Proceedings of the Ninth International AAAI Conference on Web and Social Media Understanding Musical Diversity via Online Social Media Minsu Park∗1, Ingmar Weber2, Mor Naaman1, Sarah Vieweg2 1Jacobs Institute, Cornell Tech 2Qatar Computing Research Institute (QCRI) fminsu, [email protected] fiweber, [email protected] Abstract matter of individual choice and expression; however, to a great degree, it is hypothesized and tested that the diversity Musicologists and sociologists have long been inter- of musical tastes can be explained by external factors. For ested in patterns of music consumption and their re- lation to socioeconomic status. In particular, the Om- example, previous research has identified a relationship be- nivore Thesis examines the relationship between these tween musical tastes and social factors, and produced the variables and the diversity of music a person consumes. cultural omnivore thesis. This thesis describes “a shift in Using data from social media users of Last.fm and Twit- the orientation of high-status individuals toward an inclu- ter, we design and evaluate a measure that reasonably sive range of musical preferences that traverses the tradi- captures diversity of musical tastes. We use that mea- tional boundaries between highbrow, middlebrow, and low- sure to explore associations between musical diversity brow genres (Peterson 1992; 1997; 2005).” However, sym- and variables that capture socioeconomic status, demo- bolic boundaries between musical genres have been erod- graphics, and personal traits such as openness and de- ing (Goldberg 2011) in recent years, which provides an op- gree of interest in music (into-ness). Our musical diver- portunity to rethink the high-to-lowbrow cultural categories sity measure can provide a useful means for studies of musical preferences and consumption. Also, our study in relation to musical diversity. This can lead to a better of the Omnivore Thesis provides insights that extend understanding of the impact of social conditioning on di- previous survey and interview-based studies. verse musical tastes, and by proxy, a better understanding of the connection between socioeconomic status, demograph- ics, and the diversity of musical preferences. Introduction To date, the social computing community has examined The cultural and social significance of music is universal; online listening activity as source of information and recom- music is found in every known human culture, and plays mendations for music (Bu et al. 2010; Zheleva et al. 2010; a role in rituals, wars, ceremonies, work, and everyday Farrahi et al. 2014; Turnbull et al. 2014). However, com- life (Wallin, Merker, and Brown 2001). Tia DeNora (De- putational tools and online outlets such as social media can Nora 2000) noted that “Music is not merely a meaningful make further contributions toward understanding human be- or communicative medium. It does much more than convey havior related to musical consumption and help to elabo- signification through non-verbal means. At the level of daily rate user-centric music retrieval systems by analyzing per- life, music has power. It is implicated in every dimension sonal characteristics. We focus on exploring a new means of social agency.” As social media become more ingrained of measuring the diversity of individual musical tastes by in our lives, it follows that connections between social me- using data collected from social media, and examine the re- dia use, and habits and norms regarding music consumption, lationship between musical diversity and various individual will occur. In this paper, we present an empirical analysis of factors including socioeconomic and demographic informa- social media data as they relate to and reveal details of users’ tion, as well as social and individual information that can be musical tastes. collected from social media. A person’s musical consumption can reveal a lot about Through a multi-platform analysis of a dataset of U.S. their personality, preferences, and sense of self. One can Last.fm1 users and their corresponding Twitter accounts, have limited tastes; they may listen to a single genre like pop we examine music consumption together with demograph- or rap, and not diverge into other genres. On the other hand, ics (e.g., age and gender) and other descriptive variables for another individual may be eclectic in their musical choices a community music fans who have an online presence. Using and have a playlist filled with jazz, hip-hop, indie rock, clas- 1 sical, and so forth. We often think of such differences as a Last.fm is a music recommendation service. The site builds a detailed profile of each user’s musical consumption by recording ∗The majority of this work was done while Minsu Park was a details of the tracks the user listens to, either from Internet radio research intern at Qatar Computing Research Institute. stations, or the user’s computer or many portable music devices. It Copyright c 2015, Association for the Advancement of Artificial also offers some social networking features such as recommending Intelligence (www.aaai.org). All rights reserved. and playing artists to Last.fm friends (Wikipedia 2015). 308 Twitter-derived information for these users, we inferred their aspects of western culture lead to a ‘highbrow’ status, co- socioeconomic information (e.g., income, education level, pious research has investigated the relationship between so- and area of their residence) as well as other social and per- cioeconomic position and musical tastes (Coulangeon and sonal variables (e.g., how diverse their friends and interests Lemel 2007). The majority of the current studies on the are, and how ‘open’ and ‘into music’ they are). We then de- omnivore thesis in relation to musical tastes, proposed by fined a measure for musical diversity by applying the no- Richard Peterson (Peterson 1992) show that people with a tion of shared understanding as socially perceived distances higher socioeconomic status have broader (omnivorous) mu- between genres. We suggest that designing a diversity mea- sical tastes than those with a lower socioeconomic status sure can provide a useful means for studies in recommen- who have limited (univorous) musical preferences in low- dation systems. Moving from designing a measure to analy- brow music. There are generally two definitions of omniv- sis of associations between diversity and individual factors, orousness, referred to as the volume and the compositional we suggest this type of analysis can provide meaningful in- definitions (Warde, Wright, and Gayo-Cal 2007). The first sights that are complementary to those provided by previ- refers to higher socioeconomic status people favoring more ous survey and interview-based studies regarding the musi- musical genres than those of lower socioeconomic status. cal omnivore thesis. Our main contributions therefore are as The second refers to the situation that people with higher so- follows: cioeconomic status tend to have more eclectic tastes across the spectrum of high-to-lowbrow music than people with • We propose and validate a novel diversity measure that lower socioeconomic status. borrows the concept of Rao-Stirling diversity for music More recently, however, Peterson (Peterson 2005) con- consumption. While recent studies (Hurley and Zhang ducted comparative research and noted that “despite the at- 2011; Farrahi et al. 2014) define diversity (as it relates tention paid to the concept by numerous scholars, the sub- to music consumption) as the total number of unique gen- types of omnivorousness suggested by them were diverse res associated with all artists listened to, we go into more and fall into no recurrent patterns due to changes in the detail, and define diversity as a multidimensional prop- socio-cultural world.” Indeed, though there is a little dis- erty that has three main attributes: variety (the number agreement that the contemporary era has witnessed shifts in of unique genres one listened to), balance (the listening the ways cultural preferences and practices are mapped onto frequency distribution across these genres), and disparity social locations, the extent to which this implies changes (the degree of distance between musical categories). in the functioning of cultural capital remains unclear (Rim- • We investigate the relation between musical diversity and mer 2012). In addition, Peterson (Peterson 2005) raised a various other variables including socioeconomic factors. question regarding the traditional measurement of omnivo- In particular, we find that followers of high-profile news rousness, and recent qualitative studies identified a number media are more likely to have diverse musical tastes. We of limitations in conventional survey-based studies (Warde, also consistently find a weak, but robust trend for peo- Wright, and Gayo-Cal 2007; Rimmer 2012): First, the sim- ple who are more ‘into’ music to have less diverse tastes. ple or compositional volume of genres preferred by an indi- Along with these findings, our results also show that de- vidual is insufficient to show the full picture of one’s form mographic factors such as age and gender are associated of engagement and social status since different conceptual with musical diversity rather than conventional socioeco- frameworks may provide different understandings. Second, nomic status such as income and education level. there is a tendency to discriminate genres within preferred genres (i.e., even though one answers ‘rock’ as a preferred We begin by reviewing the primary