Recently, Mastodon has added a recommended followers carousel into users’ feeds. This feature is the first time that many users noticed that Mastodon has some form of algorithmic discovery. This feature has recently been reworked to be a “friends of friends” recommendation model.
While many people have argued that discovery is necessary for the Fediverse to appeal to a wider user base, some are concerned with the power that recommendation systems have bestowed on the owners of social media platforms. Acknowledging these concerns, and to provide transparency, the Mastodon team has included descriptions that explain how follower suggestions are chosen.
While these descriptions are helpful, they deserve further explanation. The words “popular” and “similar” describe algorithms, or ways of using data to make decisions without human input. How does Mastodon decide who is similar to you or your friends? And who gets to be highlighted as popular?
In this blog I will go through each of these algorithms and explain them without code or math. Supporters of Mastodon claim that the open source nature of the platform prevents the downsides of algorithmic suggestions. Theoretically, anyone could look at the code and show how decisions are being made. So that’s exactly what we will do here!
How do I control MY account being recommended?
You can access the discovery features through the web portal for your instance. Go to Preferences, click on Public Profile, then click on the Privacy and reach tab. There you will see a Reach section with a checkbox to turn the discovery features on and off. If you uncheck this, you will not appear as a suggestion to other people (we will discuss later how this happens). This will not change whether you see suggestions, this only controls the visibility of your personal account.
What data is used to give me recommendations?
The data used in these suggestions is mostly from your follower and following lists. Your follower and following lists are generally public and accessible information, unless you have specifically suppressed this option. In order to know who is similar to or popular with your friends, your Mastodon server has to take a look at who your friends are.
Who will I see?
The only people who will factor into the algorithms are profiles that are searchable and have discovery turned on. Other people who have turned off visibility as described above, will not appear.
Who won’t I see?
This important question brings us to what is called the “base account scope”. A user who does not meet the “base account scope” should never be shown to you. This is how Mastodon uses your past choices to remove people who you don’t wish to see from your recommendations.
If a user is part of an instance that is defederated by yours, they will not be shown to you. You should also not see people that you have blocked, muted, silenced, are on a domain that you have personally blocked, or who you have previously rejected as a recommendation. It is clear that Mastodon is putting in effort to respect the personal choices that you have made on who you wish to see.
How do the algorithms work?
- Popular on your server
These suggestions come from anywhere on the Fediverse, as long as they match the base account scope. These users have the most followers and interactions from your local server, meaning other people on your home server think these are good accounts to follow. Follower count is just what it sounds like! Interactions are likes, replies, boosts, etc. Accounts that you see suggested to you under this heading might be recognizable as notable on the Fediverse, or in the world beyond.
- Popular with the people you follow
This is the most complicated algorithm. First, your Mastodon server reviews the list of people that you follow. It then takes each of those people and pulls in the people that they follow. This second group is combined into one large list. This second list will have duplicates: if two of your friends follow the same person, that person will be in the list twice. This list is then sorted by the profiles that appear the most. The more of your friends that follow the same profile, the more likely it is to be recommended. If there is a tie, the profile with fewer followers gets recommended.
- Handpicked by your server
These are profiles that your instance administrators want you to see! If you have any concerns about these profiles, your instance admin and mod team should be willing to have a conversation with you about why they think those people are a great follow.
- Similar to people you follow
This algorithm starts by looking at people you have recently followed. It takes their bios and uses their content as keywords to search for similar new profiles. When looking at the results of the search, Profiles that have verified links in their profile are prioritized for recommendation.
Here’s a handy infographic that describes each of the algorithms.
