YouTube’s Recommendation Algorithm
At YouTube’s “beta” launch in May 2005, the platform started off around 30,000 site visitors per day. Only half a year later with its official launch in December 2005, YouTube was reeling in two million video views a day. Only a month after that in January 2006, the daily views jumped up to 25 million views, and just a year after its initial launch, the platform was garnering over 100 million videos being posted per day. YouTube’s massive growth brought about a myriad of issues which led to Google’s purchase of the platform.
Since its inception, YouTube has become a major platform where creators from anywhere in the world can share videos. YouTubers’ ability to monetize their content and the spread of social media influence throughout the decade has also made it possible for content creation to become a career today.
Because of YouTube’s massive popularity and ability to provide income to creators, the search algorithm and the recommendation algorithm have been dissected and analyzed. Alongside this, YouTube’s goal is to keep users engaged for as long as possible as they make money from the ads that users come across, so the more time users spend on the site, the more income the platform can generate.
YouTube’s recommendation algorithm is where the majority of users find the videos they will ultimately watch rather than searching up a topic they are specifically looking for. In 2016, YouTube implemented a recommended algorithm that employed the use of deep neural networks. The system works in two main networks. The first of the two networks generates video candidates that the user would potentially be interested based on their watch history and videos that other similar users have watched. Even just from this network, the video recommendations are incredibly accurate. The ranking network is then used to determine the order in which the videos will be displayed to the user. Similar to PageRank algorithm, content is assigned a score and the content with the highest score is presented to the user first. During the ranking system, the video is analyzed much more thoroughly than the candidate generation system where they are only analyzing hundreds of videos as opposed to billions. During the ranking system, everything is analyzed down to the thumbnail to gauge users’ interest.
https://www.britannica.com/topic/YouTube
https://tinuiti.com/blog/paid-social/youtube-algorithm/
https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/45530.pdf