TikTok and its Back-End Algorithms
It’s surely acceptable to say that anyone who has ever used the social media app known as TikTok has experienced the inexplicable moment when a video that’s eerily similar to the conversation you were just having shows up on your feed. I hate to burst your fantastical and conspiratorial bubble, but no, there is not a secret FBI agent peering through the other side of your smartphone’s camera or listening to you through the built-in microphone. Contrary to these imaginative speculations, TikTok is not listening or watching us, it is reading us; every action we perform on the app, whether its liking a video, commenting, sharing, rewatching a video, interacting with other comments, navigating to the video creator’s page, searching a hashtag in the search bar, or swiping to the next video before even five seconds have elapsed, is recorded and interpreted by the platform’s software. TikTok is not a person, it does not listen or watch, it simply interprets our actions as if they are one-to-one data points. The app then uses this information to make predictions about who we are as users and what we want to see. To understand this process and how TikTok’s algorithm is able to interpret user data and utilize it to tailor one’s content to their individualistic interests and preferences, I conducted a thoughtful research study. This particular study spanned a total of 4 days and required that I mindfully view 30 videos a day on TikTok’s platform. Typically TikTok, like many other social media platforms, is not the type of content that we approach with the intention of decoding the hidden messages and patterns behind what we view. Instead, TikTok is the type of platform that we engage with when we want to mindlessly scroll through entertaining content that can distract us from reality. However, in order to understand the back-end mechanics of a machine, one must think like the machine. Thus, for this project, I had to approach my content as if each interaction with the content I was shown, as well as the analytics of each video, was an objective, quantifiable marker of who I am as a person. Though this sounds dehumanizing, this linear mode of interpretation is exactly how an algorithm is able to articulate its subjects and determine their preferences, interests, and hobbies. With this notion in mind, I was able to effectively examine the topical, temporal, and various other patterns that spoke to my personal practices online and reveal how the TikTok algorithm understands me as a person, as well as interpret the complexities behind the conclusions that the app drew about who I am as a quantifiable object of data points. Ultimately, after extensive research, I was able to identify 3 notable patterns in the content presented to me on the app’s “For You Page”. In general, the content that the algorithm consistently presented to me spanned across a total of 6 overarching topics which I deemed “Sports”, “Beauty”, “Lifestyle”, “Relationships”, “Nutrition/Recipes”, and “Animals”. However, by the fourth day of research, very clear patterns began to emerge, revealing how my practices within the app led the app to develop a conceptualization of who I was as a person- and let me say, it was scarily accurate.
Algorithmic Pattern 1
The first of the three pronounced algorithmic changes could be categorized as a topical content shift. Let me preface this information by stating that I am not new to TikTok, therefore, the app already has quite a bit of data collected about my preferences. However, I am not a habitual user so the amount of screen time spent solely on the app over the past 4 days was quite an increase from my typical usage, leading to drastic shifts in my usual content. Because of this previous usage, TikTok’s algorithm had already delineated that I enjoy sports-related content, and it is not wrong. I thoroughly enjoy sports including college sports, professional sports, football content, basketball videos, running videos, soccer content, etc. The algorithm knows this because I have liked and interacted with many sports videos in the past. However, based on the first day of data collection, the app had not yet discovered the complexities behind my interest in sports content including which sport I enjoy the most and what level of sports I most identify with. On the first day of research, I was shown a total of 7 sports-related videos- the majority of which were about collegiate basketball and NFL football teams, as can be seen in the images below.
In contrast, only 2 of the videos were about soccer, which, as a college soccer player myself, are the videos that I particularly favor. Out of habit, I was quick to double-tap and therefore liked the 2 college soccer videos while I watched the full basketball and football videos before liking them. At the moment, I didn’t intentionally act this way to skew the algorithm, but it is clear that this is the exact effect my mindless actions had because, by the fourth day of research, the sports content I was shown, drastically narrowed to not just soccer-related content but college soccer content. Some of these collegiate soccer videos I was shown were even pulled directly from my own teammates’ pages which I follow within the app- hence the tag which reads “Your Friend” that is circled in the image below.
By day 4, after liking multiple college soccer videos from the previous days, it seemed the app’s algorithm had even further zeroed in on subtopics within the realm of college soccer which matched my profile with unimaginable accuracy. This particular soccer season with Coastal Carolina Women’s Soccer has been particularly difficult for two specific reasons (1) I have spent the past 7 months recovering from a severe knee injury and (2) those who can play have struggled in terms of performance resulting in a losing record. According to TikTok’s algorithmic interpretation of my user tendencies, these two reasons are incredibly predictable as well as ironically humorous because, on the fourth day of research, I was presented with videos about both knee injury rehab and poor conference performance.
Algorithmic Pattern 2
The second interesting pattern that unraveled by the end of the research period was a substantial change in the demographic characteristics of the creators. By the fourth day of data collection, it was clear that TikTok was curating a feed dominated by users it thought were a demographic match to me including their age, race, gender, and education. In fact, by the final day, the creators whose videos were appearing on my “For You Page” even remotely resembled attributes of myself down to their physical features and relationship status. Approximately halfway through the research period, I began to notice a shift from a mixture of both male and female creators to predominately female creators. For example, on the first day, 17 of 26 total videos (excluding 4 advertisement videos), or 65% were posted by female creators. In stark contrast, by the fourth and final day, 21 of 24 total videos (excluding 6 advertisement videos), or 88% of videos shown were posted by female creators. By assessing these statistics, it is clear that the algorithm had determined my gender was likely female and therefore I would prefer content created by females. Furthermore, of these 17 female creators who appeared on my “For You Page” during the first day of research, I vividly recall one in particular who I resonated with. This creator, known on TikTok by her handle @alixearle, is very similar to me in terms of demographics- she is also a blonde-haired, blue-eyed college student in her early twenties who often openly posts about the relationship she is in, as well as beauty-related content including makeup tutorials and “get ready with me” videos as well as videos about her fitness routines, etc. The striking similarities in appearance and hobbies initially intrigued me and kept me watching for the entirety of the 3-minute video. Following that first day, @alixearle became a frequent creator on my “For You Page” and was 1 of the 30 videos I viewed for the next 3 research days, and continues to appear daily on my TikTok feed.
However, Alix was not the only creator with visible demographic similarities that began infiltrating my TikTok feed. On the fourth day, of the 88% of my feed dominated by female creators, nearly every single one shared multiple demographic characteristics with me including physical appearance, age, education level, relationship status, and even their fashion sense which typically consisted of athleisure clothing.
At this point in the data collection process, it was obvious that the app’s algorithmic software had deciphered the demographic profile I generally belonged to. But this shift in demographic patterns didn’t end with the typical and routine content shown on my “For You Page”, even the advertisements began to change based on the profile TikTok had created for me based on my actions within the app. On the first day of research, 4 of 30 videos had been advertisements. Of these 4 advertisements, each one was sponsored by a multimillion-dollar company and consisted of professionally produced media, such as this Apple ad.
However, by the fourth day, each of the 6 advertisements was instead user-generated content (UGC), meaning they were videos created by micro-influencers within the app featuring a company’s product(s) rather than created by the company themselves. In these instances, the creators used to film these sponsored ads, not surprisingly also aligned with each of the demographic characteristics that TikTok had clearly assigned to me in terms of who exactly TikTok thinks I am as a person.
Based on this clear pattern of demographic changes, it is reasonable to claim that TikTik’s algorithm has assumed a relationship between me and the creators pictured above. It is not correct to assume that these patterns assert that TikTok is stereotyping me as a human; As I stated previously, a machine cannot think through the nature of such complexities. The app’s software is simply assessing me as a set of data points rather than a living, breathing, human being. I likely watched the videos of creators who demographically aligned with me for longer periods of time, and liked them at a faster rate, maybe I interacted with the comments of such videos because they subconsciously intrigued me. These small details of my practices within the app may go undetectable to me, but not to the software. TikTok converts these practices to data, interprets it, then pulls future content to align with the characteristics that these videos share which, in this case, are the demographic characteristics of the users.
Algorithmic Pattern 3
By the final day of data collection, I noticed a drastic change in the temporal patterns of the content that made up my feed. By the final day of research, it seemed that the app’s algorithm, which curates a custom feed based on my preferences, had adjusted to cultivate a consistent flow of videos much shorter in length than the videos shown in the first half of the research period. I have a particularly short attention span, so if a video does not grab my attention within the first few seconds, I tend to skip past it to the next video. This tendency of mine was quickly detected by the TikTok algorithm resulting in a much different feed by day 4. For instance, on the second day of research, 18 of 26 videos (excluding 4 advertisement videos), or 69% were 60 seconds or more. Whereas, on day 4 of research, only 5 of 24 videos (excluding 6 advertisement videos), or 21% were 60 seconds while the remaining 79% were 30 seconds or less. One of the 6 overarching topics where this change was particularly noticeable was in nutrition-related and recipe videos. On the second day of research, a 2-minute video describing a dinner recipe appeared on my “For You Page”.
While this video is about a topic that I find entertaining, it was not particularly enjoyable for me to watch for 2 whole minutes. Eventually, I found myself scrolling on before the video finished. Comparatively, here is a similar video that appeared on my feed on the fourth day of research.
This recipe is condensed into just under 30 seconds. Because of the shortened timeframe of this piece of content, it was able to keep my attention and I didn’t find myself wanting to skip to the next video in my feed. From this example, among many other similar instances, I was able to understand the TikTok software’s initial difficulty in deciphering the complexities behind human interests. On the first day of data collection, TikTok’s algorithm had correctly identified that I enjoyed watching nutrition and food-related content but it had not yet figured out the complexity behind the interest which are my temporal preferences: I prefer a video with a much shorter time frame than one with an extended time frame. After noticing my incapability of finishing long-form videos, the algorithm began curating a feed with the same overarching topics while also adhering to my preference for short-form videos.
Ultimately, after conducting this study and analyzing the data gathered, it is undeniable that TikTok and many other social media software with similar back-end algorithms have very sophisticated and intelligent software that can predict our interests and preferences nearly as well as we can ourselves. The topical, temporal, and demographic pattern change that I noticed in just a short span of 4 days and a total of 120 videos, illustrate the app’s ability to quickly interpret our seemingly meaningless and often subconscious tendencies within the app as data points and use them to curate a sometimes scarily accurate and specific feed. So, the next time you open your TikTok and see a video about the food you were just talking about going to eat, don’t be concerned for your privacy, just your predictability.
Works Cited
Alix Earle (@alixearle) “It’s a no from me honey” TiktTok. October 21, 2022. https://www.tiktok.com/@alixearle/video/7156772494240730410?is_copy_url=1&is_from_webapp=v1&lang=en
Alix Earle (@alixearle) “They’re a little scandalous though so not for trick or treating with the fam” TiktTok. October 22, 2022. https://www.tiktok.com/@alixearle/video/7157022172307819819?is_copy_url=1&is_from_webapp=v1&lang=en
Alix Earle (@alixearle) “30 second makeup/no makeup look 🤍 #grwm #umiami #makeupfinds #amazon” TiktTok. October 11, 2022. https://www.tiktok.com/@alixearle/video/7153334626168130862?is_copy_url=1&is_from_webapp=v1&lang=en
Alix Earle (@alixearle) “#stitch with @balancedmakena rawr” TikTok. October 24, 2022. https://www.tiktok.com/@alixearle/video/7157825193036057902?is_copy_url=1&is_from_webapp=v1&lang=en
Briley (@briley.mb) TikTok. October 12, 2022. https://www.tiktok.com/@briley.mb/video/7153810371952905514?is_copy_url=1&is_from_webapp=v1&item_id=7153810371952905514&lang=en&q=briley.mb&t=1666885187184
Briley (@briley.mb) TikTok. October 2, 2022. https://www.tiktok.com/@briley.mb/video/7149763356470332718?is_copy_url=1&is_from_webapp=v1&item_id=7149763356470332718&lang=en
Cassie (@cassafrass86) “Dinner on the fly! These have been a go to for a quick dinner for us the last couple years. Soooo yummy and so easy to make! #chickenbaconranch #quickdinner #sliders #dinneronthefly #rotisseriechicken #alfredo #fyp #yummy #easydinner #kingshawaiian @kingshawaiian @hiddenvalleyranch” TikTok. October 14, 2022. https://www.tiktok.com/@cassafrass86/video/7154565379996618027?is_copy_url=1&is_from_webapp=v1&lang=en
Char (@char_harr) “the fact this is not an ad and i have this much passion for this lip product #fyp #lipcombo #lip #tarte #tartecosmetics #sephora” TikTok. October 13, 2022. https://www.tiktok.com/@char_harr/video/7154135278989413674?is_copy_url=1&is_from_webapp=v1&item_id=7154135278989413674&lang=en&q=char_harr&t=1666886166549
Chiefs (@chiefs) “i’m a photographer can i take ur picture?? #photography #imaphotographer” TikTok. October 23, 2022. https://www.tiktok.com/@chiefs/video/7157397513278950698?is_copy_url=1&is_from_webapp=v1&item_id=7157397513278950698&lang=en
Doosi (@doosifit) “Crispy buffalo chicken rolls for fat loss 😳 #eggroll #caloriedeficit #weightloss” TikTok. September 25, 2022. https://www.tiktok.com/@doosifit/video/7147435007471111467?is_copy_url=1&is_from_webapp=v1&lang=en
Duke Basketball (@dukembb) “He really pulled up wirh a 🎷🎷🎷🤔🤫😅” TikTok. October 24, 2022. https://www.tiktok.com/@dukembb/video/7157916615097552174?is_copy_url=1&is_from_webapp=v1&lang=en
ESPNW (@espnw) “#greenscreen Maroons head coach Julianne Sitch is breaking barriers AND posting results 😤👏 #uchicago #soccer #maroons #ThatsaW” TikTok. October 12, 2022. https://www.tiktok.com/@espnw/video/7153699934695345451?is_copy_url=1&is_from_webapp=v1&lang=en
G (@gworkk) “Back by popular demand #food #d1athlete #macros #cut #GenshinTeleport #OLAFLEX” TikTok. October 12, 2022. https://www.tiktok.com/@gworkk/video/7153479271149735211?is_copy_url=1&is_from_webapp=v1&lang=en
Jordy (@jordyshawrty) TikTok. October 22, 2022. https://www.tiktok.com/@jordyshawrty/video/7157143281770564907?is_copy_url=1&is_from_webapp=v1&item_id=7157143281770564907&lang=en
Katie Schmidt (@katie_schmidtyyyy) “I could make so many of these #paris #eiffeltower #europe” TikTok. October 6, 2022. https://www.tiktok.com/@katie_schmidtyyyy/video/7151291804313488686?is_copy_url=1&is_from_webapp=v1&lang=en
Maggie Williams (@maggie.williams61) “#ad bright skin begins with skincare! #skintok #targetstyle #lorealdermtalks #lorealvitaminc @target @L’Oréal Paris” TikTok. September 22, 2022. https://www.tiktok.com/@maggie.williams61/video/7146367109595925803?is_copy_url=1&is_from_webapp=v1&lang=en
Marcus Williams (@musclenmotiv8) “I became an athletic truth group coach because the solutions are REAL! #faith #fitness #personaltrainer” TikTok. October 20,2022. https://www.tiktok.com/@musclenmotiv8/video/7156274747997752618?is_copy_url=1&is_from_webapp=v1&lang=en
Rocio Fernandez (@rocii_fr) “she a real one🤍@mego9 #greenscreen #foryoupage #foryou #fyp #soccer #d1” TikTok. October 24, 2022. https://www.tiktok.com/@rocii_fr/video/7157797724224556294?is_copy_url=1&is_from_webapp=v1&lang=en
Vic (@vic_kyr) “Home = on the field playing ❤️🩹 #acl #aclrecovery #aclsurgery #athlete #injury #soccer #injuredathlete #d1 #d1soccer #soccergirl #injuryrecovery #ncaa #mentalhealth #mentalhealthmatters #athletementalhealth” TikTok. October 21, 2022. https://www.tiktok.com/@vic_kyr/video/7156775468975377710?is_copy_url=1&is_from_webapp=v1&item_id=7156775468975377710&lang=en
