|Instagram's headquarters in Menlo Park, California. / Photo by: Anthony Quintano via Flickr|
To clarify how its feed ranking algorithm works, Facebook subsidiary Instagram gathered a group of reporters at its San Francisco office to explore the internal workings of the said algorithm, according to Josh Constine, writing for TechCrunch.
The algorithm was rolled out by Instagram in July 2016 as the replacement of a reverse chronological feed algorithm that was causing users to lose 70 percent of their posts and 50 percent of their friends’ posts. The result was that its 800 million-plus users were seeing 90 percent of their friends’ posts and were spending more time on the app
The three main factors behind the machine learning-based algorithm are:
The algorithm predicts how users care about a particular post. Higher rankings are given to what matters to users, determined by their past behavior on similar content and the actual content of the post being analyzed by machine vision.
The algorithm gives priority to timely posts over weeks-old ones.
The algorithm estimates how close the user is to the person that shared the post. It gives a higher ranking to people that users have interacted with a lot in the past on Instagram. This includes commenting on their posts or being tagged together in photos.
Besides these three primary factors, the algorithm makes use of three minor factors that influence ranking.
How often users visit Instagram because the algorithm will try to show them the best posts since their last visit.
If users follow a lot of people, the algorithm will be choosing from a bigger list.
The amount of time that users spend on Instagram determines if they will be seeing the best posts during short sessions.
During the meeting with reporters, the Instagram staff belied rumors that it will be reverting to the old reverse chronological feed because it has no intention of adding unnecessary complexity to its feed. The Instagram team added that no posts are being hidden in the feed, and users will see everything posted if they keep on scrolling.
The feed algorithm is not biased towards either the video or photo format but users’ feeds are dependent on the kind of content that they have been seeing.