Machine learning algorithms form the backbone of social media platforms. Keeping users hooked and presenting a curated slice for brands to leverage effectively. But how do they work?
Facebook, Twitter and Instagram are the top three most visited social media platforms after Google and YouTube. Social media platforms are a fertile ground for brands to advertise, engage with their audience, and add new consumers to their bucket.
But what really drives engagement over these platforms? The answer is content. An algorithm is a set of mathematical rules describing how data collection should behave. Social media algorithms help maintain order and rank search results and advertisements. For instance, an algorithm on Facebook directs the order in which the content is displayed.
Let us understand the algorithms working behind the top social media platforms.
Every time someone checks their Facebook feed, the posts they see and the order in which they appear are determined by the algorithm. Considering the volume of users on its platform, it is nearly impossible that Facebook’s ranking system relies on a single algorithm. Instead, the team has to apply multiple layers of machine learning (ML) models to predict the most relevant, meaningful and engaging content for its individual users. The Facebook algorithms follow four steps process, namely:
- Step 1: The step looks at the inventory – all the content available on the platform. It then considers the likes and dislikes, posts from friends, and groups one has joined.
- Step 2: Next, it moves to consider thousands of active and passive signals. Active signals are likes, comments, shares, and replies measure engagement. Passive signals include view time, the time of posting, internet connection, etc..
- Step 3: Then comes the prediction stage. Here, the algorithm uses thousands of signals to determine the engagement rate with a particular post. Say, for example, the likelihood of a user commenting, how much time they would engage with the post if the user watches the entire video, etc.
- Step 4: In the final step, the algorithm will assign a relevancy score to a particular post, i.e., how much interest people will show to a particular post. Once the relevancy score is determined, the content is distributed.
Twitter offers its users two different timelines: Home Timeline or Latest Tweets. A real-time chronological timeline of Tweets from the people you follow is displayed in Latest Tweets. Home moves posts into what it thinks is a better order using its ranking algorithm. Twitter’s ranking algorithm uses multiple ranking signals based on:
- Recency: Depends on how popular a topic is, rather than looking for popular topics on a daily basis or for a while.
- Relevancy: Users’ engagement history, from the posts they like, and topics they follow, to even the trends going viral at their location.
- Engagement: This one takes care of the popular tweets, how people in one’s network react to those tweets, to like, retweet, etc.
- Rich media: Depending on users’ interaction, images, videos, GIFs and polls get into the feeds.
The truth is that although the platform uses machine learning, not even Twitter is certain of the exact results of its algorithms. As part of its “responsible machine learning initiative,” Twitter is now investigating the outcomes of its algorithms.
As Instagram describes it, there are multiple algorithms at play — each one for a different part of the platform, whether it’s feed posts or Reels. The algorithms take the users’ preferences, termed signals, to create content that helps users to remain hooked to the app. Hence, thousands of signals across four critical aspects (in descending order of importance) come together for a personalised view, which includes,
- About the post: This one considers how well-liked a post is, as well as less important details about the content itself, such as when it was posted, length of a video, and location.
- About the person who posts: This step allows the algorithm to get a sense of how interesting a person can be for a particular user, which includes signals like how frequently individuals have communicated with that person over the past several weeks or how well they have responded to their content.
- Users’ activity: The task here is to find a particular user’s likes and dislikes, and understand the posts they like the most depending on what they share.
- Users’ interaction history: This helps the platform understand the users general level of interest in reading posts from a certain person. Take, for example, whether they leave comments on one other’s posts.
Beyond the interest score, various other variables may affect where a post appears in a user’s home feed. These include the quality of the photo or video, the originality of the post, the violation of Instagram’s community guidelines, or if the content is reported.
The TikTok algorithm makes personalised content recommendations for each of its users. Based on the specific tastes of each user, it chooses which videos to show them on their page. The algorithms take several signals into account, which include,
- Behavioural signals of users: TikTok takes many factors into consideration, but the most crucial one is how users act on the platform. With the help of data such as video likes and shares, comments made, accounts followed, and the content created – the platform can better comprehend each user’s distinct tastes, and how they evolve over time. Based on how closely these behavioural signals match with the films, it ranks the clips.
- Content discoverability: TikTok learns how to rank the content by using information from the video itself, such as captions, hashtags, and sounds. This is mostly because it uses the data to determine what the content is about, and when to display it on users’ pages.
- Account indicators: TikTok analyses each user’s device and account settings to maximise performance. It primarily considers country choices, preferred languages, device kind, category choice, etc. However, they don’t carry as much weight as the other two criteria because people don’t express them as preferences.
LinkedIn has its own algorithm that can connect users, identify jobs, surface relevant material, and surface any other information a user would be looking for on the platform. The LinkedIn algorithm prefers relevance over recency. Its objective is to present readers with the information they are interested in, before more current updates. This enables LinkedIn to show users the articles, videos, jobs, and other content that matches their search criteria or are most likely to find interesting. The signals the LinkedIn platform uses include:
- Connections: The algorithms consider connections as a reference to identify users’ tastes and preferences. More connections on LinkedIn equals more possibilities to rub shoulders with other professionals digitally. This allows people outside one’s connections to see their posts and vice-versa. Similarly, the LinkedIn algorithm considers the users one has interacted with through comments, likes, shares, and responses.
- What is the content: The algorithm used by LinkedIn determines how pertinent a user’s content is to your interests. It “reads” the hashtag, post, page, and profile kinds users have interacted with most frequently. Again, based on users’ prior online activity, LinkedIn may even provide users with content unrelated to their current connections.
- Engagement: The likelihood that a user will interact with further content is assessed by the LinkedIn algorithm using recent likes, shares, comments, Connection requests, and other engagement indicators.
To sum up,
An algorithm’s job is to provide users with pertinent content. It is used to filter through the vast amount of content available on each social media network. On the flip side, there is a call for large social media giants to shift towards ‘Explainable ML models’ for justifiable and transparent decision-making. What’s your take?