Datawatch: How social media platforms keep us scrolling

Ever wondered what keeps us endlessly scrolling on platforms like Instagram, TikTok, and Twitter? Find out, in our latest instalment of Datawatch.

Social media plays an increasingly significant role in almost everybody’s lives – whether you’re a casual user who likes to stay in touch with friends and family, or a power user with thousands of followers. 

But these platforms aren’t just successful at getting people to download and sign up – they also find incredibly effective ways to keep people hooked and scrolling. Their secret? Data science. 

The power of the algorithm

While every social platform has the same goal of keeping its users engaged, each one has its own approach. Every platform has its own unique algorithm to help curate a near-perfect personalised feed to meet its individual users’ needs and interest – using millions of data points from every account. 

Instagram’s algorithm, while consistently changing based on the latest trends, is perhaps the most renowned for its effectiveness. The platform’s algorithm is based on a combination of who you follow, how long you spend on the platform, and how you react to content. Even if you and another user follow the exact same accounts, your personalised feeds will still be different. 

It might sound like a straightforward algorithm, but the further you dig, the deeper it gets. Instagram’s ‘Explore’ page – a page of curated content for users from accounts they don’t currently follow – uses complex machine learning models to select the right content for its users, including one powerful technique called ‘word embedding’.

Word embedding traditionally decides the order in which words appear in a text by measuring how connected they are. But in Instagram’s case, the same machine learning model is used to identify how connected images and videos are.  

Instagram’s word embedding model looks at seed accounts – accounts users have interacted with in the past – looks for accounts similar to them, and selects 500 pieces of content. These pieces of content are then filtered to remove spam and ranked by their probability of the user interacting with them. And even then, only the top 25 of these content pieces reaches the user’s Explore page.

It sounds like a long-winded process, but this machine learning model is running every second for every user of the platform, ingesting millions of data points to make it work effectively.

But how does Instagram actually compare and match content accurately if all it has to go off is images and videos?

Powerful, AI-driven image recognition

The answer is with well-trained, AI-powered image-recognition software that captures granular data from every post to inform Instagram’s word embedding model. 

Meta – the parent company of Instagram – uses its highly-powerful image-recognition software to identify unique qualities in content, such as the types of objects, colours, locations, people, and even facial expressions found in images and videos. 

Most of the time, this all happens behind the scenes to help curate the perfect feed for every user. But one time just a few years ago, everybody got a peek behind the curtain at how Instagram’s algorithm finds the right content for them. 

When Instagram briefly crashed in June 2019, Instagram users saw the image-recognition metadata that’s used to help curate their feeds, instead of the actual content itself. Users’ posts simply appeared as blue text in grey squares, reading:

• “Image may contain: One person smiling, close up”
• “Image may contain: Night, Sky, Outdoor”
• “Image may contain: Cat, Cup of Coffee, Sun”

And millions of other combinations. For many users, it was just a short, frustrating period where they couldn’t use the platform. But for those interested in the power of data science, it was a fascinating look at how Instagram knows the interests of its users.

Engagement monetised at every step

Of course, platforms like Instagram, Twitter, and TikTok aren’t going to these lengths just to provide high-quality entertainment to their users. They’re aiming to monetise their engagement, using dozens of methods to generate income.  

Both Instagram and TikTok now seamlessly insert highly-personalised ads in between their user-created posts, and they’re often indistinguishable from organic content.

TikTok has refined the use of ads on social platforms even more in recent years, offering brands access to a wide range of advertising tools. The most popular paid ads appear as in-feed videos, looking almost identical to organic content, with the only exception being their convenient call-to-action button. Other, more unique methods, include tools like branded effects such as 2D filters and 3D lenses – made available to all users, and spreading brand awareness through other user-created content.

Even simple techniques such as hashtags can help brands easily get into users’ curated content feeds. For example, John Deere, the US manufacturing company, doesn’t have a TikTok account, but the #JohnDeere hashtag on the app has more than 3 billion views.

These opportunities to monetise users’ scrolling have expanded even further over the past two years, with both Instagram and TikTok introducing dedicated shopping portals as part of their platforms. Now, when a user interacts with a product they see on one of these platforms, they often don’t even need to exit the app to make their purchase – they can make a swift and easy payment, and get back to their scrolling. 

Analytics at The Smart Cube

Social media platforms offer some of the best examples as to how brands can use data to understand every nuance of their customers’ behaviour. 

Here at The Smart Cube, we offer bespoke, end-to-end analytics capabilities, from data engineering through to machine learning and AI – to help equip organisations across every industry generate the intelligence they need.

To read about some of the ways we’re helping our clients, or learn how we can help you achieve your own business goals, visit our website.