Accounting for Personalization in Personalization Algorithms: YouTube’s Treatment of Conspiracy Content

This article investigates under which video watch conditions YouTube’s recommender system tends to develop a preference for conspiracy-classified videos. Whereas existing research on so-called filter bubbles and rabbit holes rely non-personalized recommendations standard patterns, this study puts personalization diversified user strategies at the center of its design. 20 authenticated bots have been instructed YouTube content based four distinct strategies. In baseline strategy, watched non-conspiracy videos only. Treatment involved watching content, selected either non-personalized, partly-personalized, or fully-personalized input. Bots total 15 videos, after each their top homepage were collected classified as conspiracy-related not. allowed us measure impact strategy proportion recommended step. The same experiment has reverted, exposing treatment groups only, assess persistence pattern. Our results show that users primed with tend quickly receive much larger recommendations. Inverting pattern proves significantly more difficult than generating it. There are also indications relying personalized input might produce stronger effects. contributes evidence argument recommendation is prone strong, potentially pernicious patterns. Moreover, it replicable methodology stage in algorithms.