How Does Recommendation Systems of Netflix, Amazon, Spotify, Tik Tok and YouTube Work?
Artificial intelligence (AI) and machine learning are buzzwords that are increasingly appearing through online articles. The largest websites operating in various industries use them to gain a strategic advantage over the competition and fight to maintain their position on the market. AI-based solutions are able to improve the user experience with a given website, and for its owners it is associated with increased engagement, better CTR ratios or higher sales.
In this article, we look at the machine learning-based recommendation systems of the largest online services and streaming platforms, and we consider whether the solutions they use can be used in online stores.
Netflix is a trendsetter of modern recommendation models. The platform implemented the original version of its recommendation system in the year 2000. In 2006, Netflix organized the "Netflix Prize" - a competition to select the best data filtering algorithm with a top prize of $ 1 million.
Currently, Netflix uses an advanced system of personalized recommendations based on 4 complex algorithms that take into account both descriptions and target age ranges of movies, as well as ratings and popularity among users. Generated recommendations are arranged in several rows sorted by their categories and user preferences. Netflix estimates it gets around 80% of total watch time thanks to its recommendation system, which is an impressive result. Moreover, the platform is constantly focused on enhancing the user experience to improve the retention rate, which will translate into savings in acquiring new customers (approx. $ 1 billion annually since the year 2016).
It is estimated that around 35% of Amazon's revenue is derived from its recommendation system. It is considered to be one of the most popular and effective recommendation systems on the market. Its success lies in the multitude of recommendation models and their clever use in various store tabs. Models such as "other customers also bought" using up-selling strategies are great on the shopping basket page, and "similar" or "popular" models located on the product page have a positive impact on engagement. Amazon applies filter grouping algorithms that are scaled to huge datasets and capable of generating high-quality, real-time recommendations. Thanks to the great optimization of the entire system, Amazon achieves very high returns on investment, and customers have greater chances to discover interesting products that they would not otherwise notice.
Many Spotify users are impressed with how well the Discover This Week playlists fit their music tastes. Every Monday, the platform generates 30 new song suggestions for millions of its users. Playlists are created in a fully automated manner by a recommendation system based on artificial intelligence algorithms. Most of the personalized content on the Spotify platform is handled by the proprietary "Bandits for Recommendations as Treatments" system or simply "BaRT". It consists of 3 main algorithms found in other systems: collaborative filtering, the natural language processing model, and the audio path analysis model. All 3 work together to analyze descriptions, songs and the behavior of other like-minded users in order to provide the most relevant recommendations.
Taking into account that Spotify users have listened to over 2.3 billion hours of music from "Discover Weekly" playlists since the beginning of its operation, it can be concluded that those algorithms are extremely effective in recommending new music. We wrote more about the secret of Spotify's recommendation engine success on our blog.
Tik Tok is a relatively new social networking platform that allows its users to upload creative 60-second videos. Two things set this platform apart from others - a much higher engagement rate and one of the best recommendation systems on the market. The user does not need to search or know in advance what they want to watch. A personalized feed will provide content that will most likely interest them within one click. The algorithm analyzes real-time data about the videos themselves, as well as the preferences and viewing time of other users.
Tik Tok users typically spend 52 minutes a day on the application. By comparison, Snapchat, Instagram, and Facebook users spend 26 minutes, 29 minutes, and 37 minutes respectively. This clearly demonstrates how much advantage AI-based solutions offer.
Over 500 hours of videos have been uploaded to YouTube every minute since May 2019. Accessing millions of videos of all kinds, from fun few-second long cat videos to specialized tutorials, requires an effective system to help people choose what they are looking for. YouTube endorsements are powered by Google Brain, which recently launched as TensorFlow. This made it much easier for Google to train, test, and deploy AI-based solutions in a variety of environments.
YouTube's recommendation engine consists of two main neural networks. The first one is responsible for generating candidates. It analyzes millions of videos and narrows their collection down to just a few hundred depending on the user's viewing history, user inquiries, and other demographic characteristics. The second neural network is responsible for the next step - the ranking system. It takes into account data from the candidate generator system and other available data, such as the user's language, the language of the video, and the user's personal preferences. It then classifies the set of videos in order of the likelihood that the video will be interesting to a particular user. These suggestions are presented to users in real-time and are undoubtedly very effective.
Significance for the eCommerce industry
The significance of the solutions based on machine learning and artificial intelligence for the largest websites on the web clearly demonstrates that they are currently one of the most important trends in online activity. Content recommending systems are able to attract users' attention for longer, and websites that have implemented comprehensive recommendation engines have a significant advantage over the competition. For the constantly evolving eCommerce industry, solutions that increase website personalization based on machine learning are a very good way to develop and maximize profits.
The largest online platforms are increasingly using artificial intelligence to improve user experience. These solutions are surprisingly effective and can help the eCommerce industry as well.