Wanna know how much Instagram selects recommended posts to highlight in user home feeds?
A new article on Social Media Today reports that if you're looking to get a better handle on how Instagram's feed algorithms work, you can optimise your content approach accordingly.
Instagram has just published a new overview of how it ranks content for its 'Suggested Posts', or the posts that you see in your Home feed from accounts that you don't follow in the app. This element became a key focus recently, after Instagram began inserting a lot more AI-based content recommendations into user feeds, which prompted widespread user backlash, and has since seen IG scale it back, as it works to refine its algorithms. But even with that shift, Instagram does see AI recommendations as a key element of its future, and in maximizing user engagement.
So how does Instagram select which additional content to show you in your Home feed?
First off, Instagram's engineering team outlines the focus of its recommendation system, and underlines the key aims of its approach:
That then sees Instagram's post recommendations fall into two categories - 'Connected' and 'Unconnected', with the latter being the posts that Instagram's systems find and highlight, based on your interests.
The aim is to build more directly on your stated interests, as opposed to simply adding in the latest trending content:
"In order to ensure that our recommendations feel similar to posts in Home Feed we prioritize accounts that are similar to accounts a user encounters in Home. In the candidate selection step while training and evaluating our ranking models we ensure that the overall distribution is not skewed away from Home-based sources. We follow the same freshness and time sensitivity heuristics as Home Feed to ensure that suggested posts provide a similar kind of fresh feeling as the rest of Home Feed. We also ensure that the mixture of media types (like photos/videos/albums etc.) are relatively similar in Home and suggested posts.
For users whose immediate engagement graph is relatively sparse, we generate candidates for them by evaluating their one-hop and two-hop connections. Example: If user A hasn't liked a lot of other accounts, we can probably evaluate the accounts followed by the accounts A has liked and consider using them as seeds. A â†' Account Liked by A â†' Accounts followed by the accounts A likes (Seed Accounts)."
Instagram has just published a new overview of how it ranks content for its 'Suggested Posts', or the posts that you see in your Home feed from accounts that you don't follow in the app. This element became a key focus recently, after Instagram began inserting a lot more AI-based content recommendations into user feeds, which prompted widespread user backlash, and has since seen IG scale it back, as it works to refine its algorithms. But even with that shift, Instagram does see AI recommendations as a key element of its future, and in maximizing user engagement.
So how does Instagram select which additional content to show you in your Home feed?
First off, Instagram's engineering team outlines the focus of its recommendation system, and underlines the key aims of its approach:
- Users spend a lot of time crafting the perfect home feed for themselves. How can we do some of that work for them and make it feel like they crafted these recommendations themselves
- Anecdotally speaking, users who stay engaged keep finding newer sources of interests to follow. Can we help in this act of progressive personalization a bit?
That then sees Instagram's post recommendations fall into two categories - 'Connected' and 'Unconnected', with the latter being the posts that Instagram's systems find and highlight, based on your interests.
The aim is to build more directly on your stated interests, as opposed to simply adding in the latest trending content:
"In order to ensure that our recommendations feel similar to posts in Home Feed we prioritize accounts that are similar to accounts a user encounters in Home. In the candidate selection step while training and evaluating our ranking models we ensure that the overall distribution is not skewed away from Home-based sources. We follow the same freshness and time sensitivity heuristics as Home Feed to ensure that suggested posts provide a similar kind of fresh feeling as the rest of Home Feed. We also ensure that the mixture of media types (like photos/videos/albums etc.) are relatively similar in Home and suggested posts.
For users whose immediate engagement graph is relatively sparse, we generate candidates for them by evaluating their one-hop and two-hop connections. Example: If user A hasn't liked a lot of other accounts, we can probably evaluate the accounts followed by the accounts A has liked and consider using them as seeds. A â†' Account Liked by A â†' Accounts followed by the accounts A likes (Seed Accounts)."
-
Wealthkey -
[ 1 ] Thanks - Reply
{{ DiscussionBoard.errors[11726679].message }} -
-
petervirdee -
Thanks - Reply
{{ DiscussionBoard.errors[11726958].message }} -