Tuesday, July 14, 2020

Explaining TikTok’s Machine Learning Algorithm

tiktok algorithm

TikTok isn’t just home to “Renegade” dances—it’s also revolutionizing the world of machine learning through its proprietary algorithm

 

By Alvin Mak

 

TikTok continues to sweep across the world as a young adult social media phenomenon, and it’s showing no signs of stopping any time soon.

 

In January 2020, 19% of people between the age of 13-35 years used the ByteDance-owned app. In a mere three months, that number increased to 27%. Unique visitors to the app rose by a whopping 30.1% from January to March, and 28.8 million unique people are now active on the platform.

 

Along with its Chinese counterpart Douyin, TikTok crossed 2 billion downloads as of May 2020. Having achieved the milestone just five months after it hit 1.5 billion downloads, the app is now part of an elite class of social media giants such as Facebook and Instagram.

 

The COVID-19 pandemic also accelerated the app’s growth, as schools shifted to online learning and students practiced social-distancing. 12.2 million new American users downloaded the app in March 2020 alone.

 

If ByteDance’s flagship app is generating such staggering numbers, it must be doing something right.

 

What is TikTok?

 

TikTok’s official website calls the app “the leading destination for short form mobile video.” It features full screen videos that are between 5 and 60 seconds long. Anyone with a TikTok account can upload a video, and new videos can be discovered on TikTok’s feed simply by swiping up.

 

From lip-syncing videos, where users dance or lip-sync to popular songs, to video memes backed by music, TikTok now stands at the helm of creating user-generated video content trends.

 

With its ‘Duet’ and ‘React’ features, videos can also be remixed by other users on the platform. Duet, for example, places the original video and the ‘Duetter’ side by side, letting creators put their own unique spin on viral TikToks.

 

Engagement on the platform has been so voracious that creators are getting worldwide recognition for their presence on the app, creating new social media celebrities.

 

Most notably, the name D’Amelio has suddenly become well-known thanks to two TikTok accounts that have achieved monumental fame. The youngest of two daughters, Charli D’Amelio, has an estimated net worth of anywhere from $400,000 to $4 million] at just 15 years of age.

 

The TikTok cultural phenomenon has also attracted the attention of mainstream celebrities like Jimmy Fallon, Shaquille O’Neal, and Kevin Hart, in addition to American major sports franchises such as the NHL and the NBA, who are looking to expand their social media reach.

 

TikTok’s Key Differentiator

 

TikTok averages 46 minutes of total daily screen time per user, beating comparative figures for YouTube and competing closely with the likes of Facebook and Instagram. The secret behind TikTok clocking in a high daily usage figure is their use of algorithmic personalization.

 

TikTok’s main page is segmented into two, allowing users to move between the ‘Following’, and ‘For You’ pages. The ‘Following’ page is the classic self-curated content feed based on the creators you follow. Content order is determined by the publication time of each video–the newer the video, the higher its place on the feed.

 

Conversely, the ‘For You’ page is where users surrender most content curation abilities to TikTok’s algorithm. The videos located in this feed are estimated to be chosen based on several factors such as view time, completion rate (how many users watch the entire video), trending hashtags and sounds, users’ locations, and users’ saved videos.

 

All of these indexes are combined into TikTok’s algorithm, determining which videos will finally be displayed on users’ feeds.

 

Algorithms are the cornerstones for app technology, and are in no way unique to TikTok. Instagram uses algorithmic personalization in its ‘Home’ and ‘Search & Explore‘ pages. ‘Search & Explore’ is similar to TikTok’s algorithm, in that it displays posts based on past engagement.

 

Instagram’s main ‘Home’ page is still dominated by a user-curated selection mechanism, ultimately determined by the user and who they follow, with an algorithm controlling viewing order.

 

YouTube’s algorithm, on the other hand, controls much of its ‘Home’ page and constantly recommends content based on a user’s viewing history, whereas Facebook customizes viewing order in its home feed as well.

 

However, what makes TikTok’s personalization system stand out is its ability to feature an endless algorithm-curated stream of videos on one page and keep users scrolling, despite not offering any previews of the content.

 

When a video ends, the only way of knowing what the next video will be is by swiping into it directly, since TikTok does not show previews or allow for an expanded view of home page content. By the time a user scrolls to the next video, it has already started playing, giving it a better opportunity to hook the user’s attention.

 

TikTok discovered that the act of thinking about what to watch next is too much effort, and unnecessarily risks the user disengaging from the platform. In other words, users won’t switch out of the app if they don’t have time to even consider it.

 

This ingenious concept behind the ‘For You’ page has paid off, and has now become a flagship feature of the platform. While users prefer to stick to self-curated content pages on Instagram and Facebook, they are strangely complacent when using TikTok, preferring to remain under the influence of the algorithm.

 

Transformation of Content

 

Creators on the platform have adapted to this new cross-platform algorithm domination, and beating the algorithm has become the goal as they strive to leverage TikTok’s automated customization mechanisms for increased viewer counts.

 

Similar to how YouTubers started creating longer videos after recognizing that the algorithm rewarded longer watch times, TikTok videos feature more often on the ‘For You’ page when people have watched them multiple times, and completed the entire video.

 

Subsequently, creators started putting punchlines of memes at the very end their videos to keep users watching, thereby encouraging an increase in both the completion rate and the number of views.

 

In the era of Covid-19, TikTok’s machine learning has found a new purpose in promoting useful resources during the pandemic. Its algorithm is able to recognize Covid-19 related hashtags, and responds by including a link to useful verified Covid-19 information next to the video.

 

Big Data

 

TikTok’s influence on machine learning is a comparatively small yet important driving force in the arena of cloud computing.

 

Its ability to make complex and high-speed computations is largely owed to new-age cloud computing, made infinitely faster by Big Data cloud services such as Amazon Web Services (AWS) which are free from a mobile phone’s hardware limitations.

 

Data is gathered using cookies, on-platform surveys, and user behavior, stored and processed on the cloud, and returned as a content stream that is curated for users’ most recent interests. With the power of cloud computing, the algorithm continuously computes each user’s preferences and learns how to optimizes the feed in real-time.

 

TikTok’s ‘For You’ page is just one small example of the sheer amount of data that is collected and rapidly processed for user optimization. This amount of information could never be analyzed without the help of cloud computing services, let alone analyzed in real-time.

 

The huge success of algorithmically personalized feeds effectively demonstrates the potential of Big Data. This trend of user optimization through data processing on cloud data services is expected to generate new revenue flows and more marketing opportunities.

 

Further, the Big Data industry will begin to see increased demand coming from social media content personalization requirements, making cloud services such as AWS, Microsoft Azure, and Google Cloud even more powerful.

 

Processing Hardware

 

TikTok’s demonstration of the success of algorithmic personalization also opens the door to further expansion of the market for intense processing capabilities. Computing hardware manufacturers such as Intel and Nvidia will undoubtedly continue to see demand for their computing hardware.

 

It is this superior simultaneous processing power delivered by manufacturers such as Nvidia that is driving the world of cloud computing and ultimately making AI and machine learning viable. What was once a distant dream from science fiction, is now closer than one would think.

 

From online ads and personalized feeds, to self-driving cars, much of our recent automation owes itself to cloud computing and AI. Algorithmic personalization has made cloud computing more important than ever, and TikTok has placed dancing teenagers at the forefront of these monumental steps in innovation.

 

Header image by Amanda Vick on Unsplash

Email This Post Email This Post