TikTok and Information Cascades
TikTok differentiates itself from other engagement algorithms as it doesn’t rely simply on active behavior and user networks in order to curate content. Rather, it follows a more decentralized recommendation network. In order to explain TikTok’s addictively entertaining algorithm and the novel nature of its structuring of social media, I will explore it in the context of the phenomena learned in class.
For those unfamiliar with TikTok, the majority of the user experience occurs on a “for you” page, a stream of highly curated content. As a user watches a video that appears on their feed, information about their interests and dislikes are critically analyzed and collected: how long they remain on the video, whether they pause or rewatch it, and how they engage with the content. Who a user follows has some effect on what they are recommended, but it is only one part of a larger set of information. A major aspect of TikTok’s algorithm is its approach to discovering engaging content- all videos will appear on the feed of a small sample group, and if it is popular among the sample set, will be promoted more. In this way, videos from users with little to no following can become an instant hit, resulting in a rise to fame seemingly out of nowhere.
Due to this, TikTok cannot be modeled as a typical network where certain users have more power depending on the number of neighboring edges; this is a major departure from more conventional algorithms in which large swathes of content go unnoticed, and its viewership is limited by the size of its user network. Instead, it follows a sequential information cascade in which TikTok makes the decision for its user with little knowledge of the users’ preference- it judges the probability that a user will enjoy a certain video by how much previous users did, and promotes them based on this information. This is an informational effect, as the fact that other people liked a video does not directly result in your own enjoyment of it; this only affects the probability of your own enjoyment. Because of this highly directional cascade, users will almost immediately be herded into a niche cluster of their preferences, where they are shown videos tailored entirely to their demonstrated interests.
https://towardsdatascience.com/why-tiktok-made-its-user-so-obsessive-the-ai-algorithm-that-got-you-hooked-7895bb1ab423