Instagram revealed details of the recommended content section algorithm
The algorithm consists of three main blocks of ranking, which “extracts” 65 billion functions and makes 90 million models every second so that you and I can slip on the recommended dogs and cats.
Instagram defines profiles similar to each other by analyzing texts. The system does not analyze so many keywords, but their order in which words appear in the text and how much they are connected.
This is the first key and important knowledge that Instagram evaluates not every single post, but entire profiles among themselves.
The system builds recommendations based on the “initial profiles” (seed accounts) with which you interacted before. It analyzes what content you like and save and collects a cloud of similar profiles for you, from where it selects 500 posts. These posts are filtered out from spam, misinformation, and “content that MAY violate Instagram rules” (hi Shadow Ban) and ranks the remaining content based on the probability of interaction with the posts.
As a result, the algorithm selects 25 posts that you see immediately when you open the section with recommended content.
So simple. At the same time, in our opinion, this indirectly confirms the theory that Instagram does not determine the coverage of each individual post, but distributes the “basic” coverage to the profile for the posts that they will receive anyway. And, provided that the post works well (based on a bunch of factors), it expands the reach of subscribers as well and allows it to get viral.