Technology | Jason Lee Oakes| April 13, 2017
EBSCOhost® is an ideal resource for exploring music scholarship. But can EBSCO be searched “musically?” And what might be the advantage of doing so?
These questions and more are explored in my webinar Mastering the Mix, viewable in its entirety below. In this blog post I will summarize, and expand upon, a number of points from the webinar. While I’ll be using RILM Abstracts as a case study, the ideas presented here could be applied beyond music-centered databases.
Academic databases have two key components: data and algorithms. The data is the information compiled in the database, while an algorithm is a sequence of procedures applied to the data with a certain end result in mind. Algorithms are often designed to create order out of seeming chaos.
A third component of the academic database, the search engine, mediates between the two components above. The search engine accepts an input from the user, selects an appropriate algorithm to search the database and outputs targeted data.
Algorithms are just as crucial to music as they are to search engines. To comprehend a given musical system is to comprehend the musical procedures applied to sound that separate the resulting sounds from random noise, and that separate one musical genre from another in stylistic terms. But this is not to say that most musicians follow rote instructions, of course, which would make for a lot of boring music.
Instead, musical algorithms are constantly elaborated, amended and re-invented. According to cognitive musicologist Robert Jourdain, it’s just this dynamic interplay between algorithmic regularity and human creativity that lies at the heart of music’s emotional impact — engaging the listener through a “flux of musical anticipations.” Likewise, Oliver Sachs has argued that the transformative effect music has on many victims of Parkinson’s disease is rooted in music’s interplay between sequences of procedures and human agency.
This interplay has taken on added complexity in the digital era, now that a vast array of new algorithms has entered the picture. Digitized music, which takes sound and transforms it into digital data (all ones and zeros), lends itself to all kinds of algorithmic manipulations that wouldn’t have been possible before. These algorithms impact both the production and the consumption of music as demonstrated in this video.
One digital algorithm that changed music for a lot of people was the Napster algorithm. The peer-to-peer file-sharing service quickly turned into the world’s biggest music swap meet between 1999 and 2001. Using a simple algorithm that searched the metadata of millions of digital files, Napster popularized the idea that music is something to be widely disseminated and acquired online. And from that point forward there was no turning back. Like other truly game-changing algorithms, Napster struck a magic balance between automated data algorithms and the subjective human touch. It was Napster that first got computer users familiar with file sharing and social networking on a mass scale. From then on, the public’s overall comfort level with sharing intimate details of their lives on the Internet, building distinctive online identities in the process, shot up steeply. With Napster, information exchange was made musical and, as such, more deeply human.
Long before huge databases of music recordings could be shared and filtered with such efficiency, academic databases like RILM Abstracts developed a similar approach to information about music. Drawing on a “peer-to-peer” network of shared music research, today there are nearly a million records about music in RILM Abstracts, searchable through the EBSCO interface. But how can searches of this massive database be made as “musical” as possible, quite apart from the content itself? Taking a page from Napster and from other digital music algorithms, how can we best enhance the quality and the impactfulness of information retrieval in academic databases through increased musicality?
A good starting point can be established through a simple observation: Music is defined by relationships. A single note doesn’t mean much in isolation. Even Tuvan and Mongolian throat singers subtly alter timbres/overtones over time to make a “single note” musical. In the broadest possible sense, then, music acquires meaning through how notes are arranged relative to other notes: arranged pitch-wise in relative intervals to form melodies and harmonies; arranged relative to time through structured rhythms, metrical systems and other temporal modes; and through the relative arrangement of voices and instruments to create compelling timbres and textures. Musical meaning is also found in how humanly-organized sounds are used to organize people — acting as a powerful symbol for cultural identity, social belonging, individual uniqueness and other methods of negotiating human relationships.
Moving from music itself to music scholarship, database search results are usually at their most effective and appealing when a query is posed in relational terms. Taking an inverse example at first, if you search RILM Abstracts for records on “popular music” with Major Topics chosen from EBSCO’s pull-down search menu, more than 60,000 records are returned. The search result isn’t likely to be “effective or appealing” to anyone due to its single-note quality and the lack of focus that results.
But now let’s try turning this into a multi-parameter search. One quick, easy and useful parameter that can be added to the mix is “Full Text.” By clicking the Linked Full Text box on the left side of the screen, only records with attached PDFs are returned, saving the user a trip to the library in the process. At the time of writing, this search returns more than 5,000 entries. It’s still a large number, but a lot less than 60,000 and the content is at least just a click away.
From here it’s easy to take more steps to get a more “musical” search result by throwing more parameters, and thus a broader array of relationships, into the mix. Adding an EBSCO Subject parameter to the parameters already chosen, the search is narrowed to records where the chosen word or phrase appears in RILM Abstract’s indexing for a given record. For instance, choosing “heavy metal” as the subject returns around 100 full-text citations, a much more manageable number than 5,000.
Most important of all, the results are musical. They strike a useful balance between uniformity and diversity, a balance likewise found in music that strikes an aesthetically appealing balance between repetition and variation. While all the records in the dataset are uniform in addressing heavy metal directly and thoroughly, there’s a good bit of variation otherwise: spanning writings that examine “metal studies” as an academic field, sonic traits of drone metal in light of genre theory, the sociology of Caribbean heavy metal scenes, and perceptions of sexuality and gender around female metal fans, among many other topics.
From EBSCO to Excite, the ultimate goal of most search engines is to return a good mix of results. This helps explain the shift from the directory model of first-wave search engines like Yahoo Directory, to the second wave of webcrawler search engines (Google most famously) that utilized algorithms to locate sites, collect metadata and build an index. I would submit that the latter won public favor due to two main factors: 1) it was more likely to deliver exactly the results the user was looking for (indexes are more granular than top-down categories); and 2) it was more likely to return unexpected results.
Needless to say, random and irrelevant results are not widely desired. They are equivalent to “wrong notes” in a melody and just about as popular. Instead, results that provide a novel yet purposeful perspective on a query are often the most impactful — like the surprising yet logical-after-the-fact twist in a melody that serves as the “hook.” Returning to the example of Napster, it hooked users not just because it found the music they already knew they wanted, but also because they ended up discovering new and unfamiliar music they went on to fall in love with — often by searching laterally through a given user’s music collection (equivalent “lateral” searching in RILM index is discussed in the webinar). This mix of the familiar and the novel is a sure-fire formula for a successful search interface.
With this in mind, the digital-age database manager must work to be a master of the mix — all the more so when it comes to popular music studies and other interdisciplinary fields. The popular music researcher is sure to need materials published in non-music journals and publications. What’s more, she is likely to seek out other important data strewn across magazines and fanzines, posted on blogs and other websites and located across a range of other non-traditional sources. As a result, one major initiative at RILM taken up lately has been to seek out and compile more of these “outside the box” materials, curated for potential use value as primary and/or secondary data.
Given the risk of information overload that comes with the widening and the blurring of traditional boundaries, effective curation becomes all the more important. Approaching a database from a musical point of view offers a step in the right direction. Editors at RILM and at other databases are increasingly placed in the role of “record collectors” who don’t just “collect” but who also filter, organize, and interpret the data we collect. Like the crate-digging DJ, we dedicate ourselves to digging for data and creatively integrating new materials. This DJ mindset also highlights the necessity of working across various old and new media and of delving into unexplored spaces to find hidden gems.
As there isn’t enough room for me to get into the second half of the webinar here, I provide a few pointers below to the main questions addressed. Thanks for reading and watching!
Jason Lee Oakes oversees popular music literature at RILM Abstracts. He holds a PhD from Columbia University in Ethnomusicology. Dr. Oakes was selected to be one of two co-editors of the Oxford University Press Handbook of Hip Hop Music Studies. His extensive list of other writings includes contributions to Current Musicology; The Ashgate Research Companion to Popular Musicology; Queering the Popular Pitch; and Access All Eras.
Your comment will be reviewed by a moderator for approval.