Critical inquiry into recommender systems for music.

Recommender systems are ubiquitous in today’s connected world. They influence what news we read, what music we listen to, or what results we receive when searching something online based on our previous choices and queries. Yet not many people know about recommender systems and how they influence our media consumption.

The FAV-5 research project aims to make recommender systems and their algorithms tangible. However, the focus of the project is not to improve these algorithms but to make us think and reflect on their impact in our daily lives. Specifically, Rachel’s research asks: what do we actually think of the fact that these recommender systems influence what we see, hear and read?

The FAV-5 is a music system made out of 5 radios and a web application, shared between 5 people. Using the web application, each user creates a playlist consisting of several songs from which the radios pick their songs to play. The radios themselves are simple devices with a single knob which can be used to turn the radio ON or OFF and set the volume. Turned ON, the radios loop a single song from a user’s playlist for the whole day, alternating between the playlists throughout the week. This way, each person is part of the recommender system, having an influence on what the other radio owners can listen to. After a weeks time, each user can re-create their playlist, reacting to the music of the previous week and influencing the other radio owners.

FAV-5 is a design research project that aims to make recommender systems and their algorithms tangible.

My work in this project was to provide technical support with the setup of the music system. Here’s how the setup worked: The web-application to create user playlists is a simple HTML input form styled with selectize.css. The individual playlists are then stored in form of a text file. Each day, a PHP script is executed on the server as a cron job and the script selects a playlist and a random song from the playlist. This selected song is then copied to a specific folder on the server under the name currentSong.mp3.

The radios, on the other hand, are made out of a RaspberriPi 3, a loudspeaker and a potentiometer with an ON/OFF switch. The radios are constantly connected to the website with the currentSong.mp3 and turned ON, play back the file. The communication between the website and the potentiometer on the RaspberryPi is established through WebIOPi, an IOT toolkit for the RaspberryPi.


Rachel van Berlo




  • Research