On YouTube’s recommendation system [Cristos Goodrow]
Cristos Goodrow, VP of Engineering at YouTube, published a blog post about YouTube’s recommender system that provides “a deeper look into how YouTube’s recommendation system works”. It includes a general introduction on what recommender systems are, how they are used on YouTube, and an overview of what information YouTube uses for the recommender system. Cristos also devotes a good part of the blog post to talking about bias and filter bubbles, and how YouTube tries to prevent extreme content to be promoted through the recommender system.
Recommendations drive a significant amount of the overall viewership on YouTube, even more than channel subscriptions or search. […] When YouTube’s recommendations are at their best, they connect billions of people around the world to content that uniquely inspires, teaches, and entertains. […] Clicks, views, watchtime, user surveys, shares, likes and dislikes work great for driving recommendations for topics like music and entertainment—what most people come to YouTube to watch. But over the years, a growing number of viewers have come to YouTube for news and information. Whether it’s the latest breaking news or complex scientific studies, these topics are where the quality of information and context matter most. Someone may report that they’re very satisfied by videos that claim “the Earth is flat,” but that doesn’t mean we want to recommend this type of low-quality content.
https://blog.youtube/inside-youtube/on-youtubes-recommendation-system/
Case Study on how VistaPrint uses Amazon Personalize for product recommendations
Netflix’s Journey: Early Challenges and the Absence of Recommender Systems in Marc Randolph’s Vision
Spotify considers the interests of artists when recommending music to users
About The Author
Joeran Beel
I am the founder of Recommender-Systems.com and head of the Intelligent Systems Group (ISG) at the University of Siegen, Germany https://isg.beel.org. We conduct research in recommender-systems (RecSys), personalization and information retrieval (IR) as well as on automated machine learning (AutoML), meta-learning and algorithm selection. Domains we are particularly interested in include smart places, eHealth, manufacturing (industry 4.0), mobility, visual computing, and digital libraries. We founded or maintain, among others, LensKit-Auto, Darwin & Goliath, Mr. DLib, and Docear, each with thousand of users; we contributed to TensorFlow, JabRef and others; and we developed the first prototypes of automated recommender systems (AutoSurprise and Auto-CaseRec) and Federated Meta Learning (FMLearn Server and Client).