If your music collection contains everything from Elvis to Elgar and Franz Ferdinand to Franz Liszt, you should not be surprised. According to two statistical physicists in Belgium, our musical tastes may be more diverse than we realise. Renaud Lambiotte and Marcel Ausloos of the University of Liège have analysed data from music-sharing websites and found that people like a wide variety of music that belong to a mixture of several pre-defined categories. The results could provide a new way of classifying musical genres and be useful for analyzing market trends (physics/0508233).
The study by Lambiotte and Ausloos was based on data from audioscrobbler.com — a website where users upload their favourite musical recordings to a personal “library” on a central server. The website is designed to encourage users to discover new music by proposing recordings in other people’s libraries that it thinks meet their taste. Using data from the site in January 2005, the two physicists were able to examine the music listened to by a total of 35916 people, who together owned 617 900 different “music groups” in their libraries. Each music group refers to a particular artist, the top five of which were Radiohead, Nirvana, ColdPlay, Metallica and the Beatles.
Lambiotte and Ausloos realised that the list of users and the music they listen to can be analyzed using methods from “complex-network theory”, in which the people and the music form two different types of “nodes” in a network. The physicists started with a “fully connected” network, in which any two people are connected if they share at least one song by the same artist. They then applied a “filter” to their network, which takes into account correlations between peoples’ entire music collections rather than just between the individual music groups they like.
As the value of the filter increased, the Belgian duo found that disconnected structures or “branches” formed along the original network structure, thus revealing collective trends and cliques in the form of a “map” of different musical genres (see figure). Moreover, the study revealed that each listener is characterized by a very diverse mixture of music groups, which Lambiotte and Ausloos call the “individual musical signature”. The fact that our musical tastes are becoming more varied and unconventional may be because people can easily download music from the Internet to form personalized databases of favourite recordings on iPods and other devices.
The physicists found listeners who liked such diverse groups as The Jon Spencer Blues Explosion, Galaxie 500, Prince and the Revolution, Uriah Heep and Laurent Garnier — an unexpected mixture of indie rock, funk, hard rock and dance. “These structures do not fit the neat usual genres defined by the music industry and represent the non-conventional taste of listeners,” explains Lambiotte. “Our method accounts for the fact that music perception is driven both by the people who make music and also the people who listen to it.”
The results could also provide a new way to redefine these genres based on the choices of the listeners themselves. Lambiotte and Ausloos also present a simple “agent-based voter growth model” that highlights how listeners form opinions about different types of music, which could help to explain how particular artists becomes trendy. “Our method allows us to quantify the music signatures of a large sample of individuals and visualize their collective behaviour, that is the emergence of sociological communities,” says Lambiotte.
