Designing Recommendation Systems on Content Platforms: Trading off Quality and Variety

Abstract

User-generated content platforms use advanced recommendation algorithms to deliver content from a vast number of creators to a wide audience with diverse preferences. Recommendation systems match their offers to users' tastes while taking into account content quality. The literature has focused on the impact of recommendation systems on demand-side factors, such as the quality and variety of users' content consumption. However, the literature has largely overlooked their influence on supply-side factors, including creators' entry and quality decisions, and how they interact with consumers' content diet. The literature has also neglected the “blockbuster effect”, whereby very few “trendy” content generates disproportionate consumer interest. We address the above omissions and explicitly model creators' entry. We reveal a tradeoff between the provision of content quality and variety on a platform. While increasing the recommended content's fit to users' preferences benefits consumers and the platform, pushing higher quality content may hurt consumer welfare by reducing the variety of content produced. Additionally, consumption variety is a U-shaped function of the emphasis the recommendation system puts on content quality. In the presence of trendy content, emphasizing content quality may reduce creators' participation and hinder the emergence of trendy content.

JEL Classification: D43, L82, M31, L13