The Closed Loop Between Opinion Formation and Personalized Recommendations

In online platforms, recommender systems are responsible for directing users to relevant content. order enhance the users’ engagement, adapt their output reactions of users, who are, in turn, affected by recommended this article, we study a tractable analytical model user that interacts with an news aggregator, purpose making explicit feedback loop between evolution user’s opinion and personalized recommendation More specifically, assume is endowed scalar about certain issue receives it from aggregator: her influenced all received pieces news, which characterized binary position on at hand. The confirmation bias, is, preference confirms current opinion. aggregator recommends items goal maximizing number clicks (as measure engagement): fulfill its goal, has compromise exploring preferences exploiting what learned so far. After defining suitable metrics effectiveness (such as click-through rate) impact opinion, perform both extensive numerical simulations mathematical analysis model. We find recommendations markedly affect opinions favor emergence more extreme ones: intensity these effects inherently related recommender. also show tuning amount randomness algorithm, one can seek balance system opinions.