Developers of recommender systems are often confronted with popularity bias. This is the tendency of models to select more popular content, even in situations where less popular content would be more appealing to the user. While this behavior is acceptable in a moment, it’s strategically important to support content from more niche authors. Doing so helps increase the platform’s content diversity and attract new audiences.
The Zen development team faced a question: How do we keep newcomers active on the platform? Well, you could subsidize them with impressions, but how do you know how many impressions will retain an author? In her report, Anastasiia will show what mechanism they have developed to rank small authors. She will share the technical details of the mechanism and the results of its operation in production. Spoiler alert! The platform enjoyed growth in the number of active authors.