Original title: On personal recommendations in social networks
Philipp Grünter, Karl Henrik Johansson, Angela Fontan · September 2026
Social media algorithms constantly recommend content to change how we think. However, existing math models do not show how people actually respond to these suggestions. To fix this, researchers created a new mathematical model based on confirmation bias, which is our habit of accepting ideas that match what we already believe. In this model, people react to content using their past beliefs and a personal sensitivity setting. The researchers then tested two ways to recommend content. When an algorithm sends the exact same message to everyone, it can create two opposite stable groups, causing unintended results. When an algorithm targets people individually, the researchers figured out how to pick optimal recommendations that still guide users reliably, even if their sensitivity is uncertain.
Imagine a room of people trying to decide which door to leave through. If someone announces the same vague direction to everyone, half might turn left and half right because each person interprets it differently based on where they are standing. Here, the broadcast recommendation is that vague direction, and each person's current position is their prior belief. The model shows that such uniform advice can permanently split the group.
Social networks constantly recommend content, shaping what we see and believe. This work shows that sending the same recommendation to everyone can inadvertently split the group into opposing camps—a form of algorithmic polarization. Understanding this helps designers choose between uniform and personalized strategies to avoid unintended consequences.
The abstract does not specify empirical validation; the model is theoretical and demonstrated only through numerical simulations. It assumes a specific Gaussian influence function and focuses on uncertainty in the sensitivity parameter alone. Real-world recommendation dynamics may involve additional complexities not captured.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
On personal recommendations in social networks — Philipp Grünter, Karl Henrik Johansson, Angela Fontan
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.