12 Jul Pinterest Updates Algorithm to Surface More Content Types via @MattGSouthern
Pinterest is introducing a new ranking model to its home feed in an effort to surface certain types of content more often.
Traditionally, Pinterest ranks content in the home feed using a click-through prediction model.
Pins that a user is most likely to click on, as determined by past activity, are prioritized in their home feed
While that model is effective at maximizing user engagement, it’s not the best model for surfacing a variety of content types.
For example, if a user never clicks on video content then they’ll never be shown pins with video in their home feed.
But that doesn’t necessarily mean they wouldn’t engage with video content if it were to be surfaced.
Pinterest found itself with a problem of wanting to boost more content types while still keeping content recommendations relevant.
To solve this problem, Pinterest is introducing a real-time ranking system for its home feed called “controllable distribution.”
Controllable Distribution
Pinterest describes controllable distribution as a “flexible real-time system.”
It’s not a complete algorithm overhaul. Rather, controllable distribution is only applied after the traditional home feed ranking algorithm.
Pinterest will still use its click-through prediction model to find relevant content. Then it will apply controllable distribution to diversify the types of content being displayed.
Controllable distribution makes it possible to specify a target for how many impressions a certain content type should receive.
For example, controllable distribution could be used to specify that 4% of users’ home feeds should contain video content.
This is done through a system that tracks what percentage of the feed was video in the past. Then, the system boosts or demotes content according to how close that percentage is to the specified target.
Pinterest says this can be accomplished while still respecting users’ content preferences.
What Does This Mean for Marketers?
As a real-time system, the controllable distribution model will be continuously adjusted.
On one hand, that means the home feed won’t get stale for users.
On the other hand, it’s not exactly possible to optimize for an algorithm that changes in realtime.
Perhaps the best piece of advice for Pinterest marketers to take away from this is to follow Pinterest’s lead.
Pinterest is diversifying the types of content in the home feed. If you want more opportunities to show up in peoples’ feeds then diversify the types of content you publish.
For example, if you only publish photos, then consider adding some videos or GIFs to the mix. Maybe some product pins if you’re an e-commerce retailer.
Pinterest’s target for displaying certain types content will be changing all the time.
Publishing a wide variety of content will help ensure you have the right type of content available at the time Pinterest wants to display it.
Additional Notes
Pinterest’s home feed ranking team used to do manually what controllable distribution is designed to do algorithmically.
Yes, Pinterest’s home feed ranking team actually used to step in and adjust how often certain types of content appeared in users’ home feed.
Yaron Greif of Pinterest’s home feed ranking team describes the old process as “painful for both practical and theoretical reasons.”
“In practice, these hand-tuned boosts quickly became unmanageable and interfered with each other. And worse, they often stop working over time — especially when ranking models are updated. We regularly had to delay very promising new ranking models because they broke business constraints.
In theory, controlling content on a per-request basis is undesirable because it prevents personalization. If we show each user the same number of video Pins we can’t show more videos to people who really like to watch videos or vice versa.”
Pinterest says it’s committed to investing in the post-ranking stage of surfacing content. So it’s possible we may see this model applied elsewhere on the platform in the future.
Source: Medium
Sorry, the comment form is closed at this time.