Post by Regolyth on Jul 20, 2015 12:18:37 GMT -5
Here’s How Facebook’s News Feed Actually Works
At Facebook headquarters in California, about 20 engineers and data scientists meet every Tuesday in the “John Quincy Adding Machine” room—“Abraham Linksys” and “Dwight DVD Eisenhower” are nearby. They’re tasked with assessing the billions of likes, comments and clicks Facebook users make each day to divine ways to make us like, comment and click more. In Knoxville, a group of 30 contract workers sit in a room full of desktop computers, getting paid to surf Facebook. They are tasked with scrolling through their News Feeds to assess how well the site places stories relative to their personal preferences. Their assessments, as well as ratings from about 700 other reviewers around the United States, are later fed back to the team in California, all in the service of improving Facebook’s News Feed algorithm, the software that delivers personalized streams of content.
In its earliest days, Facebook was essentially a directory of profile pages. Users could list their favorite bands, post pictures or write on each others’ profiles, but these activities were mostly discrete. As the social network grew, Facebook engineers noticed that some people were navigating the site in unexpected ways. Every user had access to a page showing when all their friends had last made a change to their profiles. A growing number of people began bouncing from this page to different users’ profiles to figure out what their friends were up to. “Users are usually pretty lazy. They’re not really willing to jump through a lot of hoops to do most things,” says Ari Steinberg, an early Facebook engineer and former manager of the News Feed team who now runs a travel startup. He and others at Facebook realized they needed to provide an easier solution.
The first iterations of the News Feed algorithm were pretty crude. Based largely on their own intuition about what people liked, engineers assigned point scores to different story formats (a photo might be worth 5 points, while joining a group was worth 1 point). Multiplying the score of the post type with the number of friends involved in the story would yield a general ranking order for posts. The formula might be tweaked based on emailed complaints from users or problems staffers saw in their own feeds. “We would just make all these arbitrary judgments,” recalls Steinberg.
As Facebook grew, News Feed became more flexible. Eventually the algorithm ranked content considering recency, post type and the relationship of the poster to the end user in a formula that came to be known as EdgeRank. The debut of the “Like” button in 2009, which let users endorse specific pieces of content for the first time, helped News Feed hone in even more on which stories people actually enjoyed.
In Karahalios’ 2013 study, which involved 40 subjects who were selected to mimic the demographics of the U.S. population, 62% of people didn’t know that their News Feeds were being filtered. When the algorithm was explained to one subject, she compared the revelation to the moment when Neo discovers the artificiality of The Matrix. “We got a lot of visceral responses to the discovery when they didn’t know,” Karahalios says. “A lot of people just spent literally five minutes being in shock.”
In its earliest days, Facebook was essentially a directory of profile pages. Users could list their favorite bands, post pictures or write on each others’ profiles, but these activities were mostly discrete. As the social network grew, Facebook engineers noticed that some people were navigating the site in unexpected ways. Every user had access to a page showing when all their friends had last made a change to their profiles. A growing number of people began bouncing from this page to different users’ profiles to figure out what their friends were up to. “Users are usually pretty lazy. They’re not really willing to jump through a lot of hoops to do most things,” says Ari Steinberg, an early Facebook engineer and former manager of the News Feed team who now runs a travel startup. He and others at Facebook realized they needed to provide an easier solution.
The first iterations of the News Feed algorithm were pretty crude. Based largely on their own intuition about what people liked, engineers assigned point scores to different story formats (a photo might be worth 5 points, while joining a group was worth 1 point). Multiplying the score of the post type with the number of friends involved in the story would yield a general ranking order for posts. The formula might be tweaked based on emailed complaints from users or problems staffers saw in their own feeds. “We would just make all these arbitrary judgments,” recalls Steinberg.
As Facebook grew, News Feed became more flexible. Eventually the algorithm ranked content considering recency, post type and the relationship of the poster to the end user in a formula that came to be known as EdgeRank. The debut of the “Like” button in 2009, which let users endorse specific pieces of content for the first time, helped News Feed hone in even more on which stories people actually enjoyed.
In Karahalios’ 2013 study, which involved 40 subjects who were selected to mimic the demographics of the U.S. population, 62% of people didn’t know that their News Feeds were being filtered. When the algorithm was explained to one subject, she compared the revelation to the moment when Neo discovers the artificiality of The Matrix. “We got a lot of visceral responses to the discovery when they didn’t know,” Karahalios says. “A lot of people just spent literally five minutes being in shock.”