Reactive user policies can beat social recommender opinion drift

A new paper on arXiv (2508.13473) by Atefeh Mollabagher and Parinaz Naghizadeh examines whether users can protect their opinions from the influence of social recommender systems. The authors model two content consumption strategies: a passive policy where the click probability on recommendations is fixed, and a reactive policy where the probability adaptively decreases following large changes in a user's opinion. They analytically derive the expected opinion and user utility under both policies, considering influence from both the social network and the recommender.

The key finding is that the adaptive policy can prevent opinion drifts induced by recommendations, and when a user values opinion preservation over pure engagement, the reactive policy yields higher expected utility than the fixed one. Numerical simulations validate the theoretical results. The study highlights how individual user-level strategies—not just platform design—can mitigate algorithmic bias, offering a game-theoretic lens for recommender system research and potential tools for users to maintain autonomy.