What should I watch this night after a hectic day at office?
This is the question that pops into your mind once you are back home from the office and sitting in front of the TV with no remembrance of what kind of shows you watched recently. Today, everyone wants an intelligent streaming platform that can understand their preferences and tastes without merely running on autopilot. From Netflix to Amazon Prime – recommendation systems are gaining importance as they directly interact (usually behind the scenes) with users every day.
With over 139 million paid subscribers(total viewer pool -300 million) across 190 countries, 15,400 titles across its regional libraries and 112 Emmy Award Nominations in 2018 – Netflix is the world’s leading Internet television network and the most-valued largest streaming service in the world. The amazing digital success story of Netflix is incomplete without the mention of its recommender systems that focus on personalization.
Machine Learning- Making Netflix Win the Personalization Battle
Have you ever thought why the Netflix artwork changes for different shows when you login to the account? One day it might be an image of the entire bridge crew while the other day it is the Worf glaring at you judgingly. If you are Netflix user you might also have noticed that the platform shows really precise genres like Romantic Dramas where the leading character is left-handed. How does Netflix come up with such precise genres for its 100 million-plus subscriber base? How does Netflix artwork change? It’s machine learning, AI, and the creativity behind the scenes that guess what will make a user pick a particular show to watch. Machine learning and data science help Netflix personalize the experience for you based on your history of picking shows to watch.
Did You Know?
Netflix began using analytic tools in 2000 to recommend videos for users to rent.
Netflix just has a 90-second window to help viewers find a movie or a TV show before they leave the platform and visit some other service. That’s one of the major reasons why Netflix is so obsessed with personalizing recommendations to hook users.
Netflix’s personalized recommendation algorithms produce $1 billion a year in value from customer retention.
Majority of Netflix users consider recommendations with 80% of Netflix views coming from the service’s recommendations.
Netflix has set up 1300 recommendation clusters based on users viewing preferences.
Netflix segments its viewers into over 2K taste groups. Based on the taste group a viewer falls, it dictates the recommendations.
With over 7K TV shows and movies in the catalogue, it is actually impossible for a viewer to find movies they like to watch on their own. Netflix’s recommendation engine automates this search process for its users.
Personalization of Movie/TV Show Recommendations
Netflix’s chief content officer Ted Sarandos said –
There’s no such thing as a ‘Netflix show’. Our brand is personalization.
Personalization begins on Netflix’s homepage that shows group of videos arranged in horizontal rows. Each horizontal row has a title which relates to the videos in that group. Most of the personalized recommendations begin based on the way rows are selected and the order in which the items are placed. Recommender systems at Netflix span various algorithmic approaches like reinforcement learning, neural networks, causal modelling, probabilistic graphical models, matrix factorization, ensembles, bandits.
Netflix’s recommendation systems have been developed by hundreds of engineers that analyse the habits of millions of users based on multiple factors. Whenever a user accesses Netflix services, the recommendations system estimates the probability of a user watching a particular title based on the following factors –
- Viewer interactions with Netflix services like viewer ratings, viewing history, etc.
- Information about the categories, year of release, title, genres, and more.
- Other viewers with similar watching preferences and tastes.
- Time duration of a viewer watching a show
- The device on which a viewer is watching.
- The time of the day a viewer watches -This is because Netflix has the data that there is different viewing behaviour based on the time of the day, the day of the week, the location, and the device on which a show or movie is viewed.
For every new subscriber, Netflix asks them to choose titles they would like to watch. These titles are used as the first step for personalized recommendations. Later as viewers continue to watch over time the recommendations are powered by the titles they watched more recently along with other factors mentioned above. Netflix’s machine learning based recommendations learn from their own users. Every time a viewer spends time watching a movie or a show, it collects data that informs the machine learning algorithm behind the scenes and refreshes it. The more a viewer watches the more up-to-date and accurate the algorithm is.
Personalization of Artwork/ Thumbnails
The main goal of Netflix is to provide personalized recommendations by showing the apt titles to each of the viewers at the right time. But, why should a viewer care about the titles Netflix recommends? How does Netflix convince a viewer that a title is worth watching? How does Netflix grab the attention of a viewer to a new and unfamiliar title? Answering these questions is important to understand how viewers discover great content, particularly for new and unfamiliar titles. Netflix tackles this challenge through artwork personalization or thumbnails personalization that portray the titles.
Netflix differs from a hundred other media companies by personalizing the so-called artworks. They say an image is worth a thousand words and Netflix is tapping on to it with its new recommendation algorithm based on artwork. The artwork for a title is used to capture the attention of the viewer and gives them a visual evidence on why it could be a perfect choice for them to watch it. The thumbnail or artwork might highlight an exciting scene from a movie like a car chase, a famous actor that the viewer recognizes, or a dramatic scene that depicts the essence of the TV show or a movie. For every new title various images are assigned randomly to different subscribers based on the taste communities. Netflix then presents the image with highest likelihood on a user’s homepage so that they will give it a try.
Netflix makes use of thousands of video frames from existing TV shows and movies for thumbnail generation. The images are then annotated and ranked to predict the highest likelihood of being clicked by a viewer. These calculations depends on what other viewers with similar taste and preferences have clicked on. For instance, viewers who like a particular actor are most likely to click on images with the actor.
Other Applications of Machine Learning at Netflix
- Machine learning shapes the catalogue of TV shows and movies by learning characteristics that make content successful among viewers.
- It powers the advertising spend, advertising creative, and channel mix to help Netflix identify new subscribers who will enjoy their service.
- Optimize the production of TV shows and movies.
- Optimize audio and video encoding, in-house CDN, and adaptive bitrate selection.
We have to thank machine learning and data science for having totally disrupted the way media and entertainment industries operate. It is pretty clear that Netflix’s amalgamation of data, algorithms, and personalization are likely to keep users glued to their screens. It will be interesting to see how the media and entertainment industry will reshape with machine learning and artificial intelligence.