Two months after appearing in front of the U.S. Senate, Facebook (FB) CEO Mark Zuckerberg finally turned in 229 pages of answers to the Senate’s unanswered questions. The CEO had promised to answer all the lawmakers’ questions at a later date, as the lawmakers’ time was limited.
Amid 76 pages of answers addressed to the 118 questions posed by Sen. Ted Cruz (R-Tex.), Facebook gave a very clear explanation of exactly how the News Feed prioritizes content. For a long time the News Feed promoted articles and videos from news outlets, and in January Facebook tweaked the News Feed to a model that favors content from family and friends.
In an answer to Cruz, Facebook said the News Feed is made up of “stories from their friends, pages they’ve chosen to follow and groups they’ve joined.”
“Ranking is the process we use to organize all of those stories so that users can see the most relevant content at the top, every time they open Facebook,” the company wrote.
Facebook identified four elements to the ranking. First, “the available inventory of stories” gives the algorithm a pool of stuff to choose from.
The next criterion, something Facebook calls “signals,” is not particularly clear. In the answers to Cruz and others, the company merely refers to it as “data points that can inform ranking decisions.”
Facebook elaborated on signals in a May 22 video, noting the data points into account could be “how old a given story is, or who posted it, or how fast your internet connection is, or what phone you’re on.”
The next point is simple: Facebook’s algorithm weighs how likely a person is to comment or share a story with a friend (“predictions”).
To Senators, Facebook didn’t elaborate on these criteria, but the company’s recent explainer video did explain these things. The “predictions” category looks at a user’s past behavior, how likely they are to report, share, like, and interact with a piece of content. Given that Facebook identifies interests and can measure the strength of connections between people and pages, these likely affect Facebook’s assumptions about what content they’d respond to and enjoy.
This is rolled up into a “relevancy score.” According to Facebook’s explainer, stories are directly ordered by those relevancy scores.
Absent from this list — “inventory,” “signals,” “predictions,” and “relevancy score” — is the criterion Cruz appeared to be looking for in his questions, which probed the company for a liberal bias.