The Effects of Diversity in Algorithmic Recommendations on Digital Content Consumption: A Field Experiment
73 Pages Posted: 25 Feb 2023 Last revised: 13 Feb 2024
Date Written: February 11, 2024
Abstract
While social media platforms adopt personalized recommendations to stimulate user consumption and engagement, content diversification is commonly added to recommender systems, with the belief that it will encourage users to explore unfamiliar content on platforms. Yet, the consequences of this practice on user behavior and platform trade-offs remain unclear. This study addresses this gap through a partnership with a leading global music-streaming service, wherein the authors enhanced its algorithm to amplify content diversification. They then conducted a large-scale field experiment where users were randomly assigned to receive recommendations either from the platform’s current algorithm or the modified one. Surprisingly, contrary to industry assumptions, diversified recommendations did not necessarily increase users’ content consumption diversity. Overall consumption actually declined. However, focusing on active users who account for the majority of the platform’s content consumption, a 1% uptick in recommendation diversity resulted in a .55% rise in their consumption diversity without impacting overall consumption levels. The authors demonstrate that the mechanism underlying this uptick for active users is consistent with accurate predictions of user preferences. Based on the insights, the streaming service tailored its recommender system to focus on content diversification for its active users.
Keywords: Recommender Systems, Social Media Platforms, Recommendation Diversity, Field Experiment
Suggested Citation: Suggested Citation