The "In" on TikTok's Infamous Algorithm

As the short-video streaming app "TikTok" has become increasingly popular, inquiry about its algorithm is equally infamous. The "for you page," similar to an explore page on Instagram, is a feed of new videos unique to each user. The for you page or fyp exposes users to new content based on the type of content that the user seems to enjoy. While some are amazed by the individualized curation of their "for you page," others have concerns about how these recommendations come about. These concerns range from questions about privacy to speculations about algorithmic biases or prejudices of their own. Many people feel that on the right day, their for you page is perfectly tailored to their humor and interests. Although most people feel satisfied having an app specializing in entertaining you exactly how you want to be entertained, several of us are left wondering every day, how does this app know me so well?

TikTok employs a recommendation system to suggest content to users after gathering information about what content the user likes (TikTok, 2019). The app pulls this information through three different sources: user interactions, video information, and personal and device information. User interactions include which accounts the user follows, the videos they tend to "like," "share," and leave comments on, and the content the user creates themselves. Video information would be filters, sounds, and effects used in the video and hashtags in the captions. Personal and device information casts a broader net; however, the language settings, location settings, and type of device you are using TikTok on still play a significant role in the content you see. The algorithm will consider all of these factors and weigh them depending on how much each element seemingly matters to the user. While TikTok doesn't disclose its specific algorithm, we know that TikTok's unique algorithm follows the general overview outlined above. We can delve deeper into recommendation systems to get a good idea of how their algorithm gets inside our minds.

As mentioned above, recommendation systems gather data about the user to make suggestions. The data collected can be split up into two categories, explicit interactions and implicit interactions (Chua, 2022). Likes, accounts followed, content shared, and personal information falls under the label of explicit interactions. Implicit interactions would be more along the lines of where they are using the app, the type of device, and the links clicked on. Then once the data is gathered, the recommendation engine stores and analyzes the content ("How Do Recommendation Engines Work? What are the Benefits?", 2022). The recommendation engine can find content with similar user data through the analysis process. Then comes the filtering. Filtering is where all the pieces are put together, and the engine goes through all the data received to make recommendations.

Filtering can occur in various ways based on the type of recommender system. The two most common forms of recommendation are content-based filtering and collaborative filtering (Gubareva & Yakovleva, 2021). Content-based filtering mainly pulls information from the user's profile and settings, aka, what the user says they like. The premise is straightforward; the engine recommends content similar to content the user seems to like. Collaborative filtering is slightly more complex. This method takes a deeper dive into the behavior of the user. Collaborative filtering suggests content based on the content the user interacts with and their preferences, and how similar they are to other users. TikTok and other Social Media platforms likely employ a hybrid engine that might combine two filtering systems and different algorithms.

Social media platforms have no interest in publicizing their algorithms. These companies are in competition with each other, but there is also a possibility that they don't want concerns raised over how their algorithms operate. The public has already raised concerns about these algorithms violating user privacy. With the increasing conversation about biases being present/promoted in these algorithms, TikTok and their competitors surely should be fearful of what will be discovered about their engines in the future.

---

Bibliography:

Chua, R. (2022). A simple way to explain the Recommendation Engine in AI. Retrieved 25 April 2022, from https://medium.com/voice-tech-podcast/a-simple-way-to-explain-the-recommendation-engine-in-ai-d1a609f59d97

Gubareva, O., & Yakovleva, V. (2021). Recommender Systems and The Way The Recommendation Algorithm Work. The World of Science and Innovation, 23–25. https://doi.org/https://sci-conf.com.ua/wp-content/uploads/2021/07/THE-WORLD-OF-SCIENCE-AND-INNOVATION-1-3.07.2021.pdf#page=23

How Do Recommendation Engines Work? What are the Benefits?. (2022). Retrieved 25 April 2022, from https://marutitech.com/recommendation-engine-benefits/

TikTok. (2019, August 16). How Tiktok recommends videos #ForYou. Newsroom. Retrieved April 24, 2022, from https://newsroom.tiktok.com/en-us/how-tiktok-recommends-videos-for-you