In 13 days, Twitter will release the source code used by the service to make recommendations.
A large social media service like Twitter faces a significant challenge in what posts it shows to users, given the incredible volume of content on the service. In the early days, this problem was easy, show users every post in reverse chronological order and let the user scroll through the list.
When you think about the possible ways to present a feed of content to users, this is absolutely not the best way to achieve a good user experience or engagement on the service.
Imagine you don’t launch the Twitter app for about 12 hours. Since you last logged in, there have been 5,000 posts by the people you follow. Twitter could show you the most recent, through to the oldest, but that certainly leaves you scrolling through many posts that will not be important or interesting to you.
A better way is for services like Twitter to create an algorithm that identifies what content topics you engage with the most, that is taking likes, retweets, and replies as indicators of engagement. Twitter could also use the items you search for as an indicator, along with the people you follow and the content they post.
Now here’s the hard part, how do you rank this content to show the best content to the user? Similar to the challenges Google has with its search ranking algorithm, social networks have also experimented with different approaches over the years.
In a perfect world, we’d only follow accounts that are posting relevant content 100% of the time, but the reality is that many great accounts post about a range of content and there’s no way to follow only a subset of topics posted by that account.
Twitter’s owner Elon Musk says they will release all code used for recommended Tweets on March 31st (US time), however, it’s not clear if that includes the code used to determine which advertisements you see, if it does, we’ll likely learn a lot about the demographic/interest/location slicing that occurs internally.
Musk goes on to say that the code will likely be embarrassing, obviously written at a time well before he purchased the company. If you’re a developer, this will be a fun day to take some to read through the code (assuming functions are at least commented).
It is important to note that this is not open-sourcing all the Twitter source code, which is important so as to not expose any vulnerabilities, should they exist.
