How Recommendation Algorithms Influence What We Watch Online

Abstract

Online video platforms provide access to an extremely large amount of digital content, making it difficult for individual users to manually search and evaluate all available videos. Recommendation algorithms address this problem by predicting which content a user may find relevant, interesting, or engaging and by ranking that content in a personalized feed. This paper examines how recommendation algorithms influence what people watch online, with particular attention to personalization, viewing history, user interactions, content characteristics, popularity, ranking, and feedback mechanisms. The study uses a narrative literature review based on academic research and selected first-party platform documentation. Existing research shows that large-scale recommendation systems commonly use multiple stages to reduce a very large content catalogue to a smaller set of potentially relevant items. User activities such as watching, skipping, liking, sharing, following, and searching can provide signals that influence future recommendations. These mechanisms can improve content discovery and reduce the effort required to find relevant material, but they may also reinforce existing interests, increase the visibility of popular content, and affect the diversity of content encountered by users. Evidence concerning filter bubbles, polarization, and algorithmic amplification is more complex than the assumption that recommendation systems completely determine user behaviour. Therefore, recommendation systems should be understood as part of a continuous interaction between users and algorithms rather than as systems that completely control viewing decisions. Future research should evaluate recommendation systems using multiple dimensions, including engagement, satisfaction, diversity, transparency, user autonomy, and long-term effects.