Part
01
of one
Part
01
What are interesting ways Netflix uses data science (including machine learning and AI) to manage its business?
Interesting ways that Netflix uses data science (including machine learning and AI) to manage its business include using algorithms to provide video recommendations, using AI to ensure quality streaming even at lower bandwidths, utilizing predictive modeling to optimize the quality control process for content, and conducting A/B testing process before introducing every change. The Meson system was also created to support the above workflows. Below, I will outline how Netflix uses data science to manage its business.
VIDEO RECOMMENDATION SYSTEM
Netflix uses algorithms to recommend videos to its users. The company estimated that this system helped to save $1 billion a year in value from customer retention. Data such as what users watch, when they watch, which recommendations the users did not choose, and the popularity of videos are fed into several algorithms powered by statistical and machine learning techniques. These algorithms produce recommendations such as the personalized video ranker that selects the order of videos in genre rows, the Top N ranker that selects videos in the Top Picks row, and also an algorithm that selects videos in the Trending Now row that uses short-term trends, such as the user’s interest in holiday movies.
Netflix recently changed its video rating system from a 5-star rating system to a thumb-based rating system (thumbs-up or thumbs-down). Although the five-star rating system was meant to be objective, many users rated videos “subjectively” by rating shows that are considered “great” such as Orange Is the New Black, poorly because they do not personally like the show or the genre. Thus moving to a thumb-based, fully subjective system was Netflix’s way of adapting to its users. Changing to a simpler binary input also reduces complexity and provides cleaner, more accurate data. Netflix claimed that user ratings increased by 200% since the change. Thus, more data could be fed into the system to produce more accurate recommendations.
QUALITY STREAMING AT LOWER BANDWIDTHS
Netflix has used AI to provide specialized video coding to deliver cleaner streams to users in regions with low-bandwidth. This tool is important as the company has expanded its services worldwide, including countries where internet speeds may be slower and thus potentially affecting streaming performances. The Dynamic Optimizer uses AI algorithms to review every video frame and compress it without degrading the quality of the image. This system also tailors content for users that watch the videos on other platforms such as tablets and phones, which is more common in India, South Korea, and Japan.
OPTIMIZING CONTENT QUALITY CONTROL
Besides using AI to maintain the quality of streaming, Netflix also uses predictive modeling to optimize the quality control (QC) process for content. Content refers to assets such as video, audio, and text (subtitle and closed captions). The predictive quality control model is trained by a supervised machine learning approach to predict if the content quality “fail” or “pass.” The company’s QC process included automated and manual inspections to identify and replace assets that do not meet its standards such as audio-video sync issues or poorly placed subtitles.
A/B TESTING FOR EVERY INNOVATION
Every innovation or change done by Netflix has undergone a rigorous A/B testing process before becoming the user’s default experience. Changes included the redesign of the app’s UI layout, new personalized homepage, and new products/videos. Instead of opinionated and vocal Netflix employees, A/B testing allowed Netflix to determine if the risks of making changes are justified by using actual data to guide its decisions. A/B testing is done by creating an experiment with a control group and one or more experimental groups which would receive alternative treatments. When the test goes live, specific metrics such as streaming hours and retention are tracked. When there is sufficient data, the results are studied to draw statistically meaningful conclusions as to whether the company should proceed with the proposed changes.
MANAGING MACHINE LEARNING WORKFLOWS
To support the creation of machine learning (ML) workflows and make efficient use of resources, Netflix created Meson, “a general purpose workflow orchestration and scheduling framework built to manage ML pipelines that execute workloads across heterogeneous systems.” This system manages the lifecycle of several ML pipelines, including those described above. One of the goals of creating Meson is to “increase the velocity, reliability, and repeatability of algorithmic experiments while allowing engineers to use the technology of their choice for each of the steps themselves.” Meson is a key tool for allowing Netflix to efficiently use data science (including ML and AI) to manage its business.
CONCLUSION
In conclusion, Netflix uses data science to provide video recommendations, ensure quality streaming even at lower bandwidths, optimize the quality control process for content, and conducting A/B testing process. Meson is a key tool for powering these workflows.