You are here

The Roles of Music Recommendation Systems

Milwaukee.
Up next in this panel at AoIR 2009 is Simone Pereira de Sá, whose focus is on music recommendation systems; such systems are mediators or translators to which we delegate the task of recommendation. They promise something else for the different actors in the process: artists are presented to the right people, while listeners find new music they should enjoy, and this is further enhanced through social networking tools and tagging functionalities.

Labelling systems deal with the complex issue of music classifications, choices, and tastes, and this ties into the question of musical genres - so, how do recommendation systems work on this basis, and strain, support, or overcome the idea of musical generes? As Simon Frith has suggested, one of the greatest pleasures of entertainment culture is the discussion of different values and tastes; different opinions have different levels of credibility here. This is also connected to subcultural theory, of course, which ascribes certain subcultural capital to agents in contact with the media and refers to consuming certain exclusive information and the 'right' cultural products.

Classification or labelling systems are far from neutral, then - musical genres, for example, go beyond technical or business attributes but perform an important mediator role for different groups engaged in music; these systems are used to create borders between genres and define subcultural communities. Additionally, music has a ritual role in society, of course.

So how do last.fm tagging and recommendation systems fit into this context? There are a number of different user experience levels to consider here: using the last.fm player, which allows favouriting, banning, skipping, and tagging songs; and music scrobbling, which reports one's listening activity back to the system and extracts new recommendations from these data. Both are forms of collaborative filtering, where data from the user enables the system to make better recommendations.

Further, there are the social resources on the last.fm platform, which include user profiles, tagging, friends lists, neighbouring users, internal mail, etc. But through such systems, the network becomes more complex and the feedback more intense; on the one hand, the system uses the information already collected, while these functions also reintroduce the symbolic dispute that is central to music subcultures.

Tagging could make this system highly chaotic and random, in principle - but research shows that tagging follows relatively fixed conventions which use tags that may be little known outside of a subcultural community but are clearly established within it. Indeed, there is a certain direct or mediated peer pressure to listen to the 'right' music through such systems, resulting in a substantial level of conformity in classifications between expert and amateur classifications - especially in relatively narrowly defined genres, less so in generic genres like 'pop' and 'rock'.

So, musical genre remains an important debate marker and a way of socialising and distributing social capital. However, it is extended and complexified by non-genre-related tags like 'sad songs', songs for studying', etc. And while platforms like last.fm are based on collaborative filtering, engaging the user in their social networks are crucial to the recreation of taste and value discussions on the site. Recommendation systems thus become another class of agents engaged in the field of musical taste: non-human, but intelligent.

Use of these systems is a delegation activity, where the system asks its users to participate in classifications and recreate taste communities - and these systems constitute a highly sophisticated network full of new mediators which enable their users to navigate the musical universe.

Technorati : , , , , ,
Del.icio.us : , , , , ,