YouTube Algorithm and the Creation of Rabbit Holes
Source: https://techcrunch.com/2019/08/28/youtube-to-reduce-conspiracy-theory-recommendations-in-the-uk/
This article describes the ways in which YouTube’s recommendation algorithm leads users into a “rabbit hole”– that is, the more engaged the viewer is to a certain topic, the more YouTube recommends videos related to those topics to appear in the “Recommendations” sidebar for viewers to watch next. While this seems logical in order for YouTube to keep its audience engaged and wanting to watch more content on the website (as well as generate the most ad revenue), it is also a way for viewers to become swathed in misinformation and harmful content, such as videos propagating conspiracy theories. The algorithm awards clickbait, in which the most shocking, controversial, and often harmful content receives the most attention as more people choose to click on these videos, especially in digital spheres spreading misinformation and conspiracies.
For example, if a user on YouTube begins watching a political video leaning towards one political spectrum, the algorithm tends to recommend videos guiding the user towards increasingly politically-extreme videos on that side of the spectrum, until the user find themselves in a “rabbit hole,” in which most of the next recommended videos are supportive of extremist views. The article suggests that this is harmful, especially for younger impressionable viewers, who would be much better informed if they were also recommended videos that showed views from the other side of the political spectrum.
If we were to map out YouTube’s recommendation algorithm using a directed graph based off how the article describes it, we would see that many strongly connected components exist, in which every node in the subset has a path to each other, but each subset is not part of a larger set in which every node can reach each other. Each node in the subset represents a video, and each subset represents videos all related to a certain topic. These subsets also then branch off and lead to other strongly connected components, and so on, until we reach a strongly connected component in which all of its nodes are videos of extreme topics or radical views all related to each other. This illustrates how YouTube users can find themselves led into rabbit holes — the farther they stray away from more mainstream content, the more extreme and radical the videos become, with few ways to get back to the center of the graph, due to the properties of the strongly connected components of the directed graph.