- Netflix last week said that it no longer needed to borrow money to fund its ambitious content plans.
- Crucial to Netflix's future growth is helping audiences find the TV shows and movies it invests in.
- Netflix outlined in a 57-slide presentation how it approaches personalization and the challenges ahead.
- Visit Business Insider's homepage for more stories.
Netflix last week announced two major milestones: It surpassed 200 million paid subscribers and said it no longer needed to borrow money to fund its ambitious content plans.
The streaming-video company's stock soared 70% in the last year as the service lured subscribers in greater numbers with buzzy originals like "Bridgerton," "Lupin," and "The Queen's Gambit."
Content is crucial to Netflix's remarkable growth, but so is making sure audiences find and watch the TV shows and movies Netflix is spending heavily on.
Its personalized content recommendations are an important piece of that puzzle as they help subscribers sort through the thousands of titles on the service.
Justin Basilico, a director of research and engineering at Netflix, explained in a 57-slide presentation during a virtual conference in September how the company thinks about personalization, including ways to improve its recommendations and challenges still to be solved. (Netflix regularly presents research to the tech industry and publishes a blog on its work.)
Netflix said in the presentation that it designs its recommendations to maximize "long-term member joy," a rather lofty goal that the company also said can be tough to measure and understand. It generally involves making the platform as frictionless to use as possible, so people find titles they like with ease and regularity.
One challenge Netflix is trying to solve for, which it described on slide 19, is preventing a recommendation system's feedback loops from overinflating certain titles or types of content.
Netflix, for instance, may recommend "Bridgerton" because other people on the platform are watching the show in greater numbers. As more people accept Netflix's recommendation and watch the series, the show climbs higher in the system's rankings. It then causes a bias in the algorithm to recommend "Bridgerton" over other series that aren't as popular.
The presentation outlined a few ways Netflix could approach problems like these in personalization.
For one, Netflix wants to go beyond recommendations based on correlations between different data points to try to understand causality, or what drove a user's decision making. In other words, did you watch "Bridgerton" because Netflix recommended it to you? If you had to watch it, would you still like it? How can this information train Netflix's models?
Netflix can also calibrate its recommendations to help expose people to a wider range of programming. Algorithms tend to accentuate dominant interests. While a person who watches 70% romances and 30% action movies should get recommendations in roughly that ratio, algorithms that prioritize accuracy may only suggest romances because the user is most likely to watch them. Calibrating the recommendations can help balance out that mix.
Netflix is training its systems to get better at personalizing the service based on context, as well. That context includes what device a person is using, the account they're logged in through, or the page they're viewing. Netflix might display different artwork for "Bridgerton" on a smart TV than a smartphone, or when a user is scrolling the homepage compared with a search-results page, or even in different rows on the same page.
The overarching evolution of Netflix's recommendation system, the presentation shows, is toward personalizing the full experience, from the titles shown on screen, to the artwork and synopses displayed alongside them, and to the emails sent to users with suggestions for what to watch next.
Disclosure: Mathias Döpfner, CEO of Business Insider’s parent company, Axel Springer, is a Netflix board member.
Source: Read Full Article