Algorithms often get a bad rap. And for good reason too, because algorithms, being human-made, are biased and in fact, reinforce biases. So algorithms aren’t necessarily better than humans, even though they exist through computers and power social media tools and other technology.
And algorithms have other real-life consequences. For instance, the Apple Card was found to be biased against women, Amazon screened resumes and was also biased against women, police use predictive software with facial recognition technology that discriminates by race, judges give out harsher sanctions in criminal trials according to race, Google Translate too has gender problems, and perhaps most relevantly, social media algorithms are used by politicians and their election campaigns to target voters with advertising.
And the problem isn’t going away. In the midst of the current pandemic, the American Bar Exam is being conducted remotely and the facial recognition technology that is used means a Black woman has to shine a bright light on her face for two days in order for it to work. HEH?!?!
Let’s be clear. The problem is not the math. The problem lies in the unconscious biases that are in the data that’s fed into the algorithm, which then produces biased results. So these algorithms are a form of Artificial Intelligence (more here).
Part of the problem is that there simply isn’t the diversity, inclusion, equity, or justice needed for people who are not white, not male, and not middle-class to be in the room creating these algorithms, and I don’t think it’s good enough to “de-bias” existing algorithms either. A little leaven leaveneth the whole bread and all that.
In any case, it’s still important to understand how the algorithm works. For most users of social media tools, it can help you understand exactly how tech companies use algorithms to create their profits through advertising. And for marketers, it can help you understand exactly how to use the algorithms to your advantage. And for everyone, I hope we can all learn to use social media tools with integrity.
All of these algorithms are subject to change, and curiously, people who work at these companies often say they don’t actually know how these algorithms work. And, all of these algorithms are designed to keep your attention so as to get you watching as many ads as possible, and thus make money. Which also means inflammatory content often does well. In any case, the most important thing to understand about algorithms is that they rely not just on compelling content, but on you creating community.
The News Feed is not chronological unless you change the setting, but rather is based on:
- Likes
- Comments
- Shares
Every piece of content in the News Feed is ranked according to:
- Content available
- Content itself
- Person itself
And then given a score.
For each post, aim to achieve:
- Engagement (likes, reactions, comments, shares)
- Link sharing (through Messenger and other apps)
- Shared posts on personal profiles
- Multiple replies to a comment (with lots of writing)
Facebook prioritises posts from friends over posts from pages (aka. businesses), which is problematic for unpaid posts from business pages (aka. organic reach) and social media marketers.
Where Facebook is a popularity contest, Instagram is all about the individual. Instagram actually has a few algorithms, but they’re all separate. My top tips are to use as many hashtags as possible, use all the different features and mix it up, post to Stories every day, and have a play with the newest features. Just don’t be spammy.
The Explore feed (go to the search bar, where you’ll see a mix of Reels, IGTV, Video and photos) is based on:
- Individual preferences
- What you do
- Posts by non-followers
- Non-trendy posts
- Artificial Intelligence
The Hashtag feed (type in any hashtag into the search bar, which will come up with Top Posts, which is where most of your new audience will find you, and Recent Posts) is based on:
- Individual activity and preferences
- Followers and non-followers
- Timing
- Saturation (ie. how many other people are also posting at the same time)
The Stories feed (the bubbles at the top of the feed when you open the app; also when you scroll through the Home feed) is based on:
- Personal interactions are prioritised
- Lives are prioritised
- New Stories are prioritised over already watched
- Stickers gain more interactions and views
The Home feed is based on:
- Individual interests and activity
- Recent content
- Popularity
- Others’ activity
The IGTV feed (consisting of For You, Following, and Popular) is based on:
- Who you follow
- Your interests
- Popularity
- Others’ activity
- Artificial Intelligence
The Reels feed (though unconfirmed, but similar to TikTok; also when you scroll through the Home feed) is based on:
- Who you follow
- Your interests/activity
- Your location
Note that Reels, as a new Instagram feature, is highly prioritised by all the algorithms, and it seems like it’ll be a new tab in the app, along with the Shop tab (along the bottom). So it seems likely that Instagram wants its users to use Reels, even if there’s no advertising capability in it just yet.
Twitter has two feeds, a Latest Tweets (chronological) feed and a Top Tweets feed, though even if you switch to the chronological feed, it will still give you top news through the “For You” in search and the “In Case You Missed It” in notifications. The feed is less algorithmically based than Facebook or Instagram, but still relies on indicators of:
- Recency
- Relevance
- Engagement
- Rich media (photo, video, link preview)
- Followers
- Location
TIKTOK
TikTok is probably the most problematic, in that the TikTok algorithm doesn’t even use what your individual preferences are, but just gives you content and then works out your preferences. It’s For You page uses:
- User interactions
- Video info
- Device and account settings
And then all of this data is individually weighted so it is completely unique to you. More info has just been released here (updated 24 October 2020).
SNAPCHAT
Snapchat does use an algorithm, though it is shrouded in secrecy. It’s based on the feed on the left of the camera of “friends”, “best friends” and “friends according to emojis” (see examples here). In essence, it looks at:
- Your interactions with a friend over the last 7 days
- Group chat participation
- Types of interactions (hence types of friends as marked by emojis)
Snapchat, unlike pretty much every other social media tool we’re looking at here, also divides between private publishers and professional publishers. The pro feed (to the right of the camera) is based on:
- All premium publishers
- Pro social media stars you follow
- Aggregated stories from search
- Snap Map in the Discover section
- Curated by humans
- Sorted by past viewing behaviour
LinkedIn often flies under the radar a little, but it’s got an algorithm too! It prioritises personal connections based around interests. LinkedIn is quite complicated, as it uses many factors to predict how relevant your content will be to your audience and then rank it accordingly, with particular emphasis on a smaller audience first. When you post, LinkedIn filters by:
- Spam
- Low quality
- High quality
Once you’ve got likes, comments, and shares, LinkedIn will share your post more broadly. There’re lots of details here, but it’s all about:
- Personal connections
- Interest relevance
- Engagement probability
YOUTUBE
The YouTube algorithm can be complicated and there are lots of unknowns, and even more so in 2020 with the pandemic when YouTube has rolled out things like “Breaking News”, “Music at Home”, and “Premieres”. But the YouTube algorithm decides what people watch on YouTube 70% of the time, so you should probably get to know how it works. The algorithm wants to find the right video for each viewer, and get viewers to keep watching, so it takes note of viewer behaviour as well as video performance.
The algorithm is used in search results and recommendation streams. Search results are based on:
- Your video’s metadata (title, description, keywords) and how well those match the user’s query
- Your video’s engagement (likes, comments, watch time)
Recommendation streams are based on:
- Performance analytics data (impressions vs views, aka thumbnail; watch time vs retention; how many likes, dislikes, comments or shares a video gets, aka engagement; view velocity, rate of growth; how new a video is; how often a channel uploads new videos; session time)
- Personalisation through watch history and what similar people have watched (which channels and topics have you watched in the past, what have you engaged with in the past, how much time do you spend watching, how many times has this video already been surfaced for you, what don’t you watch)
The algorithm also affects:
- Your YouTube homepage
- Trending videos
- Your subscriptions
- Your notifications
Suggested reading: