NEW YORK, Nov 27 — In a blog post published Monday, Facebook has revealed the process by which Instagram selects the content to suggest to users in the Explore tab.
Creating Instagram’s Explore tab was, according to Facebook, quite the technological feat. In short, the purpose of the page is to recommend to users “the most relevant content out of billions of options in real time” so that they can discover new content that piques their interest.
.@Instagram is solving unique #machinelearning challenges: Recommending the most relevant content out of billions of options in real time at scale. Today we’re sharing the first detailed overview of the elements that make Explore work: https://t.co/lAxG2X2Gmq pic.twitter.com/yNbpPMENY2
Advertisement— Facebook AI (@facebookai) November 25, 2019
The company explained in a blog post how the Explore tab gives each user a unique and personalised experience; tools to do this were developed based on three main needs:
- “The ability to conduct rapid experimentation at scale”
- A method to “obtain a stronger signal on the breadth of people’s interests
- A “way to ensure that [their] recommendations were both high quality and fresh”
The team then developed a series of algorithms and techniques that together address all these needs and assembled them in one place.
One of the main steps of finding relevant content for an individual is to tag “interest-focused accounts based on specific themes”; by doing this, Instagram is able to determine which accounts are topically similar to one another whether the focus of the account is “Arabic calligraphy”, “model trains”, or even simply “slime”. Based on these tags, the platform can predict which accounts out of millions an individual will likely interact with.
To ensure that content in the Explore tab is “high quality and fresh”, it passes through a ranking distillation to preselect posts that could potentially become one of the 25 items that make it to the top of a user’s Explore page.
Using and reusing these processes, ranking and reranking content, and combining these results with how people engage with such content, Instagram is able to filter out material that wouldn’t interest a user — or is spam or misinformation — and offer them only relevant publications with the top 25 being the most personalised.
With over 110 million active users using Instagram and each of those individuals being provided with a personalised Explore tab, the platform is generating billions of content suggestions on a regular basis within seconds of someone tapping on the page.
At the end of the day, this method of generating unique, tailored suggestions for users becomes more accurate the more an individual interacts with content. — AFP-Relaxnews
