Two-sided value-based music artist recommendation in streaming music services

Abstract

Most work on music recommendations has focused on the consumer side not the provider side. We develop a two-sided value-based approach to music artist recommendation for a streaming music scenario. It combines the value yielded for the music industry and consumers in an integrated model. For the industry, the approach aims to increase the conversion rate of potential listeners to adopters, which produces new revenue. For consumers, it aims to improve their utility related to recommendations they receive. We use one year of listening records for 15,000+ Last.fm users to train and test the proposed recommendation model on 143 artists. Compared to collaborative filtering, the results show some improvement in recommendation performance by considering both sides' value in conjunction with other factors, including time, location, external information and listening behavior.