Deep learning might make your Netflix recommendations a lot better

Streaming video giant Netflix is experimenting with the advanced artificial intelligence method commonly known as deep learning, it explained on its tech blog Monday. The authors didn’t get into how it might use the technology (other than “to improve our product”) but the chances are it’s eyeing deep learning as a method for making its movie recommendations more accurate.

As we have covered pretty extensively recently, deep learning and similar approaches to artificial intelligence are primarily focused on computer vision/object recognition and language understanding right now, and web companies of all stripes are trying to take advantage of them. Google, Facebook, Dropbox, Pinterest, Yahoo, Microsoft, Amazon — they’re all trying to train systems that can recognize the features of images and text in order to better understand what’s in them. Systems that better understand what text means and what’s going on in images can better perform tasks such as tagging content, but they also can give a new level of insight into customers’ preferences.

It only makes sense that Netflix would try to ride this current wave of innovation, as well. The company already analyzes every little thing about what its subscribers are watching and how they’re interacting with its site, and the types of deep learning models it’s trying to build could take those efforts much further. Perhaps they would help Netflix realize that horror movie fans prefer cover art with screaming virgins over scary monsters, or that certain on-screen imagery or language tends to turn certain people off. Understanding how content providers and subscribers use certain words or phrases could improve search by making more about what users really want and less about what they actually typed.


And of course, because this is Netflix, it’s using the Amazon Web Services cloud to power its research.

The cloud — as we’ll discuss at our Structure Data conference in March with speakers from IBM’s Watson division and AlchemyAPI — will be a key to making deep learning and other types of artificial intelligence consumable by the mainstream. Cloud computing can provide the short-term computing power necessary to analyze all the content (it’s often a very computationally intense process, which is why Netflix and others use GPUs for their parallel processing capabilities) but might be even better as a place to host services that application developers can consume via API.

Even quantum computer manufacturer D-Wave Systems (whose CEO Vern Brownell will also be speaking at Structure Data) believes the cloud will ultimately be the biggest delivery model for its new methods of processing data. Among the other speakers at the conference discussing real-world applications of artificial intelligence are Expect Labs (creator of the MindMeld messaging app) Founder and CEO Tim Tuttle and SwiftKey Co-founder and CTO Ben Medlock.

MindMeld listens to conversations and recommends relevant content.
MindMeld listens to conversations and recommends relevant content.

Although the technology might be revolutionary, we might be surprised how it trickles into our lives little by little, and even how small a role it plays among the numerous predictive models that a company like Netflix uses. There won’t be one day it decides to turn on its deep learning models and all of a sudden we’re seeing the results of them at work. More likely, if it works as advertised, we’ll probably just realize at some point down the road that we’re seeing more stuff that we like and that search seems to work better.

But for the companies doing the research and fighting for attention in a web world where he who has the best personalization and the best user experience wins, every little improvement helps.

Feature image courtesy of Shutterstock user phipatbig.