YouTube is undoubtedly the world’s leading video sharing social platform, with 3.9 billion videos, 2.49 billion active users and 100 million subscribers. (as of 2024)
Recommendations drive a significant amount of the overall viewership on YouTube, even more than channel subscriptions or search. The success of YouTube’s recommendations depends on accurately predicting the videos you want to watch.
The paper titled “Deep Neural Networks for YouTube Recommendations” provides significant insights into the internal implementation of YouTube’s recommendation logic, which might be particularly interesting to advanced data science learners.
The system is comprised of two neural networks: one for candidate generation and one for ranking.
The candidate generation network takes events from the user’s YouTube activity history as input and retrieves a small subset (hundreds) of videos from a large corpus. These candidates are intended to be generally relevant to the user with high precision. The candidate generation network only provides broad personalization via collaborative filtering.
The ranking network presents a few “best” recommendations, by assigning a score to each video according to a desired objective function using a rich set of features describing the video and user. The highest scoring videos are presented to the user, ranked by their score.
Access the full text of paper here: Deep Neural Networks for YouTube Recommendations [Paul Covington, Jay Adams, Emre Sargin]
References:
https://blog.youtube/news-and-events/youtube-music-premium-100-million-subscribers/
https://blog.youtube/inside-youtube/on-youtubes-recommendation-system/
https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/45530.pdf