Advertisement

How Netflix's global growth is making recommendations better

It's all about the algorithms.

Netflix recommendations might appear simple on the surface, but there's a specific science behind the lists of movies and TV shows it thinks you should watch next. Star ratings, Facebook integration and catalog size have traditionally played a part, but with the streaming company now operating in 130 countries worldwide, things needed to change.

In the company's latest technical blog post, Netflix engineers Yves Raimond and Justin Basilico explain the challenges behind serving relevant content to subscribers in different countries. It boils down to four things: availability, awareness, language and tracking.

When Netflix operated in a small number of regions, recommendations were based on where you live. If you're from the UK, your data would be cross-matched with other Brits because the company knew that the catalog of movies and TV shows was the same. If Netflix tried to do the same in 100+ countries, where content vastly differs, the results wouldn't be pretty.

To solve this, the company has updated its algorithms to recognize that subscribers "have access to different catalogs based on geography and time." It suggests that recommendations will only appear if members share the same catalogs and viewing characteristics -- someone thousands of miles away could have a say in what you watch next.

But how does Netflix choose the content that members see? Subscribers from India will be more interested in viewing Bollywood movies than people in Denmark, for example. One answer would be to base recommendations on other users in a specific country, just like Netflix did before, but data from regions where subscriber bases are still very small would be skewed in favor of countries where the service is more popular.

Netflix says it will continue to combine personal and local tastes, but accept that horror fans in one country are likely to share similar opinions with horror fans in another. From there, its global algorithm can detect new trends over time, as it learns more about locations, languages and popularity.

With all of these changes, it's entirely possible that Netflix will get things wrong. To ensure that it doesn't serve content that is respected in one country but deemed insensitive in another, the company needs a competent tracking system. "Solving [this] challenge means that we're able to detect issues at a finer grain and so that our recommendation and search algorithms help all our members find content they love," say Raimond and Basilico.

With more countries coming online, the streaming giant's updated algorithms will be fed even more unique data and identify new trends over time. Your fellow countrymen and women will still have a big say in what you see, but fellow sci-fi fans on the other side of the world could help refine those Star Trek recommendations even further.