EdgeRank Is Dead: Facebook’s News Feed Algorithm Now Has Close To 100K Weight Factors by Matt McGee (Edited my me)
EdgeRank, Facebook’s original News Feed ranking system, is dead.
Facebook hasn’t used the word internally for about two-and-a-half years. That’s when the company began employing a more complex ranking algorithm based on machine learning. The current News Feed algorithm doesn’t have a catchy name, but it’s clear from talking to the company’s engineers that EdgeRank is a thing of the past.
During a phone call this week, Lars Backstrom, Engineering Manager for News Feed Ranking at Facebook, estimated that there are as many as “100,000 individual weights in the model that produces News Feed.” The three original EdgeRank elements — Affinity, Weight and Time Decay — are still factors in News Feed ranking, but “other things are equally important,” he says.
In other words, the News Feed algorithm of today is much more sophisticated than just a couple years ago.
“The easiest analogy is to search engines and how they rank web pages,” Backstrom says. “It’s like comparing the Google of today with Alta Vista. Both Google and Bing have a lot of new signals, like personalization, that they use. It’s more sophisticated than the early days of search, when the words on a page were the most important thing.”
This has implications for marketers and business owners far beyond the wording used to describe News Feed rankings. It’s a reflection — and a cause, too — of today’s complex battle to reach Facebook users organically.
The winners? They’ll be the ones who understand how Facebook has moved past Affinity, Weight and Time Decay, and move past it themselves.
But now that job is a lot more complicated than ever.
Consider that there are more than a billion people using Facebook each month. And 128 million in the U.S. that use Facebook every day. They’re using dozens of different mobile devices with different capabilities for displaying content. There are 18 million Pages, many of which are actively looking for attention and a way to show up the News Feed as often as possible. And that number doesn’t include the numerous businesses that are using Facebook via regular accounts rather than Pages.
With all of that going on, Facebook says that the typical user has about 1,500 stories that couldshow in the News Feed on every visit.
So how does Facebook decide what users see, and what content from Facebook Pages make it into the News Feed? As you can imagine, Facebook isn’t about to give away all the details, but Backstrom did talk openly about several ways that the algorithm has grown up in recent years.
Affinity, Weight & Time Decay - These are “still important,” Backstrom says, but there are now multiple weight levels. “There are a lot of different facets. We have categories and sub-categories of affinity.”
Facebook is attempting to measure how close each user is to friends and Pages, but that measurement isn’t just based on personal interactions. Backstrom says Facebook looks at global interactions, too, and those can outweigh personal interactions if the signal is strong enough.
“For example, if we show an update to 100 users, but only a couple of them interact with it, we may not show it in your News Feed. But if a lot of people are interacting with it, we might decide to show it to you, too.”
Relationship Settings Another factor is the relationship settings that Facebook users can apply. With each friend, you can go a step further and label the person a “close friend” or “acquaintance.” With liked Pages, users can choose to “Get notifications” or “Receive updates,” and there are deeper settings to control what kind of content the user wants to see.
“We try to extract affinity naturally,” Backstrom says, “but if you go to the trouble to tell us more about your relationships, we will factor that in.”
Post Types The News Feed algorithm takes into account the type of posts that each user tends to like. Users that often interact with photo posts are more likely to see more photo posts in the News Feed, and users that tend to click more on links will see more posts with links.
Backstrom says this is also applied on a deeper level. “It’s not just about global interactions. We also look at what types of posts you interact with the most from each friend.”
In other words, Facebook Page owners that continually publish one type of post are likely not having those posts seen by fans that interact with other types of posts.
Hide Post / Spam Reporting News Feed visibility can also be impacted by users’ ability to hide posts or mark them as spam. But it’s not as simple as having a set threshold that will cause posts to stop showing in users’ News Feeds.
“For every story, we do the same computation,” Backstrom explains. “Given this story, and given the user’s history, what’s the probability that you’ll like this story? What’s the probably that you’ll hide it? We’re looking at this and trying to decide, is it a net positive to show this story in the News Feed?”
Further, Backstrom says there’s an element of decay when considering posts that have been hidden. Recent “hides” may carry more weight when deciding if a post shows in the News Feed, but those “hides” will have less impact as they decay over time.
Clicking On Ads, Viewing Other Timelines The News Feed algorithm is completely separate from the algorithm that decides what ads to show, when to show ads, and where to show them. But how a user interacts with Facebook ads can influence what shows in the News Feed.
“Nothing is off the table when we’re looking at what we should show users,” Backstrom says. “It can be clicking on ads or looking at other timelines. It doesn’t have to be just what the user interacts with in the News Feed.”
Device & Technical Considerations Yep, the News Feed algorithm even considers what device is being used and things like the speed of a user’s internet connection when deciding what to show.
“The technical limitations of some old feature phones make it impossible to show some content,” Backstrom. “We also know that some content doesn’t perform as well with Facebook users on certain devices. And if the user has a slow internet connection, we may show more text updates. We’re trying to show users content that they’ll find interesting and want to interact with.”
Story Bumping & Last Actor
Don’t forget these two changes that Facebook just announced last week. Story Bumping bends the “decay” rules by giving older, unseen posts a second chance at News Feed visibility if they’re still getting interaction.
Last Actor puts a premium on recency. Facebook is tracking a user’s most recent 50 interactions and giving them more weight when deciding what to show in the News Feed. This works on a rolling basis, so the value of an interaction will decline after the user has made 50 more recent interactions.
It should be clear that Facebook’s News Feed algorithm has developed significantly over the past few years. EdgeRank is a thing of the past, and it’s been replaced by a machine learning-based algorithm that, as Backstrom says, “only ever gets more complicated.”
That poses new challenges for brands and marketers hoping to get attention on Facebook, but the company says its advice to Page owners and others is the same: Create and publish and a variety of interesting content that will attract shares, comments, likes and clicks. That requires understanding your Facebook fans — from the types of posts they interact with to the different devices they might be using when they’re on Facebook.