Music Case Study

Spotify is a Swedish audio streaming service provider that users access through a free or paid subscription. Subscribers gain access to songs, podcasts and a panoply of playlists organized by genre, mood, activity and more. As of December 2021, Spotify had 180 million premium subscribers worldwide (Götting, 2022). One reason why so many enjoy Spotify’s service is the recommender system, which is known for creating personal playlists so users can discover new songs and artists.

As Hodgson (2021) explains, Spotify’s recommender system is made up of a few machine learning techniques such as collaborative filtering, natural language processing, and audio analysis models (also known as music information retrieval). The collaborative filtering models are used to analyse how each user interacts with a software platform and they compare it to other users' behavior (Hodgson, 2021). The metrics used as input are the times a song is played, skipped, saved to a playlist, or shared with another user and more. This data is then compared to other users’ behavior to assess how each person responds to any given song or artist recommendation (Hodgson, 2021). The natural language processing models are used to analyze text found online on artist and song names that can then be evaluated against other words like positive and negative adjectives, as Hodgson describes it (2021). The model then looks for similarities and differences between artists, and the results are combined with the collaborative filtering models to determine an artist’s or song’s overall ranking. Lastly, the audio analysis models are used to analyze “digital audio waveforms in order to determine, inter alia, time signature, key signature, tempo and dynamics” (Hodgson, 2021, p. 6). This means that each song can be compared to see similarities and differences, and they can be organized into curated playlists for moods and activities.