Facebook Reveals How Its AI Decides what to Recommend on Instagram

In a technical article published earlier today, Facebook has reveled some of the Machine Learning (AKA ‘AI’) technology that decides what photos show up on Instagram’s “Explore” tab—the main way IG users discover new accounts and content.

Getting your photography on the Explore tab can have a huge impact on your follower count, but there’s no department head at Facebook that you can schmooze to get your work included. As with so many things in the world of social media, your exposure (or lack thereof) is dependent on an algorithm.

But while most of most of these algorithms are shrouded in mystery, Facebook decided to share some details about how its Explore algorithm works.

The challenge, according to the article, is primarily one of scale.

“Over half of the Instagram community visits Instagram Explore every month to discover new photos, videos, and Stories relevant to their interests,” reads the article. And so Facebook’s engineers came up with “an AI system based on a highly efficient 3-part ranking funnel that extracts 65 billion features and makes 90 million model predictions every second.”

The algorithm’s job is to figure out what you’re interested in, and show you more content like it. The whole process takes place in a couple of discreet (if highly oversimplified…) steps: Candidate Generation and Ranking.

First, The algorithm identifies your interests by tracking what accounts you interact with and how—liking, saving, commenting. This data is used to create a list of “candidate” accounts that you might also enjoy.

Here’s just one example of how an account might make it into your “candidates” list:

Next, Instagram samples 500 of those candidates and runs them through a 3-stage ranking process that that identifies the “highest-quality” candidates. Step one takes the number from 500 to 150, step two takes it from 150 to 50, and step three identifies the 25 accounts you’ll see in the first page of the Explore grid.

The “how” behind this is where things get technical fast. Basically, the AI system predicts whether you’ll interact positively (a Like or Save) or negatively (See Fewer Posts Like This) with a particular piece of content you see in Explore, giving each predicted action a different “weight” and adding in a penalty for things like repeated content so you don’t see multiple photos from the same candidate account.

In this way, the Explore AI should winnow that list of 500 “candidates” down to the 25 accounts that you’re most likely to interact with positively.

In the end, the article doesn’t reveal too much actionable info that isn’t common sense, but it’s worth noting a few key takeaways for photographers who want to use Instagram to grow their fan- or client-base:

  1. Recommendations are made at the account level. In other words: Instagram isn’t finding photos you’ll probably like, it’s finding entire accounts you’ll probably like.
    Takeaway: If your feed is all over the place, you’ll never make it to Explore.
  2. There is a penalty to keep IG from showing multiple photos from the same account. That penalty increases the more of your photos a person sees.
    Takeaway: Only post your best work, because once you’ve made it into someone’s Explore, you may not make it in there again.
  3. Like it or not, it’s all about the likes. What gets recommended is what is most likely to receive a “positive” interaction.
    Takeaway: Aim for content that is like-worthy or save-worthy.

There’s a whole other conversation to be had about whether or not Facebook’s engineers are actively keeping people in their comfort zone by defining “interesting and relevant content” as “whatever you’re most likely to like and save.” But that’s a topic for another day.

To read all about the Machine Learning tech behind Instagram’s Explore tab for yourself, head over to Facebook’s Artificial Intelligence blog here.

(via DPReview)

Discussion