Caught in an algorithm
From CS lectures, math tutorials, food bloggers, study tutorials, latest internet gossip, to movies, YouTube has it all. Therefore, it is no surprise that over 2.5 billion people access YouTube once a month and in 2021 Youtube was valued at $28.8 billion dollars in revenue. But what has kept YouTube at the top since its launch in 2005? It is its algorithm, built off of deep neural network techniques. A research article titled “Deep Neural Networks for YouTube Recommendations” written in 2016 by Paul Covington, Jay Adams, and Emre Sargin employees of Google, covers the foundation of the current Youtube algorithm. While the algorithm is more complex as it focuses on personalization, noise, and several other factors, it still relates to the principles that we learned in INFO 2040; specifically, strongly connected components & hubs, and authorities.
Based on the article by Covington, Adams, and Sargin, the YouTube algorithm that they proposed focused on three major aspects: Scale, Freshness, and Noise. Utilizing the idea of strongly connected components, the proposed YouTube algorithm takes in the user’s recent YouTube activity, to retrieve a small subset (hundreds) of videos from a large corpus by leveraging a “rich set of features describing the video and user.” By utilizing these key features they can find the videos that the users are interested in, rank them, and the strongest connected components (those with the highest rank) are shown to the user. It is what also drives you to stay on YouTube for hours upon hours, mostly watching for entertainment purposes.
https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/45530.pdf