Just like Google has masses of pigeons calculating page ranks, so too does Netflix have its choice of avian figuring out what movies and shows to recommend to users. A few days ago, the streaming entertainment provider pulled back the curtain on exactly how its systems make those decisions.
A post on Netflix’s technical blog by the company’s Yves Raimond and Justin Basilico goes into quite a lot of detail on its recommendation algorithms. The article comes off the back of the service’s launch in 130 countries in early January — a mammoth feat and one that certainly required the company to take a careful look at how it selects media for users.
How hard is it then? As the post explains, there are a great many variables to take into account:
The Netflix experience is driven by a number of Machine Learning algorithms: personalized ranking, page generation, search, similarity, ratings, etc … Preparing for such a rapid expansion while ensuring each algorithm was ready to work seamlessly created new challenges for our recommendation and search teams.
One interesting tidbit that’s relevant to Australia is how content is added, particularly the issue of getting the legal go-ahead from license holders:
Most content licenses are region-specific or country-specific and are often held to terms for years at a time. Ultimately, our goal is to let members around the world enjoy all our content through global licensing, but currently our catalog varies between countries. For example, the dystopian Sci-Fi movie “Equilibrium” might be available on Netflix in the US but not in France. And “The Matrix” might be available in France but not in the US.
As you might expect, language and culture also play a vital role in deciding what content to prioritise, but applying this data isn’t a simple as it might seem:
For example, we expect that Bollywood movies would have a different popularity in India than in Argentina. However, should two members get similar recommendations, if they have similar profiles but if one member lives in India and the other in Argentina?
To address these challenges we sought to combine the regional models into a single global model that also improves the recommendations we make, especially in countries where we may not yet have many members. Of course, even though we are combining the data, we still need to reflect local differences in taste. This leads to the question: is local taste or personal taste more dominant? Based on the data we’ve seen so far, both aspects are important, but it is clear that taste patterns do travel globally.
If you’re in the mood for a long and informative read, hit up the post directly below.