Interaction history

Instagram uses interaction history to help order Feed posts: its explainer includes how many posts someone has liked, while Mosseri’s May 2023 explanation placed past interaction patterns first among the most important information used.78 Instagram also considers the viewer’s history with a particular author, such as whether they comment on each other’s posts.17 These records of past behavior inform predictions about the next action, including whether someone will spend a few seconds on a post, comment, like, share or tap the profile photo.1 A post appears higher when Instagram considers an action more likely and gives that action more weight, although Mosseri stresses that these are educated guesses that can be wrong.18

Instagram also uses interaction history for suggested posts in Feed, the Reels tab and Explore.3 Instagram’s Help Center says these suggestions draw on accounts someone follows, posts they have liked, saved, reposted or commented on, and their history of connecting with the account or similar accounts.3 Its recommendations guidance illustrates the connection with interests: interacting with restaurants and bookstores may lead to recommendations about food, recipes, books or reading.9 Search uses followed accounts, viewed posts and previous account interactions, usually placing accounts and hashtags someone follows or visits higher than those they do not.2

Instagram’s Your Activity feature provides a way to inspect interaction history: Mosseri explains that people can open their profile, tap the top-right menu and choose Your Activity to find Likes, comments and other past interactions.10 For previously watched reels, Mosseri gives the path Settings > Your activity > Watch history.11 To inspect accounts they have interacted with least over the previous ninety days, people can open their profile, tap Following and choose the least-interacted-with option, which Mosseri distinguishes from a mass-unfollow tool.6 For an explanation rather than a list of past actions, Meta’s Newsroom describes “Why Am I Seeing This?” as showing activity and other information behind predictions, and announced its expansion to Instagram’s Reels tab and Explore in June 2023.12

Meta patents describe possible ways to record and summarize interaction history, rather than establishing how Instagram stores the activity people can inspect.1314 A Meta patent about ordering and filtering comments describes recording newsfeed actions, such as likes and posts, in logs associated with individual users, either separately or in a shared log that identifies each user.14 A Meta patent about page recommendations describes turning each visited page into a numerical description, combining those descriptions into a summary of the visited pages, and using that summary to recommend other pages.13

Meta research examines how recommendation systems can learn from richer interaction histories while limiting storage and computing costs.1516 Meta’s SlimPer research describes repeatedly refining a compact assessment of how relevant a potential item is to someone while checking their history and other information; the paper reports deployment on Instagram Feed and Reels, with improved engagement and lower computing and memory costs.15 Meta’s Request-Only Optimization research describes training on groups of items shown in response to one user request, reducing repeated storage and computation so systems can learn from more data and use more complex models.16 For the separate task of recommending ads across Instagram and Facebook, Meta’s engineering team explains that activity histories vary greatly in length and that treating every history as though it were the longest could waste up to half the computing work.17