Our Approach to Facebook Feed Ranking

UPDATED

NOV 28, 2023

Facebook’s goal is to make sure you see posts from the people, interests, and ideas that you find valuable, whether that content comes from people you’re already connected to or from those you may not yet know. When you open Facebook and see Feed in your Home tab, you experience a mix of “connected content” (e.g., content from the people you’re friends with or are following, Groups you’ve joined, and Pages you’ve liked) as well as “recommended content” (e.g., content we think you’ll be interested in from those you may want to know). We also show you ads that are tailored to you.

Below you’ll find information about how we use ranking in Feed, focusing specifically on connected content. We’re now providing a deeper look at the types of signals and prediction models we use in this process to help you better understand the details around how our ranking systems work.

Why We Use Personalized Ranking

We personalize each Feed for our more than 2 billion users by using state-of-the-art machine learning systems to rank content. Because most people have more content in their Feed than they could possibly browse in one session, these ranking systems help ensure that people see the content that is most valuable to them. While many different factors affect the ordering of content in Feed, the information below will give you more insight into the types of predictions and signals that generally have the biggest impact on how our systems determine what you see.

How Feed Ranking Works for Connected Content

To determine the ranking of connected content in your Feed, the system works in four steps:

Inventory

First, we gather your recent inventory – all potential new posts, or posts with new activity, that you could see when you open Facebook. This includes all the posts shared by 1) the people you have connected to as “friends,” 2) the Pages you follow, and 3) the Groups you have joined, and excludes content flagged for violating our Community Standards.

Signals

Then, for each of these posts, we consider thousands of “signals” to make predictions about what you will find most interesting. Many of these signals are pieces of information you give us directly when you like or share a post, you connect with a friend or Group, or you comment on a Page’s post; others are inferred based on the actions you’ve taken on Facebook. We’ve shared more information about the types of signals used in ranking below.

Predictions

From there, we use these signals to make a series of personalized predictions about which content you’ll find most relevant and valuable. For example, our systems predict how likely you are to comment on a post, how likely it is that your friends will comment on the post if you share it, or how likely the post is to spark a conversation or back and forth discussion. We also use surveys to ask people whether a post was "worth your time" and these surveys are used to make predictions about other content you’ll find worthwhile. We also make predictions about whether a piece of content is problematic and should receive reduced distribution. All of these predictions are combined in the next step to produce the final ordering. We’ve shared more information about the predictions used in ranking below.

Ranking posts by score

Next, the system calculates a “relevance score” for each post and puts the posts in order based on this score. Generally, posts the system predicts will provide more value for you are shown higher up in your Feed. The system also tries to ensure your Feed has a balanced mix of content types. That means, for example, you are less likely to see multiple posts from the same Groups or from the same Page in a row; rather, you’ll see a range of posts from different sources.

Once our ranking system has calculated the relevance scores, the second-to-last step we take is to intersperse recommended content: we add this to help you explore and discover more about your interests through other people who share them, regardless of whether you’re already connected. Finally, we also include ads in Feed. Once this process is complete, your personalized Feed is ready!

Take a look at the video below to see more about how our ranking process works.

Giving People More Control Over What They See in Feed

When you engage with content in Feed, you give us a combination of explicit signals (e.g., liking, commenting, or resharing content, etc.), and implicit signals (e.g., viewing posts) that help us predict what’s meaningful to you. And because we believe it’s important you have even more control over your Feed experience, we’ve built tools to help you further customize what you see. These controls include:

  • Feed Preferences: provides options to fine-tune the way content is ranked in your Feed, including the ability to prioritize posts from your Favorites; Snooze or Unfollow people, Pages, and Groups to stop seeing their posts; and Reconnect with anyone you may have unfollowed.

  • Show More and Show Less: lets you directly tell us what you want to see more or less of by selecting “Show more” or “Show less” on the posts you see. Selecting “Show more” will temporarily increase the ranking score for that post and posts like it, and selecting “Show less” will temporarily decrease its ranking score.

  • Reduce: allows users to adjust the degree to which we demote problematic or low quality content in their Feed. (Our Content Distribution Guidelines outline some of the most significant reasons why problematic or low quality content may receive reduced distribution in Feed.)

  • Feeds tab: allows you to see the newest posts first; content is sorted in reverse chronological order (alongside ads).

A Deeper Dive into Predicting What You Want to See

The following two components are most important to determining what connected content you see at the top of your Feed - signals that tell us more about what you want to see, and the personalized prediction models powered by these signals that then create your unique Feed.

Signals Used in Ranking Connected Content

We use thousands of different signals to make predictions about whether you’ll find something more or less valuable. The categories of signals listed below represent the vast majority of the signals currently used in Feed ranking for connected content to make these personalized predictions. By drilling down into each category, you can learn more about the types of data we leverage in our models and some examples of individual signals.

To note, we have included categories and examples of signals specifically used for identifying problematic content, which we demote (or show lower in Feed). The information we’re sharing here around this subset of signals is purposefully more limited to guard against bad actors abusing our systems. Additionally, this information is subject to change.

Data specific to you
Data about your basic account information

Basic account information

  • How long you've been using Facebook

  • Language you are using Facebook in

  • Location-related information such as IP address and other device signals if you allow us to receive it

Data about how you're accessing Facebook

The time, frequency and duration of your activities on Facebook

  • Time of day for the location where you are

  • Number of days you've been active on Facebook, during a certain period of time

Device you're using

  • The device and software you're using, and other device characterisitics for example, the type of device, details about its operating system, details about its hardware and software, battery level, signal strength

Data about your activity on Facebook

How you've shared different content

  • The number of different posts you've shared, for example videos, photos, Reels, etc.

How you've interacted with different content

  • Types of content you've interacted with and how you've interacted with them, for example the total number of times you've clicked on photo posts or how many times you've commented on video posts

How you've viewed different content

  • Types of content you view and how long you view them for, for example how long you spend looking at photos, how much time you spend reading comments or how much time you spend watching video

Data about your connections (friends, Pages, Groups, etc)

Data about the overall available posts from your friends, Pages and Groups

  • How many new posts are available for you to see and the different types of posts that are available

  • How many of the posts from your connections have new comments

Data about Friends

  • How many friends you have

  • How often you interact with content from each friend

Data about Pages you Follow

  • Total number of Pages you follow

  • Number of Pages you've visited, during a certain period of time

Data about the Groups you are a member of

  • The number of Groups you've joined, during a certain period of time

Data about your basic account information
Data about how you're accessing Facebook
Data about your activity on Facebook
Data about your connections (friends, Pages, Groups, etc)

Basic account information

  • How long you've been using Facebook

  • Language you are using Facebook in

  • Location-related information such as IP address and other device signals if you allow us to receive it

The time, frequency and duration of your activities on Facebook

  • Time of day for the location where you are

  • Number of days you've been active on Facebook, during a certain period of time

Device you're using

  • The device and software you're using, and other device characterisitics for example, the type of device, details about its operating system, details about its hardware and software, battery level, signal strength

How you've shared different content

  • The number of different posts you've shared, for example videos, photos, Reels, etc.

How you've interacted with different content

  • Types of content you've interacted with and how you've interacted with them, for example the total number of times you've clicked on photo posts or how many times you've commented on video posts

How you've viewed different content

  • Types of content you view and how long you view them for, for example how long you spend looking at photos, how much time you spend reading comments or how much time you spend watching video

Data about the overall available posts from your friends, Pages and Groups

  • How many new posts are available for you to see and the different types of posts that are available

  • How many of the posts from your connections have new comments

Data about Friends

  • How many friends you have

  • How often you interact with content from each friend

Data about Pages you Follow

  • Total number of Pages you follow

  • Number of Pages you've visited, during a certain period of time

Data about the Groups you are a member of

  • The number of Groups you've joined, during a certain period of time

Data specific to the Post being ranked
Data about the settings or attributes of the Post

The privacy setting or visibility of a post

  • Whether the post is public or visible to only friends, custom, or only me

What type of post this is and what media it contains

  • Whether the post contains photos, video, Live video, a link, etc.

  • Is the post a live video or was the video previously live

Data about the post content

  • If the post contains a URL, the ratio of the popularity of a domain on FB vs the internet as a whole

  • The percentage of identical content between this post and other posts

  • Whether the post likely contains nudity, graphic violence, or any other likely Community Standards violation

  • If the content in the post has been rated false or partially false by one of our independent fact-checking partners

Data about the media, like photo or video, contained in the post

  • Width and height in pixels of the photo being shared

  • What visuals are contained in the photo

  • If the post contains video that loops and/or is a still image

Topic(s) the post is about

Data about the actor that created the Post

Attributes of the actor

  • Is the actor posting to a profile, Page or to a Group

  • If a Group post, the number of members in that Group

  • Number of confirmed Community Standards violations the account has accrued

Actions this actor has taken on Facebook

  • How many posts (videos, links, photos, etc) the actor has shared, during a certain period of time

How other users have viewed this actor or content from this actor before

  • How many times this Page has been viewed by users, during a certain period of time

  • Total number of people following the Page, during a certain period of time

How other users have interacted with content from this actor before

  • Total number of times content from this Page has been shared by users

  • Total comments, likes, shares on the Page

Data about the actor who shared the Post (when different from actor who created the Post)

Attributes of the actor

  • Is the actor posting to a profile, Page or to a Group

  • If a Group, the number of members in that Group

  • Number of confirmed Community Standards violations an account has accrued

Actions this actor has taken on Facebook

  • How many posts (videos, links, photos, etc) the actor has shared, during a certain period of time

How other users have viewed this actor or content from this actor before

  • How many times this Page has been viewed by users, during a certain period of time

  • Total number of people following the Page, during a certain period of time

How other users have interacted with content from this actor before

  • Total number of times content from this Page has been shared by users

  • Total comments on content from this Page by users

Data about how other users have interacted with this Post

Views on this post from all users

  • Total or average amount of time people have spent viewing this post

  • Total or average amount of time people have viewed a video in the post

Interactions on this post from all users

  • Total number of likes on the post

  • Total number of comments on the post

  • Ratio of clicks, likes, and comments on a post out of all times it's been viewed

Data about the settings or attributes of the Post
Data about the actor that created the Post
Data about the actor who shared the Post (when different from actor who created the Post)
Data about how other users have interacted with this Post

The privacy setting or visibility of a post

  • Whether the post is public or visible to only friends, custom, or only me

What type of post this is and what media it contains

  • Whether the post contains photos, video, Live video, a link, etc.

  • Is the post a live video or was the video previously live

Data about the post content

  • If the post contains a URL, the ratio of the popularity of a domain on FB vs the internet as a whole

  • The percentage of identical content between this post and other posts

  • Whether the post likely contains nudity, graphic violence, or any other likely Community Standards violation

  • If the content in the post has been rated false or partially false by one of our independent fact-checking partners

Data about the media, like photo or video, contained in the post

  • Width and height in pixels of the photo being shared

  • What visuals are contained in the photo

  • If the post contains video that loops and/or is a still image

Topic(s) the post is about

Attributes of the actor

  • Is the actor posting to a profile, Page or to a Group

  • If a Group post, the number of members in that Group

  • Number of confirmed Community Standards violations the account has accrued

Actions this actor has taken on Facebook

  • How many posts (videos, links, photos, etc) the actor has shared, during a certain period of time

How other users have viewed this actor or content from this actor before

  • How many times this Page has been viewed by users, during a certain period of time

  • Total number of people following the Page, during a certain period of time

How other users have interacted with content from this actor before

  • Total number of times content from this Page has been shared by users

  • Total comments, likes, shares on the Page

Attributes of the actor

  • Is the actor posting to a profile, Page or to a Group

  • If a Group, the number of members in that Group

  • Number of confirmed Community Standards violations an account has accrued

Actions this actor has taken on Facebook

  • How many posts (videos, links, photos, etc) the actor has shared, during a certain period of time

How other users have viewed this actor or content from this actor before

  • How many times this Page has been viewed by users, during a certain period of time

  • Total number of people following the Page, during a certain period of time

How other users have interacted with content from this actor before

  • Total number of times content from this Page has been shared by users

  • Total comments on content from this Page by users

Views on this post from all users

  • Total or average amount of time people have spent viewing this post

  • Total or average amount of time people have viewed a video in the post

Interactions on this post from all users

  • Total number of likes on the post

  • Total number of comments on the post

  • Ratio of clicks, likes, and comments on a post out of all times it's been viewed

Data specific to you and the Post being ranked
Data about how you have interacted with the Post

Your interactions with this post

  • If you've liked this post

Data about how you have interacted with Posts similar to the one being ranked

How you've viewed similar content

  • Your total views on other content of the same type

How you've interacted with similar content

  • Your total shares of other content of the same type

Data about you and the actor who created the Post

Your relationship to the actor who created the post

  • Whether you are the admin of the Page or your friend is an admin of the Page that shared the post

How you've viewed content from this actor before

  • Your total views of other posts from this Group

How you've interacted with content from this actor before

  • Your total shares of other content from this Page

Data about you and the actor who shared the Post (when different from actor who created the Post)

Your relationship to the actor who shared the post

  • Whether you are the admin of the Page or your friend is an admin of the Page that shared the post

How you've viewed content from this actor before

  • Your total views of other posts from this Group

How you've interacted with content from this actor before

  • Your total shares of other content from this Page

Data about how you have interacted with the Post
Data about how you have interacted with Posts similar to the one being ranked
Data about you and the actor who created the Post
Data about you and the actor who shared the Post (when different from actor who created the Post)

Your interactions with this post

  • If you've liked this post

How you've viewed similar content

  • Your total views on other content of the same type

How you've interacted with similar content

  • Your total shares of other content of the same type

Your relationship to the actor who created the post

  • Whether you are the admin of the Page or your friend is an admin of the Page that shared the post

How you've viewed content from this actor before

  • Your total views of other posts from this Group

How you've interacted with content from this actor before

  • Your total shares of other content from this Page

Your relationship to the actor who shared the post

  • Whether you are the admin of the Page or your friend is an admin of the Page that shared the post

How you've viewed content from this actor before

  • Your total views of other posts from this Group

How you've interacted with content from this actor before

  • Your total shares of other content from this Page

Prediction Models Used in Ranking Connected Content

The Feed ranking system has over 100 different prediction models. Generally, these prediction models fall into four categories:

Each prediction is a potential indicator of how valuable a person might find certain content. For example, sharing a post with other people can be an indication that you found that post to be valuable, so predicting whether you’ll share a post is a good signal of value for us to use to show certain posts higher in Feed versus other posts. As you might imagine, no one prediction is a perfect gauge of whether a post is valuable to you, which is why we use multiple prediction models in combination with the overall goal of making the Facebook app valuable for people in the long-term, not just in the specific moment when they're seeing this content.

Below you’ll see more details on the different prediction models we currently use most frequently in Feed ranking. In certain situations where the individual models are very similar in what they predict, we’ve combined multiple models into one description. For example, we have multiple models about predicting where you might click on a post - whether you’ll click on the post, on a photo in the post, into the comments of the post, etc. In this list, we’ve referred to all of these models as “How likely you are to click on some part of the post.”

The list below is grouped by how well these models are able to determine whether or not a post will be valuable; the models in the top group tend to be used most frequently in determining the ordering of the content in your Feed versus those in the following two groups. However, it’s the combination of all these models together that’s most important.

Because Feed ranking is personalized, the relative impact of each prediction model on Feed will vary depending on the person and the content, since everyone has different preferences about what they like and how they want to interact with content. For example, predictions about how long you might spend watching a video may be a stronger indicator of value for a video post than whether or not you will click on the video, while the opposite may be true for a post containing a link to an article. Another example is that for some people, “liking” a post is a strong indicator that they found that post valuable, whereas for others (such as people who don’t use the “Like” button), spending time reading the post may be a more useful prediction.

We continuously strive to improve our ranking systems to better deliver the most valuable experience to our users so the information outlined here might change over time. In addition, in an effort to deliver a more personalized experience for people, we continue to test and refine our approach to how political content is surfaced on Facebook.

Types of Prediction Models

Within each group, models are listed alphabetically and are not in a ranked order. Some of these predictions are only used if the post is relevant to the model, for example predictions about posts from Groups would only apply if the post being ranked is a Group post.

Used Most Frequently
  • How likely you are to be interested in content from your friends

  • How likely you are to be interested in the Group that shared the post or content from the Group, as measured by engagement with the Group or its content

  • How likely you are to be interested in the Page that shared the post or content from the Page, as measured by engagement with the Page or its content

  • How likely you are to click on some part of the post

  • How likely you are to interact with a post in some way by liking, reacting or commenting on it

  • How likely you are to meaningfully interact with the post, through some combination of commenting/liking/reacting/sharing to messenger/resharing and/or spending time viewing it

  • How likely you are to share the post

  • How likely you are to spend time viewing comments on the post

  • How likely you are to spend time viewing the post or content in the post (as opposed to just scrolling past it)

  • How likely you are to visit a Page after seeing a post from that Page

  • How likely you are to want to see more or see less content from the person or Page who shared the post

  • How likely you are to watch a video contained in the post and the predicted amount of time you'll spend watching it

  • How likely your interactions on the post from a Page will encourage the admin of that Page to share more content in the future that is valuable to you

  • The predicted amount of additional comments or replies a post will get if you comment or share the post

Used Occasionally
  • How likely you are to comment on the post

  • How likely you are to find a news article informative, if the post contains a link to a news articles

  • How likely you are to find the post worth your time

  • How likely you are to follow a Page after seeing a Page post re-shared by a friend

  • How likely you are to hide, snooze or unsubscribe on the post (used to reduce the distribution of a post)

  • How likely you are to like the post

  • How likely you are to react to the post (love, care, haha, wow, sad, angry); predictions about an angry reaction are used to reduce the distribution of a post

  • How likely you are to RSVP to an event, if the post contains a Facebook event

  • How likely you are to scroll through a post with a product listing from a Buy/Sell Group

  • How likely you are to send a message on a product sale post, where data privacy laws permit

  • How likely you are to spend time viewing the Event in a post, if you click on it

  • How likely your interactions with the Group post will encourage other members to share additional content to the Group or interact with other content from the Group in the future

  • The predicted number of additional times the post will be shared if the post will get if you share the post

  • The predicted number of likes the post will get if you share the post

Used Less Frequently
  • How likely you are to join a community chat from a Group post

  • How likely you are to like the post and to spend time viewing the post from a Page that you have favorited

  • How likely you are to make a donation on a post with a fundraiser

  • How likely you are to report the post (used to reduce the distribution of a post)

  • How likely you are to send the post in a message (where data privacy laws permit)

  • How likely you are to take an action to support a Creator, e.g. sending a Star

  • How likely you are to watch a video in full screen viewer, if the post contains a video

  • The predicted amount of time you might spend in the web browser, if you click through on a url in the post