Gambling Underdog Westerns: How Netflix uses microtagging to give you what you want

Since it's Spring Break and Netflix is FAR more tempting than thinking about metadata, here's an excuse to think about Netflix and metadata at the same time.  Imagine the football image library one could build with the infrastructure, algorithms, staff, and finances of Netflix!

Because recommendations are an important way to drive viewership and improve the user experience, Netflix uses 76,897 micro-genres to tag its videos. The combination of these tags results in those strange, wonderful, and oddly specific row headings like "Cerebral Romantic Thrillers from the 1980's."  But to someone studying metadata and digital libraries, the story of how Netflix tags those videos and organizes that information is almost as interesting as the results, and Alexis Madrigal tells all about it in this fascinating article in the Atlantic.  I guarantee that you will want to go and read this article, but I'll share a few nuggets from it here:

• Netflix uses a controlled vocabulary.  For example, it's always "Western," never "Cowboy Movie" or "Horse Opera."
• Netflix builds its headings using a defined hierarchy and order--a method anyone familiar with Library of Congress Subject Headings will appreciate. Certain descriptors, like "Oscar-Winning," always go toward the front, while other descriptors, like time periods, always go toward the end. The author sums it up with this general formula: "Region + Adjectives + Noun Genre + Based On... + Set In... + From the... + About... + For Age X to Y."
• Netflix tagging is done by humans, who watch every video and tag everything. Many of the tags are scales of 1-5.
• Netflix uses an algorithm to analyze the tags and build the "personal genres" it displays for its users.  Some genres, like "Feel-good," are based on a formula incorporating several tags.
• Netflix combines viewing data and tags to try to recommend videos it thinks you will like. The company believes that this is a better method than either relying on exclusively on your ratings or trying to recommend videos to you based on what other people watched.

And then there's the Perry Mason Effect... but you can read about that on your own!

One more bonus in the article:  it includes a "Netflix-Genre Generator" where you can generate your very own personal genres (it includes the alternative "Gonzo" setting (must be experienced) and the bland "Hollywood" setting (that explains most of the movies Hollywood churns out)).  Give it a try!