Advertisement
|

Facebook Discusses How Feed Algorithm Works

  • 504
    SHARES
  • 1.6K
    READS
Roger Montti
Roger Montti SEJ STAFF
Facebook Discusses How Feed Algorithm Works

Adam Mosseri, the Head of Facebook News Feed, provided a video walk through of how their news feed algorithm works. I then compare what he said to a Facebook patent that lists him as an author and find an interesting takeaway that may help you get your Facebook posts seen by the most people.

Adam Mosseri described how Facebook uses a process called Ranking to “organize” which stories you’ll see in your news feed. That word organize is very important. According to Adam Mosseri, the average person has thousands of things to read in their news feed every day.  The following is a description of how Facebook organizes that information and most importantly, what that means to you.

The Facebook Algorithm has Four Parts

Facebook uses an algorithm in order to organize that information for their members. Their algorithm is divided into a series of four steps. The four steps are called:

  1. Inventory
  2. Signals
  3. Predictions
  4. Relevancy Scores

Rather than summarize what he said, here’s a transcript of the relevant parts of his video describing how Facebook’s news feed algorithm works:

“In Ranking we have names for these steps. The first, we call Inventory, what’s on the menu. The menu are the stories that you have not seen from your friends and from the publishers that you have decided to follow.

The second we call Signals. The information we have available to make informed decisions. Signals are things like how old a given story is and who posted it. Even little things like how fast your internet connection is right now or what kind of phone you’re on.

And we rely on signals and in a large way feedback from our communities to understand what kind of content might be problematic. Maybe it’s violent content or spam or clickbait or false news.

And we use all of these signals to try to make informed predictions, which is the third step.

Predictions are things like how likely are you to comment or share a given story. How likely are you to hide or report a story? And we take those few dozen predictions and we weight them and roll them up into a relevancy score. A number that represents how interested we think you are in any given story in your inventory.

And then we order stories by those scores. We do this for each and every person who uses Facebook, for every story that’s in their inventory, everytime they come to news feed.

And because Recency is an important signal, News Feed is lightly chronological but not strictly so.

And we know we’re not perfect and we know we make mistakes. Which is why we are constantly iterating and trying to get better at connecting people with the stories that matter to them most.”

Algorithm’s Determine What Facebook Users Want

Algorithms figure out what kinds of things member prefers to see.  Adam Mosseri referred to predictions because the algorithms Facebook users are called, predictive modeling. In the case of Facebook, they predict the likelihood of user satisfaction with a news feed content item.

Facebook Patent
Facebook has a patent filed for just such an algorithm called, Filtering Content in a Social Networking Service that provides insights into the news feed ranking process. The process is described like this:

“1. A method comprising: receiving content items from users of a social networking service; receiving, from the users of the social networking service, user profile attributes comprising descriptive information about each user;”

This means receiving your personal information such as who your friends are, things that you like, where you live, where you work and other information you’ve given to Facebook.

An interesting part is here:

“…automatically creating,… groups relevant to a user according to one or more rules, the groups comprising users of the social networking service having one or more user profile attributes in common with the user;

…sending for presentation to the user in the user interface an updated subset of content items for the content feed, the updated subset including content items received only from users in the at least one group of users associated with the selected filter”

That’s the part that means they take all of your information and match it up to others with similar information and use that to create a news feed. They then filter all your information and generate a news feed based on what your friends tended to like.

How Facebook Ranks Content

The patent describes the ranking part of the algorithm in terms of affinity scores. Affinity means a similarity that implies a relationship between you and something else. So in other words, an affinity score can be a score of how likely something is relevant to you.

And that’s what Adam Mosseri called it, a Relevancy Score. So this part here might be what Mosseri is referring to when he describes a Relevancy Score.

“2. The method of claim 1, wherein providing the user interface comprises: selecting the subset of content items in the content feed according to affinity scores determined between the user and the content items; and sending the selected subset of content items for display in the content feed.”

Then further down in the patent application it provides a summary of how that content is filtered, naming certain “affinities” between you and the content items.

“Content is filtered based on the attributes in a user’s profile, such as geographic location, employer, job type, age, music preferences, interests, or other attributes. “Newsfeed stories” may be generated to deliver the most relevant information to a user based on a ranking of the generated content, filtered by the user’s affinity, or attributes. Applying this algorithm, newsfeed stories present the user with the most recent and most relevant content available on the social networking service. “

Takeaway

I think the most interesting aspect of Facebook’s algorithm is what Adam Mosseri described as Recency. Recency means how recently something was posted, shared, commented on, liked and so on. How recent a piece of content is very important because Facebook does not show older content. It’s default is to show the most recent content.

That means that if users aren’t commenting on your content then it will begin to decay. All those comments and shares are like the air beneath your content’s wings. The more comments there are on it the higher your content will likely rise in your follower’s news feeds.

Anecdote About Facebook’s Feed Algorithm

Just an anecdote, but a few weeks ago I was re-reading a very long discussion I’d started a year ago on my own feed. I discovered a comment that was particularly insightful that had gone unacknowledged. So I liked their comment and replied to it, expressing my appreciation for the insight they shared. That comment and like made by me was enough to push that year old (and popular) discussion back into the news feeds of many of my friends. Almost immediately other friends started posting to that discussion all over again!

Time is of the Essence

So one of the takeaways here, given what’s said in the patent and by Adam Mosseri himself in the video, time is of the essence. If you want your Facebook postings to be seen by others, post content that will generate comments.

I have about fifteen years experience as a forum owner and a moderator at someone else’s community. And I’ve noticed that the best discussions are started by users who ask questions. They ask people for their opinions. Doing something simple such as asking your followers, “What do you think?” is a great way to get a conversation going.

Don’t just sit back and watch the conversation. Participate by praising, reacting and otherwise responding to those who discuss. And give the ball a push to keep the conversation in play by asking a follow up question that may have been raised. Explore the topic thoroughly, try to identify what those users like to discuss and push those buttons.

The video is titled, News Feed Ranking in Three Minutes Flat.  It can can be viewed here.

Screenshot by Author

ADVERTISEMENT
ADVERTISEMENT
Roger Montti

Roger Montti

Roger Montti is an SEO Consultant with nearly 20 years of experience. I specialize in Site Audits to help sites ... [Read full bio]

Advertisement