While Artificial intelligence (AI) is definitely entering a revolutionary period for the technology, with companies extensively focusing on incorporating it into their products and day-to-day activities, the technology itself has been around for a while now. In lieu of that, let’s take a look at a major implementation of AI that still holds current day importance. Particularly, let’s discuss Facebook’s News Feed and how algorithms deliver relevant information to users.
Facebook’s News Feed feature has been a hallmark of both the platform itself, as well as a cornerstone of social media in general. The idea of truly creating a social forum online by featuring content and updates from your friends and close ones was, if not original, incredibly well executed by the platform and to this day continues to be a staple of most social media apps. As Facebook progressed into the billion dollar empire it currently is, the company made changes to news feed where it would now also feature posts from publications and content creators, giving users more bang for their online buck. But how does Facebook choose what piece of content goes to whom? Well, the social network’s tech webpage recently shed some light onto the matter.
The post, penned by Akos Lada, Meihong Wang, and Tak Yan from Facebook’s team, explains that it’s a matter of the AI predicting what people will enjoy beforehand. Using the example of a fictional user named Juan, the post provides a fictional scenario. Since Juan logged in last night, a friend posted a dog picture, another friend posted a video of them running, a Page he liked posted about viewing the Milky Way at night, and a group he’s part of put out a fun cooking recipe. Accordingly, if Juan likes the video as opposed to the dog photo, the Facebook AI will give that a higher predictive score, making it more likely to appear. Then again, things aren’t always so clean cut. Juan may enjoy the Milky Way post more than the running video, or prefer the dog photo over the recipes. There are also multiple other ways of showing interest in a post, such as commenting, tagging friends, or sharing it forward. That is where multiple machine learning models are integrated.
A big challenge is not only accurately predicting the multiple factors that go into curating one’s News Feed, but also making that process efficient to cater to Facebook’s 2 billion users. Accordingly, this is what the AI does. First, it collects all posts that have been shared with Juan by friends, pages, and groups. Secondly, it curates fresher posts that may not have been shared with him yet, but have still received a high predictive score. Then, all of these posts are run parallel to each other into the multiple machines, thus working on a base score. Finally, certain scores pertaining to contextual diversity and personalization are added, producing a final score that either pushes the story forward, or ranks it behind others.
Read next: Facebook is giving academic researchers access to its political advertising data
Facebook’s News Feed feature has been a hallmark of both the platform itself, as well as a cornerstone of social media in general. The idea of truly creating a social forum online by featuring content and updates from your friends and close ones was, if not original, incredibly well executed by the platform and to this day continues to be a staple of most social media apps. As Facebook progressed into the billion dollar empire it currently is, the company made changes to news feed where it would now also feature posts from publications and content creators, giving users more bang for their online buck. But how does Facebook choose what piece of content goes to whom? Well, the social network’s tech webpage recently shed some light onto the matter.
The post, penned by Akos Lada, Meihong Wang, and Tak Yan from Facebook’s team, explains that it’s a matter of the AI predicting what people will enjoy beforehand. Using the example of a fictional user named Juan, the post provides a fictional scenario. Since Juan logged in last night, a friend posted a dog picture, another friend posted a video of them running, a Page he liked posted about viewing the Milky Way at night, and a group he’s part of put out a fun cooking recipe. Accordingly, if Juan likes the video as opposed to the dog photo, the Facebook AI will give that a higher predictive score, making it more likely to appear. Then again, things aren’t always so clean cut. Juan may enjoy the Milky Way post more than the running video, or prefer the dog photo over the recipes. There are also multiple other ways of showing interest in a post, such as commenting, tagging friends, or sharing it forward. That is where multiple machine learning models are integrated.
Today, we're describing the ML models which power the ranking system personalizing the News Feed of every person around the world who uses Facebook.
— Alexandru Voica 💀 (@alexvoica) January 26, 2021
Here's how ranking works: https://t.co/cpcWRdXj5q
And here is the engineering behind it: https://t.co/bAjOwAWNIT
🧵👇 pic.twitter.com/Yiq5AsRruG
A big challenge is not only accurately predicting the multiple factors that go into curating one’s News Feed, but also making that process efficient to cater to Facebook’s 2 billion users. Accordingly, this is what the AI does. First, it collects all posts that have been shared with Juan by friends, pages, and groups. Secondly, it curates fresher posts that may not have been shared with him yet, but have still received a high predictive score. Then, all of these posts are run parallel to each other into the multiple machines, thus working on a base score. Finally, certain scores pertaining to contextual diversity and personalization are added, producing a final score that either pushes the story forward, or ranks it behind others.
Read next: Facebook is giving academic researchers access to its political advertising data
