TikTok is a short-form video app that has exploded in popularity since launching in 2016. The app was originally called Douyin and was created by ByteDance for the Chinese market. In 2017, ByteDance launched TikTok for markets outside of China. Since then, TikTok has seen tremendous growth across the globe.
According to Blogging Wizard, between Q3 2019 and Q2 2020, TikTok usage in the US grew by 123%. As of January 2021, TikTok was the 2nd most downloaded app worldwide. By August 2021, TikTok hit 1 billion monthly active users globally.
TikTok’s growth has been fueled by its addictive short videos and powerful recommendation algorithm that learns users’ interests. The app has become a global phenomenon and solidified itself as a major force in social media and culture.
What is the For You page?
The For You page is the main page users see when opening TikTok. It displays a constantly updating feed of recommended videos for each user. The For You page uses TikTok’s proprietary recommendation algorithm to select videos to show each user based on their interests, past viewing behavior, and other engagement signals.
TikTok’s powerful recommendation engine is the secret behind the addictive nature of the app. The algorithm analyzes a massive amount of data to figure out which videos are most likely to capture each user’s attention and entertain them. It considers factors like video topics, creators, captions, sounds, hashtags as well as the user’s interactions with content to improve video suggestions over time.
The goal of the For You page is to provide each user with a personalized, never-ending stream of engaging videos tailored just for them. The better TikTok’s algorithm gets to know a user through their viewing history and likes, the better it becomes at recommending relevant and interesting videos on their For You page.
TikTok’s Recommendation Algorithm
The core of the TikTok experience is the For You feed, powered by TikTok’s sophisticated recommendation algorithm. The algorithm analyzes each user’s behavior to determine their tastes and preferences. It registers metrics like videos watched, engagement, searches, and more. By studying endless signals, it builds interest profiles to serve each user a personalized feed of relevant, engaging content [1].
The algorithm is designed to quickly learn a new user’s preferences. It serves a diverse set of videos at first, registering all interactions to hone in on interests. The more a user interacts with certain video types, sounds, or creators, the more the algorithm refines its suggestions [2]. With time, it delivers a curated, hyper-targeted For You feed based on an ever-evolving understanding of each user.
Computer vision analysis
TikTok uses advanced computer vision techniques to analyze the actual video content uploaded by users. Computer vision is a field of artificial intelligence that trains computers to interpret and understand visual imagery. According to researchers at the University of Minnesota, TikTok uses computer vision algorithms to detect shapes, objects, faces, scenes, and actions within videos (https://cse.umn.edu/cs/news/jafarian-uses-tiktok-advance-computer-vision-and-machine-learning).
Specifically, TikTok’s computer vision systems can identify and track objects and people within a video. This allows it to understand what is happening in a scene and determine information like the number of people present, their ages, genders, and activities. The computer vision models can also recognize background details, lighting, camera motion, and other visual elements. By analyzing all this visual data, TikTok can categorize videos into different topics and genres in order to recommend relevant content to users.
Natural language processing
Natural language processing (NLP) is a key component of TikTok’s recommendation algorithm. The app analyzes the text, captions, comments, and other textual elements associated with videos to understand the content and sentiment. According to research by LinkedIn user Daigo Tanaka, videos on TikTok “get scored by quality” using NLP techniques.
Specifically, TikTok parses the text to extract keywords, named entities, and other semantic details. This allows it to categorize the topic and subject of the video. The NLP engine also conducts sentiment analysis to gauge the emotional tone of the text – whether it is positive, negative, or neutral. By combining textual semantics and sentiment, TikTok can better match videos to users’ interests and preferences.
The analysis of captions, comments, and other text expands TikTok’s understanding beyond just the visual content of videos. This allows its recommendation algorithm to make smarter connections and suggest videos that align with users’ tastes. As Daigo Tanaka summarized, the NLP engine enables the app to distribute videos to “matching user profiles.”
(Source)
Deep learning models
TikTok utilizes deep learning models to recommend personalized content that keeps users engaged. According to Towards Data Science, TikTok uses deep neural networks like LSTM (Long Short-Term Memory) to understand users’ preferences based on their viewing history and interactions [1]. TikTok has also developed its own deep learning recommendation system called For You that leverages multi-modal embeddings to match videos to individual users [2]. By analyzing signals like watches, likes, comments, and more, these AI models can discern user interests and recommend engaging, relevant content.
Personalized recommendations
TikTok’s recommendation algorithm is highly personalized for each user. According to an article by Content Petal, “The AI-driven algorithms that power TikTok’s personalized recommendations system can detect patterns in user behavior. This allows them to adjust the recommendations a user receives based on their interests and engagement” (https://contentpetal.com/how-tiktok-reads-your-mind-content-petal/).
The algorithm tracks many signals from each user to understand their preferences, including videos viewed, searches, likes, shares, comments, device and account settings, and more. As explained by The Verge, “TikTok takes into account a wide range of signals to deliver relevant content that keeps people engaged on the platform” (https://www.theverge.com/2021/12/16/22839453/tiktok-for-you-recommendations-harmful-content-fyp).
By analyzing all of this user data, TikTok can build an interest and preference profile for each user. It then uses this to serve up personalized recommendations catered to each individual on their For You feed. The goal is to keep users engaged by recommending content they are likely to enjoy and interact with.
Trend detection
TikTok’s algorithm is highly adept at detecting emerging trends and amplifying them through the For You feed. According to TikTok’s Creative Center, the recommendation system uses multiple signals to identify rising trends, including detecting when a specific sound, song, hashtag, or video effect is spiking in popularity.
The algorithm tracks not just how many people are using a hashtag or sound, but also the rate of growth. If a sound or hashtag begins getting used exponentially more, the algorithm recognizes it may be an emerging trend. TikTok also analyzes semantic meaning and connections between words and phrases to detect conceptual trends.
Once a trend is identified, the algorithm begins recommending related content more prominently to further fuel engagement. This creates a viral loop allowing trends to spread rapidly through the TikTok community. TikTok’s Trend Analysis tool provides creators with insights into trend history and growth to inform content strategy.
Creator amplification
TikTok helps promote popular creators through its recommendation algorithm. As a user engages with certain types of content, the algorithm learns their preferences and begins recommending similar videos. This helps expose creators to a larger audience who will likely enjoy their content.
TikTok also provides creators with analytics about their audience and tips on how to create more engaging videos. According to a job posting for Head of Community at TikTok Korea, the company has “various TikTok creator education programs and TikTok creator amplification programs” to help creators grow their audience.
Conclusion
In summary, TikTok’s recommendation algorithm and For You page are powered by sophisticated AI technologies like computer vision, natural language processing, and deep learning models. By analyzing the visual and audio content of videos as well as user engagement, the algorithm can deliver a highly personalized feed to each user. It detects trends and amplifies content from creators in a way that keeps users endlessly scrolling and entertained.
Looking ahead, TikTok will likely continue refining its AI to keep improving the recommendation engine. With the huge amount of data it has on user behavior, TikTok is in a prime position to stay on the cutting edge of AI development. This will allow it to better understand users’ interests and preferences to deliver the most relevant, engaging content for each individual. As TikTok expands further into areas like e-commerce, gaming, and augmented reality, its AI will also evolve to align with these future directions.
