The Network in Netflix
Online Resource Link: https://blog.kissmetrics.com/how-netflix-uses-analytics/
It’s fascinating to discover how networks can play such a large role in how we experience the world. They influence the people we meet, the websites we search, and even the shows we consume. Earlier this year, Netflix unveiled a new rating system that allows its customers to give shows and movies either a thumbs up or a thumbs down as a measure of enjoyment or dissatisfaction, respectively. The change is part of an attempt to provide its users with a more personalized content consuming experience. But how could the gladiatorial thumb achieve this? The answer is simple: graph theory.
Netflix formerly relied on a star-rating system that suggested shows based on overall popularity, assuming higher ratings would translate to higher enjoyment for its customers. Over time, however, it became clear that these ratings were more aspirational—what customers wanted other viewers to know about the quality of the content—than representative of actual preference and opinion. In this case, what Netflix customers were truly watching was the key to developing a media viewing experience tailored for each user. And so the thumb was born.
Its premise is simple, really. The company’s algorithms are designed to assume, much like the Strong Triadic Closure Property, that similar viewing choices denote similar user content preferences. In other words, if customer 1 (node A) enjoys watching House of Cards (node B) and Scandal (node C), and customer 2 (node D) also enjoys Scandal, then the data would suggest she might also enjoy House of Cards, creating a balanced network. Now, this is obviously an oversimplification of the complex set of algorithms at work, but it defines the basic premise of matching users to content based on compatibility within a much larger graph or network. These algorithms provide Netflix with troves of data, which allows them to see trends and form connections, or edges, among its customers’ activity, ultimately leading to hyper-personalized recommendations.