Musical similarity is typically equated with genre, whether reflected directly within various categorization systems, or indirectly within recommender systems. However, taking into account (1) the ways in which genres are constructed, (2) the common musical facets / traits (and listeners’ understandings of them) found in diverse genres, and (3) the emergence of musical “omnivores,” there exist other possibilities for defining musical similarity. The proposed research focus for my doctoral thesis will provide a foundation for understanding (1) how avid recreational music listeners perceive musical similarity and (2) the extent to which they agree with top-down genre-based models. The findings will have implications for considering how well categorization and recommender systems meet the needs of such users, and could eventually inform the development of new systems that account for broader notions of musical similarity.
1. “What Kind of Music Do You Like?”
Although music can be personally enriching and meaningful, it can also have significance to one’s sense of belonging to a broader social circle. Music can act as a signifier by which people can identify others who share common interests, whether prospective friends or (as this heartwarming story describes) intimate partners. It seems no wonder that a common suggested icebreaker for first dates is something related to musical taste. On a related note, Maslow mentioned that music and sex are two ways of achieving “peak experiences” (Gabrielsson, 2011)
Regardless of the context, responses to questions about musical taste will likely relate to specific musicians, whether performers or composers. Another possibility is to mention musical genres. Not much of a surprise, given that music is categorized that way in brick-and-mortar stores, libraries, and a variety of online systems (Cunningham et al., 2003). In library catalogs, subject headings for textual works relate to their “aboutness,” while music and other “non-text” materials have subject headings related to genre (McKnight, 2002). Although they offer a wider range of “bottom-up” possibilities for music categorization, the majority of tags have tended to relate to genre as well (Lamere, 2008).
Although they don’t operate on the basis of genre, recommender systems on Amazon.com, YouTube, and other such sites typically reflect such forms of categorization. This is because they track observable user behaviors by using collaborative filtering algorithms, with two of the most common being user-to-user and item-to-item. Systems employing user-to-user algorithms make recommendations on the basis of aggregated information about similar behavioral tendencies (e.g. purchases, views, ratings) among multiple users, which ideally predict and reflect tastes the users likely have in common. In contrast, item-to-item algorithms draw upon individual user tastes and interests, focusing on objects that are ideally similar to those previously purchased, viewed, rated, etc. (Bonhard & Sasse, 2006; Celma, 2010; Celma & Lamere, 2011; Ricci, 2011).
2. Genre Trouble
(With Apologies to Judith Butler)
Recommender systems can be helpful to people who want to explore genres unfamiliar to them. The sense of novelty can eventually wear down as one feels sufficiently well-versed in a particular genre, however, and might even wish to explore genres perceived as “very different” from what they already enjoy.
Genre itself is a problematic construct as well. For starters, musicians can take liberties with the rules, whether implicit or explicit, surrounding a pre-existing genre. With enough “transgressions,” musicians can end up creating music that’s sufficiently different from its predecessors to be considered part of a new genre. In this process, transitional works emerge, which can prompt debate as to whether they belong to an older or newer genre (Fabbri, 1982). It’s little wonder, then, that taxonomies can vary widely in (1) numbers of genres and (2) the granularity to which levels of subgenres are parsed out. To some degree, such taxonomies also reflect the vested interests of those in the music industry that use genre to identify, or even create, prospective markets for their wares (Aucouturier & Pachet, 2003; Holt, 2007; Negus, 1999). To further complicate things, lay listeners might disagree on the genre to which a work or musician belongs. The same is true for music experts, as exemplified by debate about whether bands like Led Zeppelin are “heavy metal” or “hard rock” (Cope, 2010).
Much of the debate about genre categorization tends to occur in discussion about various kinds of popular music. At least in the broadest colloquial sense, “classical music” seems imperturbable. Nonetheless, being music, it isn’t entirely immune from the confusion surrounding genre, either. And it’s certainly not immune from considering what one means by “musical similarity,” beyond genre-based categorization.
3. A Personal Interest
I developed a keen interest in classical music in high school, back in the late 1980s. So, if someone at that time were to ask me what I liked, that’s what I would’ve said. It’s what I listened to most often, and it’s what meant the most to me personally. Some complications emerge from that statement, however.
At a more obvious level, adolescents are “supposed” to like “popular” music… or at least certain kinds, depending on where one lives. Needless to say, especially growing up in a small rural town, a teenager listening to classical music and enjoying it was bizarre. I sometimes wished I could find a way to truly enjoy what “everyone else” was listening to. I figured one entrée would be to find popular musicians whose music shared some degree of similarity to classical. That didn’t really begin until I started university and met people with broad musical tastes. Since that time, I’ve stumbled over the years upon instances of direct classical influence on popular music. Even beyond “crossover” music that explicitly tries to combine classical and popular sensibilities, to varying degrees of success, they are too numerous to list here.
To further complicate things, I didn’t (and still don’t) like all classical music. My listening interests have tended to gravitate towards the “sound” of certain German composers from between the mid-19th and early 20th centuries.
So what do I mean by the “sound,” without getting too technical? Complex, layered instrumentation. A wide melodic palette. Thunderous percussion, especially in the kettledrums. Strong brass. Interesting harmonies, such as the “Tristan” chord from Wagner’s revolutionary opera Tristan und Isolde.
And those are just a few examples, which can also be found in other classical music that I like.
I gradually discovered what I took to be similar features as well upon encountering a wide range of popular music over the years. The more traits I perceived in a popular song that sounded like my favorite classical music, the better. No recommender system helped with yielding such moments, which were more serendipitous than anything else. Furthermore, efforts to figure out direct influences have been in vain. While frustrating, it makes the notion of sufficient similarities across music from different genres even more interesting. I discuss some personal observations regarding a number of them on my personal blog, too.
Nonetheless, over the years, I puzzled as to why my tastes seemed so idiosyncratic. Why I more or less knew what I liked when I heard it, even if I felt at a loss for textual description initially. Eventually, it came to me that this could be an information-related issue: Regardless of whether two pieces of music come from “very different” genres, how do (or how could) listeners actually perceive similarity, and how might that inform the process of finding new music? Since that time, whenever I’ve mentioned my research interest to others, I typically receive an enthusiastic response that indicates personal resonance with their listening interests. As someone once described it to me, their tastes were “all over the map.” So even if the possession of diverse musical tastes seems “random,” maybe there are good reasons why this occurs.
4. Other Ways of Accounting for Broad Musical Tastes
Of course, the ambiguities and confusion of genre categorization mentioned earlier could provide a plausible explanation. But thinking about music more broadly, at least within a “Western” context, all kinds of music draw upon many of the same principles. The 12-note, or 12-pitch, musical scale, carrying with it an array of musical traits, acts as a kind of foundation for musical “universality.” Or, perhaps more accurately, listeners’ understandings of that scale do so, at least in the sense of extramusical meanings they attach to the appearance and conglomeration of particular traits. That said, they might not be consciously aware of the degree to which they’re constructs. (I usually forget that myself as a listener, until I put on my scholar’s cap.) Of course, the term “universality” can be easily misused, and its nuances misunderstood. In the interest of conserving space here, and to understand what I mean, please see my commemorative posting about the 40th anniversary of conductor and composer Leonard Bernstein’s Norton Lecture series at Harvard.
Whatever the case, listeners’ understandings of what musical traits “mean,” whether heard individually or together, could also account for similarities across different genres. Prospective clues could come from the “facets” identified by Downie (2003) in our own field, as well as others identified in music psychology (Gabrielsson, 2009; Snyder, 2000; Wedin, 1972). Interestingly, there are some variations in terminology used among the aforementioned publications.
Another phenomenon worth mentioning is the emergence of “omnivorous” music listeners over the past century, as identified and discussed in papers by Peterson and Kern (1996), as well as Peterson and Simkus, (1992). This term refers to the ability of individual listeners to enjoy music that’s not typically associated with their social status, referred to in the study as low-, middle-, and highbrow. According to Peterson and Kern (1996), such factors as higher living standards, increased social mobility, broader acceptance of diversity, and the emergence of mass media seem to account for omnivorous tastes. There are some questionable assumptions, however, such as not including jazz (mainly due to its shifting social status associations) and not accounting for divisions among specific types of popular music. Nonetheless, the broader concept is useful for describing how people can enjoy music that might seem “very different.”
Additional factors could also account for broader musical tastes. In the 20 years since both “Peterson papers” were published, the ubiquity of the Internet has created more opportunities for musical omnivorousness. As well, in discussing the results of her dissertation research on young adults’ music information-seeking behavior, LaPlante (2008) describes how underground music could act as a marker of “legitimate taste,” regardless of whether an underground fan also listens to classical music. For further insights on what one could consider “popular” or “legitimate” music (or somewhere in-between), I recommend LaPlante’s discussion (2008, pp. 224-229) about sociologist Pierre Bourdieu‘s (1979 / 1984) research into the ways people use musical taste to distinguish themselves from others. More or less in line with the 1992 and 1996 “Peterson papers,” perhaps this (along with the traditional limitations of “physical” classification systems) accounts for why formal music categorization systems tend to draw upon genre, and generally do not acknowledge other traits that could indicate other forms of similarity.
When I started thinking about musical similarity as a research focus, I considered the potential implications of cross-genre similarity for recommender systems. Could they make suitable “cross-genre recommendations,” which go beyond the behavior- and market-based (and perhaps even status-based) limitations of current systems? A handful of studies have taken some tentative steps in this direction (Roos & Manaris, 2007; Hong, 2011), but nothing currently large-scale has emerged.
Many studies related to music information-seeking tend to focus primarily on systems, while very few (Cunningham et al., 2003; Inskip et al., 2008; LaPlante, 2008) ask users about their experiences. LaPlante’s 2008 dissertation comes closest to my proposed approach, with its focus on self-described avid recreational music listeners.
What distinguishes my study is a closer emphasis on how listeners define musical similarity. I suspect that it might be based on genre to some degree, but with some discrepancies possibly emerging as well.
I will conduct face-to-face semi-structured interviews with 15-20 respondents, who will also be asked to engage in an open-ended “think aloud protocol” exercise where they describe how they look for “similar” music. It will conclude with a brief questionnaire to see if responses might somehow correlate with demographic traits.
Personally, I’m quite keen on hearing responses regarding similarity and the “popular vs. classical” divide, especially with that conflict’s pervasiveness in the broader culture over the years (Ross, 2004; 2007; 2013). Nonetheless, the researcher in me thought it best to have no questions related to that. A fair number of respondents might not really think about the dichotomy at all, and I don’t want to use leading questions that would make my personal interest in the topic all too clear. Still, I do have a question at the end of the interview about noticing “surprising” similarities across different genres. It will be interesting to see how respondents conceptualize the question, and especially the word “genre.” Whatever they say, it should provide insights on how listeners perceive genre, along with an interesting point of comparison with “top-down” systems-based conceptualizations.
6. Implications and Applications
Although the research findings might have a variety of potential applications and implications, categorization practices seems like a fairly obvious place to start. By extension, depending on such factors as scalability, the findings could be used in developing recommender systems that go beyond the item- and user-based algorithms mentioned near the beginning of this posting. Of course, this doesn’t mean that such systems will act as a “solution,” as human judgment is still very much needed for something as individual (and yet still socioculturally informed) as musical taste. As well, the findings could be of interest in a variety of music-related fields, including music education and therapy.
Of course, the proposed study could act as a foundation for further research into listeners’ perceptions of musical similarity. Although the proposed study will draw upon self-described lay music enthusiasts of any background, future studies could focus on more specific demographic groups for points of comparison regarding music information-seeking practices.