Visualizing the Knowledge Space of Music · 2017-07-04 · Goal: Map the knowledge space of music I...

Visualizing the Knowledge Space of MusicIAML 2017, the 66th IAML Congress

Glenn Henshaw, Fan Yun

June 22, 2017

What is music scholarship?

I Music scholarship is hard to defineI Deeply interdiciplinary.I It’s a young field.

What is music scholarship?

Examples:

I Musicology: Richard Wagners opposition to animalexperimentation: A visionary social critic

I Ethnomusicology: A bird tradition in the west of the BalkanPeninsula

I Music pedagogy: From Mississippi hot dog to Arizonacactus

I Music therapy The use of music with chronic food refusal: Acase study

I Popular music studies Crossing cinematic and sonic barlines: T-Pains Cant believe it

Goal: Map the knowledge space of music

I How can the music be divided into subfields?

I What are the principle subfields?

I What are the emerging subfields?I How do these subfields connect/interact?

I InternallyI externally

Goal: Map the knowledge space of a given field

I Inskip and Wiering Conducted surveys of Musicologicalscholarship.

I Leech-Wilkinson considered community behavior from ahistorical perspective.

Goal: Map the knowledge space of a given field

I Expert surveys tend to be costly and slow endeavors.

I Expert bias.

Goal: Map the knowledge space of music

I Co-Word Analysis attempts to map the knowledge space ofa given field by measuring and analyzing the strength of theassociations between terms (keywords, indexed terms, orwords from a corpus)

I The strength of the association between terms is based onhow frequently they co-appear in documents.

I We use the results of co-word analysis to hierarchically clusterthe terms.

I We visualize the clusters with dendrograms, weightedgraphs, and strategic diagrams.

Visualization techniques

I Dendrograms: detailed information on the relationshipbetween terms and clustering.

I Weighted graphs: details of each cluster

I Strategic Diagram: Local and global view of the clusters

What is RILM?

I A comprehensive music bibliography featuringI abstracts and citationsI 143 languagesI 1967 – presentI 875,000 records

I RILM indexing represents hierarchical relationships withbroader and narrower topics.

The Data

I Almost all academic music articles (2000-2015)

I 263,656 Rows

I 59,908 Articles

I 25,297 distinct terms (lvl1)

15 Most common terms

Co-occurrence

DefinitionFor two terms, i and j , their co-occurrence is defined as

Ci ,j = How often i and j appear in the same article

Example

The terms ’aesthetics’ and ’popular music’ apear together in 123articles.

C’aesthetics’,’popular music’ = 123

From matrix to graph using the matrix CI Dots (nodes) are terms.I Lines between terms indicate that C > 4.I Colors indicate terms that are highly interconnected (clusters).

From matrix to graph using the matrix CI Despite its beauty, we could not obtain valuable information

from this graph!

From matrix to graph using the matrix C

I High frequency terms dominate the connections.

Cosine similarity

Cosine similarity (CS) is defined as

CS(i , j) =

√C 2ij

CiiCjj.

Example

Let

i = ’China’ j = ’popular music’k = ’mathematics’ l = ’scales’.

Then,

Cij = 151 CS(i , j) =√

1512/(5084 · 2388) = .04

Ckl = 15 CS(k , l) =√

152/(309 · 303) = .05.

Hierarchical Clustering (a rough sketch)

I Distance between individual terms: d(i , j) = 1− CS(i , j).

I At the start, each term, ti , is contained in its own clusterci = {ti}.

I Define distance between clusters D(ci , cj).I Many ways to do this

Repeat until there is only one cluster:

I Find two clusters of minimal distance, i.e. minD(ci , cj).

I Merge the clusters ci and cj into a cluster ci&j .

I Delete the clusters ci and cj .

We visualize Hierarchical clustering with a dendrogram.

Hierarchical Clustering (a rough sketch)

A

BC

D

E

Hierarchical Clustering (a rough sketch)

A

BC

D

E

A B

Hierarchical Clustering (a rough sketch)

A

BC

D

E

A B E C

Hierarchical Clustering (a rough sketch)

A

BC

D

E

A B E C

Hierarchical Clustering (a rough sketch)

A

BC

D

E

A B E C D

Dendrogram (49 Clusters)

0.00.51.01.52.02.53.03.5conservation and restorationreligious institutionsinstruments--keyboard (organ family)instrument builders--organsound recordingsperformers--popular musicperforming groups--popular musiceconomicssound recording and reproductionmass mediaadvertising and marketingsociologyrockpopular musiccultural studiesUnited States of Americablack studiessong textship hopsexuality and genderwomen's studiesmanuscriptsnotation and paleographychant--Christian (Western)source studieseditingSchubert FranzSchumann RobertBrahms Johannesinstruments--keyboard (piano family)accompanimentinterpretationChopin FrédéricarrangementsLiszt FranzBeethoven Ludwig vanMozart Wolfgang AmadeusHaydn Josephperformance practicetempoBach Johann Sebastianinstruments--keyboardcompositionelectronic music and computer musicsonoritymixed mediatimeDebussy Claudescienceaestheticsphilosophyanalysiscomputer applicationscounterpointcreative processStravinsky IgorSchoenberg Arnoldserialismmathematicstheoriststheoryintervalsmodalityscalespoetrytext settingrhythm and metermelodyharmonytonalitytextureformmotive and themeperceptionpitchsingingacousticsanatomy and physiologyvoicetherapychildrenpedagogycareersaudiencesAustraliaamateursmovement and gestureconductingaptitude and abilitypsychologyperformancemedicine--by topicinstrumental playingpracticing and rehearsingensemble playingfilm and videofilm music and television musicdirectors producers etc.performing organizationsperformers--conductingoperaVerdi GiuseppeWagner Richarddramaturgychoreographers

ajkovskij Pëtr Il'imusical theater--popularpremieresliteraturelibretto--by librettistimprovisationperformers--jazz and bluesjazzanecdotes reminiscences etc.academic institutionspedagoguesperformers--pianoobituariesmusicologistsperformers--voiceperformers--traditional and neotraditional musicperformers--organsocieties associations fraternities etc.--nat.CanadaGermanycultural policiespoliticswars and catastrophesreceptionFrancecriticismperiodicalslibraries museums collectionscorrespondencesong--popular and traditionalLithuaniaethnomusicologyethnomusicologistsHungaryBartók Bélasong--by composerdiscographiescatalogues and indexes--by nameinstrumental musicvocal musicgenre studiessemioticstwenty or twenty-first-century musicsemanticssong--generalUnited Kingdom--Englandreligion and religious music--Christianity (Protestant)nationalismCzech RepublicBrazilidentity--collectiveSpaininstruments--string (guitar family)AustriaItalyreligion and religious music--Christianity (Roman Catholic)religion and religious music--Christianitypolyphonychoral musicperforming groups--small ensemblesinstruments--string (violin family)performers--violinperformance venuesRussian Federationfestivalscompetitions awards and prizesmusicologycongresses conferences symposiaChinadramatic artsinstruments--mechanicalinstruments--collectionsGreecetemperament and tuningantiquityinstruments--percussion (bells)dancedance musicIndiainstruments--percussion (drum)ceremoniesinstruments--wind (free reed family)terminologylanguageWestern musichistoriographyKoreaEuropeAsiainstruments--string (zither family)instruments--string (lute family)Japantimbreinstrumentation and orchestrationinstruments--wind (woodwind flute family)instruments--wind (brass)instruments--wind (woodwind reed family)iconographyinstrumentsvisual and plastic artssymbolismnature

Density and Centrality

For each cluster we calculate:

I Density - Average strength of all the connections within acluster

I Centrality - The square root of the sum of the squares of allconnections to outside clusters

Strategic Diagram

Strategic Diagram

Centrality

Density

Strategic Diagram

Centrality

Density

Cluster: Form

motive and theme

texture

harmony

form

rhythm and meter

tonality

melody

Strategic Diagram

Centrality

Density

Cluster: Popular Recordings

performers--popular music

performing groups--popular music

sound recordings

Strategic Diagram

Centrality

Density

Cluster: Pedagogues

anecdotes reminiscences etc.

performers--piano

pedagogues

academic institutions

Strategic Diagram

Centrality

Density

Cluster: Rock

sociology

rock

cultural studies

popular music

Strategic Diagram

Centrality

Density

Cluster: Mechanical instruments

instruments--collections

instruments--mechanical

Strategic Diagram

Centrality

Density

Cluster: Classical

Haydn Joseph

Mozart Wolfgang Amadeus

Beethoven Ludwig van

Strategic Diagram

Centrality

Density

Cluster: Organ

instruments--keyboard (organ family)

instrument builders--organ

religious institutions

conservation and restoration

Strategic Diagram

Centrality

Density

Cluster: Violin

performing groups--small ensembles

instruments--string (violin family)

performers--violin

Strategic Diagram

Centrality

Density

Cluster: Antiquity

instruments--percussion (bells)

antiquity

temperament and tuning

Greece

Strategic Diagram

Centrality

Density

Cluster: Performance

ensemble playing

performance

instrumental playing

psychology

medicine--by topic

practicing and rehearsing

aptitude and ability

Strategic Diagram

Centrality

Density

Cluster: Analysis

computer applications

counterpoint

creative process

Stravinsky Igor

analysis

Strategic Diagram

Centrality

Density

Cluster: Theory

modality intervals

theory scales

mathematicstheorists

Strategic Diagram

Centrality

Density

Cluster: Black studies

sexuality and gender

women's studies

hip hop

United States of America

song texts

black studies

Strategic Diagram

Centrality

Density

Cluster: Iconography

instruments

visual and plastic arts

nature

iconography

symbolism

Future Work

I Use more of the indexing string

I User data

I Mapping the movement of clusters through the strategicdiagram over time

I Produce an online tool for scholars to browse

Thank You!