn around and sell their system to others as well and turn Netflix into so much more than just a video entertainment delivery company, but into a true powerhouse in terms of big data analysis aimed at practical implementation. Think about what kind of advertising targeting could be done if the ad networks could loosely model your own brain.