Netflix Recommendation System: Inside the Algorithm
Over 80% of the Netflix content that users watch is discovered through Netflix’s recommendation system. The goal of this blog is to give a basic understanding into how exactly this system works.
The Netflix recommendation system collects data from multiple sources. Using Machine Learning and Algorithms, Netflix recommendation system finds content that users may not have considered watching.
Overview of how Netflix Recommendation system and algorithms work
Content is organized into rows categorized usually by a genre type or a more specific term or phrase that connects all the titles in a row. These categories are personalized to match users interests. To make this even more effective, the Netflix algorithm is always striving to get better. Every time a user watches a title, Netflix collects more user data and this data is input into the Netflix algorithm.This gives Netflix feedback on to which categorization techniques were effective and which were not. Since this is an iterative process, the more time you spend watching content on Netflix, the more data Netflix collects and inputs into the algorithm, which then allows the recommendation system to give better and more personalized recommendations.
The type of data Netflix collects from users
The data Netflix collects from users is much more complex than just the genre of the titles users watch. It is true that Netflix cares about the basic information of a film but in general specific data about Netflix members is the key to how this system works. Netflix collect data about entire users’ watch histories. Netflix wants to know ‘what did you watch before a certain title?’ and ‘what did you watch after?’. They also collect information like the time of day of a watch.
How Netflix Curates their Library
Netflix’s library is made up of over 5000 titles available for the American market alone. Thus, It would be inefficient and ineffective for Netflix to blindly offer members the entire catalogue. Thus, for both Netflix and users sakes, the Netflix library has to be curated. Since quality and taste are rarely synonymous for all users, Netflix has to make sure that recommendation are specifically tailored to each unique user.
Netflix also uses member similarity techniques to help with recommendations as well. If a user matches the viewing patterns of another user, then Netflix will generate recommendations to the new user that will be very similar to the other users tastes and preferences.
If you ever wondered why some categories or genres can be oddly specific, that is also intentional. In fact, Netflix had real people tag every single title in the catalogue with highly specific categories to make the recommendation system even more personalizable.
The Importance of Rows to Netflix
Netflix uses rows because they believe doing so, allows members to more easily navigate through the catalogue. By using specific genre or categories in each row and sorting the order of titles in a row, Netflix members can easily decide whether a row is worth exploring or not. This is a way for Netflix to have members spend more time watching titles (while Netflix collects more data for algorithm) and less finding new titles. Also, keep in mind, that each device used for Netflix streaming, has different capabilities in terms of numbers of rows that can be displayed on a page. Thus, Netflix designs its interface to be best optimized for different hardware.
Challenges of Netflix’s Row System
One of Netflix’s biggest challenges is finding balance in how unique and personalized categories should be. Categories should be relevant to users but at the same time Netflix strives to broaden users tastes. Thus, Netflix aims to create groupings that can be fresh and new but also ensure that the groupings are relevant to a user’s history and preferences.
How Netflix Uses Images To Influence Recommendations
Netflix actually uses content’s artwork to help fuel recommendations as well. Surprisingly, even the images you see for Netflix content is personalized. The images are intended to keep members hooked. Netflix believes that if a title does not draw a members interest within 90 seconds, they are likely to move on. Thus, Netflix believes that images are the best way to match members with the ‘perfect title’. After performing some experimentation and testing, Netflix concluded that images that convey a certain spectrum of emotion are best at interesting members.
As you can see in the thumbnail above, members interest and values can vary internationally. Thus, the most popular thumbnail for a show can vary drastically in different countries.
As you can see in the example above, Netflix’s system learns what mood, emotion, or tone a certain user finds attractive or interesting. This information is then used to find a thumbnail that best matches that emotion,tone or mood for a recommendation.
Conclusions
Netflix’s main purpose is to get members as highly engaged and engrossed as possible. However, the goal of this blog is to look at the ideas and methods Netflix uses to keep up viewer engagement and increase interest in the platform rather than to debate the ethics of the addictive system built. Netflix’s recommendation system is effective because Netflix understands the limitations of the interface between the catalogue and users. Thus, by using the row system and iterative data collection and analysis, Netflix is able to generate quick recommendations that gets better as users gets more addicted. Also by using unique image thumbnails, Netflix has an effective tool to increase users’ interest in titles they may have skipped by appealing to their psychology.