Delivering artist messages to listeners based on predicted responses

Abstract

A digital content provider that provides content channels to users via a computer network selects an artist message. The content provider retrieves a predictive model trained to calculate an artist affinity score indicating a likelihood that a given user will provide positive feedback to a given artist message. The predictive model is trained, using machine learning, based on feedback from users for previous plays of artist messages. The content provider retrieves artist features describing the artist who recorded the artist message, and user features describing a candidate user. The content provider calculates an artist affinity score for the candidate user and selected artist message using the predictive model. The content provider determines to play a content item by the artist to the candidate user, determines to play the selected artist message associated with the content item based on the artist affinity score, and transmits the artist message and content item.