Examining similarities and differences in genre classification for identical album data in MARC records versus last.fm

 

This project evolved out of an internship at Columbia University's Music and Arts Library during which I worked with my supervisor to create original MARC records for hip-hop LPs. Throughout this process, we discovered several challenges to the bibliographic description of these materials via current cataloging standards.

 

what's the problem?

Genre and subgenre: Hip-hop, Rap, and everything in between...

 

Currently the primary LC genre/form term being used to describe these materials is Rap (Music). Hip-hop is not available in LCGFT, it is only available as a variant term, which means a search for hip-hop in subject fields will redirect to Rap (Music). According to many knowledge bases, including Encyclopedia Brittanica, the terms hip-hop and rap are not synonymous with each other, even though the LCGFT seems to structurally imply they are interchangeable. Additionally, the role of producer is difficult to accurately describe in RDA for MARC records. It is also difficult to assign subgenres to differentiate between various styles. I discuss these challenges in greater depth via this post.

Addendum: When creating this project, the only genre/form term available to describe these materials was Rap (Music). Since the creation of this project, a new genre/form term has been added: Hip-hop (Music).

 

ye olde taxonomy vs folksonomy debate

 

I developed the idea to use python script to examine library cataloging practice (taxonomy) versus user-generated metadata (folksonomy) for hip-hop and rap music. I chose to compare subject fields (650 and 655) for albums classified as Rap (Music) in MARC records versus last.fm tag metadata for the same albums.
 

Last.fm is unique in that it is based in a community of listeners, and provides a customizable listening experience. Last.fm uses a plugin called Audioscrobbler which records users' listening habits and compiles data based on their listening history. Listener data is utilized to recommend similar artists, albums, and songs. Users can tag albums and artists using existing tags or new tags, and listen to the radio stations generated from the use of those tags. The wiki definitions for tags can be edited by community members.

 

process

The comparison process began with the search of album data from MARC records, narrowing down the list to records classified as Rap (Music) in sound disc format (both LPs and CDs). Then, the album and artist information was used to query the last.fm API, searching the top tags associated with these albums. 

The graph on the left illustrates the additional terms used by catalogers to modify the albums placed under a Rap (Music) genre/form heading. The graphs are color coded to show how genres and subgenres within the last.fm tags correspond to the LC subject fields.

 

issues to note

The MARC record data is from 2014, and may not reflect current record numbers. The data in the left graph is only for the creation of records, not for the amount of items that are copy-cataloged using the same record.

I limited the range for the LC subject fields to at least 28 counts (28 MARC records using that term), and for the last.fm tags at least 500 counts (500 uses of the tag).

 

Some of the MARC records were for lesser known albums, or had nontraditional spellings for the artist and album. These incongruences were difficult to reformat with code and resulted in an "album not found" error message. As a result, no last.fm tags were generated for those albums. This error represented about 3% of the albums (346 out of 10,464). 

 

which classification method is better?

There are merits and drawbacks to both means of knowledge organization. LC vocabularies are guaranteed to be based out of a lengthy process of peer review to assure accuracy and compatibility with overall cataloging practice. Given the complexities of this process, it is more difficult to establish best practices in keeping with current trends.

 

With last.fm tags, it is easier and quicker to establish categories in keeping with current trends. Tags don't require a lengthy, scholarly process, but still have a significant base in collective knowledge. In this regard, the folksonomy of last.fm is not so different from the scholarly processes within the cataloging sphere. Both are based in community, several voices agreeing upon a label for a particular music format.

However, a drawback of last.fm tags is the lack of standardization.

For example, the disparate spellings of 'hip-hop' results in misrepresentation of numbers. For these albums, the count for the use of hip-hop (in some variation) is actually 68,555 (surpassing the count for rap at 57,467).

 

the producer question

Interestingly, the last.fm tags also provided insight into the 'producer' issue. 

 

One of the ways last.fm users connect the contributions of a particular producer or production team is by the use of tags. For this collection of album data, some of the most widely used tags were for particular producers, production teams, or record labels. Many record labels utilize the work of specific producers who contribute to the distinctive style or sound associated with that label. Indeed, all these facets are part of the stylistic differences which formulate the basis for subgenres of hip-hop and rap music such as the G-funk sound associated with the production of Dr. Dre, and often associated with West Coast rap in general. All of these facets are aptly encompassed in the tag data compiled from last.fm, and could readily be cross-referenced via other music metadata sources such as Discogs and Spotify, sites which utilize other types of music metadata organization.

 

data trends

In the categorization of these albums, the last.fm data indicates greater volume of usage for 'hip-hop' versus 'rap' tags.
 

last.fm:
East coast versus west coast rap is clearly emphasized in the last.fm data. Other geographic locations like 'bay area', 'memphis rap', 'dirty south' and 'New York' are frequently used. Several categories relating to the year or era, including 80s, 90s, old-school, golden, or classic era are utilized. Female rappers and vocalists are given unique signifiers. Tags are utilized for conscious and political rap, gangsta rap, and r&b and soul-influenced rap and hip-hop.

 

MARC records:

It is also clear by looking at the albums labeled as Rap (Music) in MARC records that the rapping vocal style veers into other types of music that are outside the sphere of hip-hop, such as various types of metal genres. This gives further evidence that rap and hip-hop are not equivalents, despite the fact that hip-hop is listed as a variant term of Rap (Music) in LCGFT.

 

At least 500 uses of the last.fm tag warranted inclusion in the last.fm graph, though I excluded terms like "favorite albums" and "fun to skateboard to". I briefly considered including some tags like "excellent lyricism" because it reflects the categorization needs of the user community, but decided against it.
 

conclusion

The last.fm folksonomy represents a radically different approach to the organization of music metadata, as it is focused in a grassroots listener community versus the hierarchical, scholarly basis of library categorization. However, the sheer volume of listener usage for these tags makes them worthy of consideration when it comes to the increase in discoverability which is afforded by precise description. As catalogers at University of Michigan have demonstrated, music metadata from external sources like last.fm can prove useful in the creation of accurate library records.