How ByteDance’s TikTok is using Machine Learning to learn and build a huge dataset of human behaviour.

Sure, TikTok is fun. Ask any ‘TikToker’ and they all say the same thing “the app is amazing”. Why? Because that provides many great features for both the creators and the consumers. Studies told us that 2hr/day is the minimum average for anyone using TikTok, that’s way more than any other available social media app.

Why do people like this app is related to many factors and I’m going to focus here on one of them which is, the apps use of cleaver machine learning algorithm(s). How TikTok uses ML algorithms to provide tools for creators and content for consumers.

What is TikTok?

If you’re living under a rock till now, let me tell you about TikTok briefly. It is a video sharing app where anyone can create small 15 second video clip for others to enjoy. The app was made by a Chinese company ByteDance and their majority of apps are somewhat related to one thing and that’s providing content using machine learning algorithms. They bought musical.ly and combined it with their ML enabled app Douyin and we had TikTok.

How Machine learning plays a vital role in TikTok?

The TikTok app is surely addictive. Time flies when you’re scrolling through this app (saying from personal experience). The app doesn’t require you do any setup like choosing interests or something, instead as soon as you download and install this app, you can start scrolling through content and it’s where the magic begins.

As soon as you’re welcomed with first video clip (that was just a random guess by ML algorithm working under TikTok servers), the algorithm tries to calculate your interest on that video. The weighting or scoring factor does depend on some parameters like –

• How much time you spent watching that video clip.
• What type of comment(if any) you left on the clip.
• Like interaction. Do you liked the music or the video.
• How fast or slow you scrolled through the video clip.
• Did you clicked on the user profile, if yes then which type of his/her other videos you’ve watched.
• Time you use the app most. Is it on day time or night time.
• How much time you used the app.

The list is very long and I’m certain that there must be some other pin point details that TikTok captures about the consumer per video clip.

With the help of ML, maybe within 5 minutes or less, you have a more engaging video clip thread than before and this cycle continues till you find yourself almost giving full time to all the videos or sometimes watching them again and again.

Ok. So you may ask, what’s the big deal? Majority of apps(specially in social media apps) are using ML algorithms to train their model to suggest more and more useful content for the consumer right?

Hmm, Yes sort of. YouTube, Netflix, Twitter, Instagram, Facebook and many others are using ML algorithms for the same, but what separates TikTok from them is the blind faith on the user to choose whether the user likes the content or not without taking his/her consent before, like these other apps do. TikTok evolve in real time by learning with your every interaction while other related apps just don’t.

What TikTok is achieving from this?

For starters, they are gaining a lots (I mean literally a lots) of data about human behaviour (majorly teens or young people). How they interact, what type of content they like etc and in the same time TikTok’s model is evolving about understanding human behaviour, emotions and engagement with faster rate and with less errors.

TikTok uses many filters(or whatever those features are called) that can guess your age, ask you relevant question to answer, tell your nationality etc. These features are making their model more robust and that’s for free. They are getting better and better in this and I’m sure they will keep on adding features like this to make their model well trained and error free.

What future holds for TikTok and ByteDance?

As ByteDance main research areas include Natural Language Processing, Machine Learning, Computer Vision, Speech and Audio, Knowledge and Data Mining, Distributed System and Networking, and Computer Graphics, I’m sure the future for them is bright. Their model is getting trained with real world data fast enough, people are spending more and more time scrolling through TikTok video feed and getting more creative.

I’m sure ByteDance was using this data from TikTok and data from their other apps to train their other related models too and that’s totally fine. ByteDance is also focused on global growth and as it’s operating on a second largest economy in the world (China) without any competition, its certain that they are here for long run.

Although, Google was considering purchasing another short-form video app called Firework, demonstrating deep interest for the short-form video format TikTok has popularized on a global scale, so it’s possible that ByteDance may get a real competition in future, and Google is already doing many great things in AI and ML segments.

Let’s see, where this take us.