The Atlantic published a fascinating article this week on how Netflix creates headings atop the rows of movie recommendations.

Here’s an example from my Netflix screen right now:

Netflix 1

You may have seen headings like:

Emotional Independent Sports Movies

Spy Action & Adventure from the 1930s

Cult Evil Kid Horror Movies

Cult Sports Movies

Sentimental set in Europe Dramas from the 1970s

Visually-striking Foreign Nostalgic Dramas

Turns out Netflix has 76,897 such headings in its system at the moment!

I have often wondered exactly how these intriguing headings – Netflix calls them altgenres and the article calls them micro-genres – are generated.

The article provides the answer: with people and algorithms.

Let’s start with the people.

Using large teams of people specially trained to watch movies, Netflix deconstructed Hollywood. They paid people to watch films and tag them with all kinds of metadata. This process is so sophisticated and precise that taggers receive a 36-page training document that teaches them how to rate movies on their sexually suggestive content, goriness, romance levels, and even narrative elements like plot conclusiveness … the “social acceptability” of lead characters, and dozens of other facets of a movie.

Many values are “scalar,” that is to say, they go from 1 to 5. So, every movie gets a romance rating, not just the ones labeled “romantic” in the personalized genres. Every movie’s ending is rated from happy to sad, passing through ambiguous. Every plot is tagged. Lead characters’ jobs are tagged. Movie locations are tagged. Everything. Everyone.

Once these micro-tags are in place, an algorithm takes over and generates the headings using a template like:

Region + Adjectives + Noun Genre + Based On… + Set In… + From the… + About… + For Age X to Y

Since a movie can have dozens of adjectives, Netflix limits the length of headings to 50 characters.

Let’s relate this back to what appeared on my Netflix screen (grossly simplified):

  • Someone in the Netflix tagging army tagged The King’s Speech with emotional, british, biographical and drama micro-tags (and, presumably, many others).
  • The algorithm used a template like the one above and generated the heading “Emotional British Biographical Dramas”.
  • The Iron Lady, it turns out, was tagged with those same micro-tags and therefore falls under the same heading. Note that The Iron Lady may have many tags that The King’s Speech doesn’t have and vice-versa. What’s important that they share the tags that make up this particular heading.
  • Since I liked The King’s Speech, The Iron Lady is recommended to me.

Do these micro-genres move the needle for the business?

“Members connect with these [genre] rows so well that we measure an increase in member retention by placing the most tailored rows higher on the page instead of lower”

I am not surprised.

To understand how this approach leads to better recommendations, we will describe in a future post how traditional approaches work and how this micro-genre-based approach addresses some of their shortcomings … please stay tuned.

(HT to my colleague Bharath Krishnan for making me aware of this article)