Measuring cultural value using social network analysis: a case study on valuing
electronic musicians
Anna Jordanous
School of Computing
University of Kent, UK
a.k.jordanous@kent.ac.uk
Daniel Allington
Department of Arts & Cultural Industries,
University of the West of England, UK
Daniel.Allington@uwe.ac.uk
Byron Dueck
Department of Music
Open University, UK
byron.dueck@open.ac.uk
Abstract
In evaluating how creative a program or an artefact
is, a key factor to consider is the value inherent in
that program or artefact. We investigate how to measure
subjective, cultural value: value which has been
expressed by members of a community towards other
members. Specifically we focus on a case study asking:
to what extent can we use social network activity
to examine the value that electronic musicians place in
each other’s work? Focusing on activity by electronic
musicians on the music social network SoundCloud, we
combined qualitative and quantitative research to understand
and trace significant ‘valuing activities’ in SoundCloud
data. Exploring interaction on the site in this
guided way has enabled us to compare, contrast and assess
what value is attributed to different members of the
electronic music community on SoundCloud. In this
paper we report our results and consider how this work
offers a methodology for computational analysis of cultural
value. We hypothesise that this methodology is extensible
to other creative domains; potentially this could
lead to a tool for automated cultural value judgement
methods on large social network datasets. Hence we
move towards computationally generated evaluations of
value, a fundamental part of creativity.
Keywords: Social network analysis, value metric, evaluation
Introduction
How can we measure the value of creative entities to a community?
(especially unquantified value, expressed through
esteem rather than money?) And how could such value
judgements be automated across large amounts of data and
implemented within computational systems?
Value judgements are a vital part of creativity; the usefulness
or value inherent in a creative system and what it does
is intricately connected to how creative it is (Ritchie 2007;
Jordanous 2012a). In computational research on creativity,
we would like our systems to be able to perform evaluation
of their own processes. Autonomous judgements of value,
integrated within a computational system, are desirable but
only occasionally realised in computational creativity.
Value itself can be difficult to identify and measure. In
particular, a distinction exists between the more easily identifiable
economic value of creative works and their producers,
compared to their inherent and intangible cultural value.
Cultural value is attributed through peer interaction and underground
expressions of esteem rather than measures such
as sales of artefacts or ticket sales. There is a ‘relative independence
of a status order built from peer esteem from one
built purely upon popularity or sales’ (Lena and Pachucki
2013, 239). For example, electronic music is a creative domain
consisting of many underground subcultures, where
economic or popular recognition is often not achieved and
quite often not even pursued to any great degree. Value attributions
become difficult to recognise due to lack of official
recognition or monetary reward for electronic musicians.
So how do you measure or evaluate cultural value? Here
we address this question through a case study on electronic
musicians. In the Valuing Electronic Music (VEM) project1
we investigated how electronic musicians show their appreciation
and value for other musicians, via qualitative interviews
and quantitative research around SoundCloud,2 a social
network for musicians (particularly for electronic musicians).
Our aim was to gauge how value is attributed and
recognised through interactions between electronic musicians.
In particular, we wanted to identify features of interartist
networking and peer evaluation contributing to value
production that are detectable in quantitative analysis of digital
interactions. The main aim relevant to computational
creativity is to determine what computational analysis could
be performed as a proxy for cultural value.
We argue that the approach developed in the VEM project
is adaptable to assessment of value in a range of cultural contexts.
We offer a method for empirical evaluation of cultural
value through analysis of social interactions.
Value evaluation in computational creativity
Where evaluation of computational creativity systems includes
some value judgements, objective metrics have to be
carefully selected to ensure value is evaluated in an appropriate
and representative manner. In computational work,
though, objective metrics have key advantages over subjective
data collection, which can be time consuming (especially
if collating user feedback) and problematic in terms of
identifying representative samples of users. Also, it is dif-
ficult to integrate such testing within a computational sys-
1
See http://valuingelectronicmusic.org
2
http://www.soundcloud.com
Proceedings of the Sixth International Conference on Computational Creativity June 2015 110
tem’s processes and respond to the feedback, particularly
if system testing is carried out towards the end of research
projects. But there is a need for autonomous value judgements
that could be integrated within computational creativity
systems; creativity is not just about new work but also the
development and refinement of this work (Boden 2004).3
The term value encompasses many different aspects such
as appropriateness, relevance, usefulness, correctness, worthiness
and/or quality. A minimum (probably insufficient)
definition of creativity could be novelty+value (Jordanous
2012b). Jordanous (2012a) defines value as:
• ‘Making a useful contribution that is valued by others
and recognised as an influential achievement;
perceived as special; “not just something anybody
would have done”.
• End product is relevant and appropriate to the domain
being worked in.’ (Jordanous 2012a, 258)
In his discussion of value, Ritchie (2007) makes extensive
use of value ratings but leaves open what type of method
should be used to generate these ratings. Domain-general
heuristics for value judgements are difficult if not impossible
to identify; value is relative to the domain and is embodied
in different ways. For example, accuracy is vital for mathematical
proof generation systems, (Colton 2008) but not for
creative musical improvisation (Jordanous and Keller 2012).
One of this paper’s authors recently reviewed evaluation
of computational creativity systems (Jordanous 2011). She
found that 43% of papers containing some content on system
evaluation aimed to evaluate the value, quality or appropriateness
of the system or system’s output. Many types of
empirical value measurements were found, as well as value
measurements based on user feedback. The value of a creative
system entails more than the value of its products; but
this perspective was not evident in Jordanous’s review. Typically,
systems were evaluated based on the value or validity
of the artefacts they produce, e.g. statistical tests for validity,
calculations of how fit-for-purpose material produced
during runtime was, how interesting their products were, or
other domain-specific indicators of validity or value.
Social and cultural value, particularly in music Blacking
noted that the existence of musical geniuses such as
Bach and Beethoven is reliant on the presence of a discriminating
audience (Blacking 1973). We push this viewpoint
further: the relationship between audience and musical
performer is both vital for appreciating musical value
and the division between audience and musical performer
can be blurred. Turino (2008) contrasts ‘presentational’ musics,
based around the quality of works and performances,
with ‘participatory’ musics, where value is within the quality
and intensity of social interaction. Turino reminds those
in a Western Classical musical mindset of a vital aspect of
music: the collective, participatory social aspect of musical
3
Some computational creativity researchers use evaluation in
the processes of creative systems (Perez y P ´ erez, Aguilar, and Ne- ´
grete 2010, engagement-reflection), (McCormack 2007, evolutionary
computing), (Pease, Guhe, and Smaill 2010, generate-and-test).
experiences, especially when incorporating collective listening,
composition, performance and dancing. For example,
social interaction and communication are key for creativity
in musical improvisation (Jordanous and Keller 2012).
Csikszentmihalyi (1988) proposes a systems model of
creativity as a dynamic process of interaction between Domain,
Field and Individual/Person:
‘[creative] is the product of ... a set of social institutions,
or field, that selects from the variations produced
by individuals those that are worth preserving; a stable
cultural domain that will preserve and transmit the
selected new ideas or forms to the following generations;
and finally the individual, who brings about some
change in the domain, a change that the field will consider
to be creative.’ (Csikszentmihalyi 1988, 325).
Csikszentmihalyi’s emphasis on interactions between domain,
individual and field (Csikszentmihalyi 1988), can be
situated within the broader area of field theory (Bourdieu
1993) where producers compete for recognition rather than
financial gain. Bourdieu posits that all agents involved in
music-making form part of the musical communities that attribute
value to music-making activities, regardless of level
of ability or profile. So ‘hidden musicians’ (Finnegan 2007)
(everyday music-makers who are key to the musical life of
communities but understudied by scholars and publics) play
a significant part in determining who and what is valuable
within musical practices (Dueck 2013). Exploring how hidden
and star musicians link together in networks of evaluation
and commentary lets us see how all depend upon one
another, jointly producing the cultural context in which their
music can have value. Although Bourdieu focused on what
he termed ‘legitimate’ culture (i.e. serious literature, art music,
etc), his ideas have since been adapted to other cultural
forms e.g. Lopes 2000 (jazz), Elafros 2013 (hip-hop).
Social networking and new media websites have provided
music makers with new spaces in which to negotiate and
produce cultural value for their work, taking on tasks that
would once have been the sphere of specialists in marketing,
publicity and criticism. These phenomena appear to
have had a particular impact on electronic music, which is
typically made by lone, but highly networked, individuals
and is often circulated non-commercially online. A recent
report across UK-based professional musicians found that
64% ‘us[e] web-based technologies to produce, promote,
and distribute their music’ (DHA Communications 2012).4
De Nooy argues that social network analysis can legitimately
‘be used to gauge the amount of... symbolic capital’
(De Nooy 2003, 325). De Nooy’s proposed approach
to the study of symbolic capital had been successfully implemented
as a methodology for studying the production of
cultural value by one author of this paper (Allington, under
review). Allington used data harvested from online sources
to study the production of value within Interactive Fiction
(stories that develop in plot through user interaction). Centrality
measures were used to assess the level of value associated
with specific creators working within that community
4
This figure may be higher for electronic music, which typically
attracts music makers highly familiar with digital technology.
Proceedings of the Sixth International Conference on Computational Creativity June 2015 111
(Allington, under review). Allington’s methodology formed
the starting point of the present project (complemented with
ethnographic research). We scale up from de Nooy’s work
with tens of producers and Allington’s with thousands, to
hundreds of thousands of users in the current work.
Identifying cultural value in electronic music
Looking specifically at electronic music, the Valuing Electronic
Music project investigates how we can gauge what
cultural value electronic musicians hold. With the above discussions
guiding our work, we looked at how peer groups of
electronic musicians showed appreciation of each other. Our
quantitative work focused on tracing activities for ascribing
value to users, through network analysis on large collections
of data. This paper reports the project’s findings, from
the perspective of developing a methodology for empirically
identifying and evaluating cultural value (that could in future
be incorporated autonomously in a creative system).
Partly inspired by successes using social network analysis
to make proxy judgements about value within a network
of Interactive Fiction writers (Allington, under review), the
research focuses on interactions between creative producers
on the music social network SoundCloud, aggregating peer
evaluations and tracing the production of value.
Our approach to cultural value judgements Our quantitative
research centred around collecting and analysing data
from SoundCloud’s API, about how users interacted with
each other on SoundCloud. SoundCloud provides a good
data source for technical reasons (a well-developed API provides
access to all public data), for social reasons (it is
widely used by amateur, semi-professional, and professional
electronic musicians for networking and publishing music),
and for ethical reasons (the data is clearly marked to site
users as public). This sits in contrast to sites such MySpace,
which has declined in popularity.
We initially collected data on all demographic information
and activities that SoundCloud made public, with the
intention of using our qualitative data to understand the relative
importance of each activity. Demographic data that
users had made publicly available include their location,
URLs and avatars relating to their online profile, number
of followers, details of record labels they were attached to,
etc.5The activities that we collected user data for were the
publishing of tracks, following and being followed by other
users, liking a track, commenting on a track, creating personal
playlists of tracks and creating or joining a group.
While the project was primarily a study of online data,
this study was contextualised and enriched through study
of SoundCloud users in the offline environments in which
they primarily perform. In particular, our initial research on
SoundCloud suggested there existed more-or-less closelyknit
communities of co-located producers of electronic music.
This implies that, even in the apparently transnational
world of electronic music and online distribution, the social
5
More details at the SoundCloud API documentation at
http://developers.soundcloud.com/docs/api/guide and our github:
http://www.github.com/ValuingElectronicMusic/network-analysis.
production of value may still be influenced by localised realtime
face-to-face interactions. Hence ‘offline’ qualitative
work was conducted alongside our quantitative work, with
each mode of research guiding and influencing the other.
We interviewed eight electronic musicians, representing
various different types of musicians in different genres from
grime to techno. We also attended three electronic music
performances and made observations, and interviewed
a panel of three musicians at a public event we organised
in London in June 2014. Informing our qualitative research,
we also actively engaged in the SoundCloud community e.g.
‘liking’ tracks we enjoyed and following musicians. The interviews
helped us to explore the performers’ perceptions of
value. Using semi-structured interviews allowed us to cover
areas of interest such as how the interviewees valued other
people’s music, while allowing the interviewee to guide the
conversation towards areas they felt important. Observation
data from gigs (e.g. order of appearance of various performers,
prominence of performers’ names on promotional
materials, audience behaviour, etc) informed the interviews
themselves as well as providing much-needed context for
our relatively abstract online data.
A common theme emerging from our qualitative research
was that rather than searching for value (as an entity to measure),
we should be focusing on valuing activities. Actions
by and interactions between musicians were reported by interviewees
as a vital way in which they perceived that people
appreciated them and their work. Similarly in observations
during gigs and in specific questions to live performers, we
often noted the importance attached to people’s body language
and responses to music. In these electronic music
communities, the status attached to people also affected to
what degree any valuing activities were. In particular, our
interviewees typically gave higher credence to interactions
with other musicians, compared to those with non-musicians
(or those perceived as a non-musician, for example if their
reputation as a musician was not known by the interviewee,
if they had not mentioned their own musical activities during
the interactions or if they had not included pointers to
their own work in their SoundCloud profile or other online
profiles). This is similar to Bourdieu’s emphasis on cultural
producers’ esteem for one another’s work (Bourdieu 1993).
Data Collection We wrote code in Python to collect public
data automatically from SoundCloud, using the SoundCloud
API and Python SDK.6 It was impractical to study the
entire network of users, which comprises tens of millions
of accounts, many of them inactive or controlled by bots,
and huge amounts of data to collect. We initially adopted a
snowball sampling method: starting with a seed individual,
collecting data for the seed and the individuals they are connected
to, then collecting data for the individuals connected
to our seed’s connections, and so on). However, we encountered
problems with this approach due to SoundCloud network
structure and sheer density of data. Many millions of
6
This code is open-source and available at
http://www.github.com/ValuingElectronicMusic/network-analysis
- it is built from existing code by Allington for social network
analysis, also available from the ValuingElectronicMusic github.
Proceedings of the Sixth International Conference on Computational Creativity June 2015 112
users would frequently be found within just two degrees of
separation of a single individual. Undeclared restrictions
placed by the SoundCloud API on downloads of information
meant that we were prevented from collecting full data
on all of those people, with an upper limit of 8199 in place.
For example, if a given user had over 100000 followers, one
would be unable to discover the identity of more than 8199
of them.
Following discussion with experts at a workshop organised
as part of the project, we decided to adopt a different approach.
We switched to a two-fold data collection approach
of (i) a sample of 150000 randomly selected SoundCloud
users and (ii) ego-networks consisting of the networks of
users around our interviewees and their followers/followees.
In each case, we collected all publicly available data about
each user, along with data on all tracks uploaded by these
users and those who followed them. Due to the download
restrictions of the SoundCloud API we could only download
up to 8199 items of data per information request, but
in practice this only affected data collection for a very small
number of highly popular SoundCloud users. Some minimal
data cleaning was needed, mainly for reconciling locations
of users where different people used different variations of a
location name (e.g. Cairo and Al Qahirah, or NYC and New
York), or neighbourhoods within cities rather than cities.
Genres of electronic music are varied and broad, including:
house, trance, techno, trap, EDM, ambient, grime, etc.
Initial research showed that while the predominant types
of music on SoundCloud are in electronic music genres,
SoundCloud tracks are often tagged as belonging to a subgenre
of electronic music, rather than as ‘electronic’. To
locate data corresponding to electronic musicians, we could
not merely search for those who published music tagged as
‘electronic’, nor would it be appropriate to treat all electronic
music genres as belonging to one community (as con-
firmed by our interviewees). Instead we made use of the fact
that most musicians actively participating on SoundCloud
(uploading music, interacting with other users) were electronic
musicians. In our data collection, then, we collected
data on randomly chosen musicians such that we could later
filter the data by genres or other pertinent factors (to be informed
by our qualitative research).
Working out what data to look for In interviews, we
asked if there were valuing activities the participants would
highlight as important on SoundCloud, and if so, which
ones. In general, even minimal acts of valuing such as playing
someone’s track were considered to have some value.
Participants highlighted indication of a longer term public
support base via number of followers, and the use of the
commenting facility for people to leave messages on individual
uploaded tracks. Further, participants valued activities
which arose from or led to offline connections and
collaborations, although this type of activity is difficult to
track quantitatively.7 Activities such as playing or ’liking’
7
Collaborations between two SoundCloud musicians are tricky
to detect in SoundCloud data, as tracks on SoundCloud can only be
attributed to a single creator. Tracks with two or more associated
creators tend to either be uploaded by one of the collaborators with
someone’s track or including a track in a personal playlist
or group were not highlighted, possibly because it is less
easy to trace the provenance of this kind of valuing activity
to individual musicians and hence less easy to judge the
credibility of the person being interacted with.
The facility to follow and be followed by other SoundCloud
users was widely used by users, and afforded analysis
or user interaction on a wider scale than at the level of individual
comments, allowing us to detect general trends in
much larger samples of data. While the follow activity does
not require much engagement compared to making a comment
on someone’s track, nevertheless this activity identifies
a SoundCloud user as showing their valuing of another user,
in a publicly accessible manner. Qualitatively, we found
that there was value attached to having large numbers of
followers, though the participants disagreed as to how important
this was to them personally. Quantitative analysis
revealed, however, that SoundCloud is not a media which
compares to YouTube or Twitter in terms of magnitude of
followers. In our 150000 user sample, only three accounts
had over 100000 followers and all of these accounts represented
agents involved in music that had enjoyed signifi-
cant commercial/popular recognition, above the subcultural
recognition that is more common in electronic music scenes.
Interim results and redirections in our quantitative research
Following Allington (under review), initial quantitative
research (Jordanous, Allington, and Dueck 2014) saw
us seek the top-ranked users according to centrality measures.
(Centrality measures highlight the most influential
nodes in a network.) We also attempted to visualise the networks
but found that graphs for samples greater than 500
users would be unreadable. We measured recommendation
and influence through indegree rankings (a measure based
around how many users follow another user). In an initial
test sample of 1500 users, we identified key users. This ranking
did find some key players in electronic music whose data
had been captured in our sample, such as Tiesto. Our results, ´
however, did not help us understand the network at a deeper
level, particularly regarding our search for cultural value
through peer esteem. While indegree is more sophisticated
than merely measuring the number of followers per account,
there was some similarity between these two rankings. A
‘Justin Timberlake’ account, for example, comes in at position
20, despite having no interactive activity on SoundCloud
and therefore no identifiable contribution to cultural
value through SoundCloud interactions.
We started to explore more sophisticated methods such
as PageRank and eigenvector rankings to help identify key
players in SoundCloud’s networks. However we started
to notice a mismatch between qualitative findings and the
shape of our quantitative data, stemming from earlier observations
about the nature of sub communities within electronic
music. In interviews, when we asked questions about
valuing and appreciation, participants often replied in terms
text pointing to the other collaborator(s), or via the creation of a
new SoundCloud account representing all the collaborators, which
is distinct from the collaborators’ personal accounts.
Proceedings of the Sixth International Conference on Computational Creativity June 2015 113
of relationships and interaction. When we probed further,
the participants tended to answer in terms of the genre(s)
they produced music in, reframing the question to focus on
that sub-community they were part of.
Understanding that we should look for subnetworks and
cliques within our data, we investigated on what grounds we
should cluster our data, through interviews and through inspection
of our data for commonly occurring links. Genre
was one important clustering factor suggested in the interviews.
Somewhat surprisingly for an online network, geographical
location was another factor we were guided to
investigate. Participants reported how offline interactions at
particular places fed back into the social network interactions.
The importance of offline contacts could not be ignored,
especially given the social network ‘fatigue’ reported
by some participants in building their profiles. In terms
of location having an influence on a musician’s perceived
value, our interviewees talked about the importance of their
location for raising their profile and credibility. Though
some had experience of being based elsewhere, many of our
interviewees were based in London, which - as we find below
- is an important centre for electronic music. One participant
in particular reported a conscious decision to base
themselves in London for profile-raising reasons.
Analysis of clusters of users and sub-networks Learning
from experience, our quantitative research focused on what
sub-communities and clusters existed in our data. We took
two directions: 1. constructing and studying multiple networks
of electronic music producers and their connections,
and 2. using the comments-based data to identify the language
used between peers to express value.
We built networks of accounts and tracks, based on ‘follow’
relationships, which we could re-apply centrality measures
to. Clusters and cliques in these networks were also
identified where possible, based on available metadata about
users and tracks such as genre. We should note here that
many users do not provide location information, particularly
if not active users (though we focus on those users who actively
engage with other users on SoundCloud).
Inspecting the data on comments about tracks, we noted
that the overwhelming majority of comments tended to be
positive, unlike commenting activity typically observed on
sites such as YouTube (Pihlaja 2012). In our analysis of the
comments data (filtered from spam where possible) we used
the Open Office dictionaries for English, French, Spanish
and Italian to identify and extract English language comments
to reasonable accuracy.8 We treated the Englishlanguage
comments on tracks as corpora based on track genres.
Corpus analysis allowed us to identify evaluative vocabularies
associated with particular genres, groups, and locations,
by comparing these subcorpora on the lexical level.
Given that SoundCloud comments were typically positive
(or spam), we posit these vocabularies as genre-specific indicators
of value as expressed in that genre.
8
Our approach did not pick up comments such as ‘wooooot!!!’
or ‘loveeeeeeeeee’, the type of which occur frequently in our data.
Table 1: Follow relationships by frequency of locations
Location of followed Location of follower n
1 London London 3799
2 Melbourne Melbourne 2274
3 Berlin Berlin 1375
4 Paris Paris 1253
5 New York New York 1190
Computational analysis: Results and discussion9
Geography Analysis of locations in our random sample
revealed London as the most common city location for music
makers (users who had uploaded tracks to SoundCloud);
200 accounts out of the 17357 eligible accounts were attached
to users based in London. London music-makers
had the highest mean number of followers, though a disproportionately
high standard deviation reveals results were
skewed by a small number of very highly followed accounts.
On analysing individual ego-networks of our participants,
we could identify clear clusters within the ego-network
based on location of the users, indicating a preference for
users to follow other users in the same geographical area
as them. This hypothesis was supported by evidence in the
larger random sample (see Table 1).
Other key cities identified through our random sample behind
London were New York (171 accounts belonging to
music-makers), Los Angeles (93), Chicago and Paris (both
81). In terms of followers, strong bidirectional links were
identified between London, New York and Los Angeles
(UK/US), and then between London, Berlin and Paris (major
European capitals). Given that this part of our analysis
was genre-agnostic, it was surprising to see cities such
as Nashville and Mumbai, with strong musical connections
to country music and Bollywood music respectively, featuring
little in the interconnected data. Perhaps this is because
these types of music do not enjoy the same associations
with online/digital technologies and, more specifically,
with SoundCloud (emphasising the need to ensure that the
social interactions you are analysing are relevant to the creative
communities you study).
Using eigenvector centrality based on a graph connected
by follow relationships, we identified similar rankings; the
central node in this graph was London (0.90093 centrality),
followed by New York (0.24838), Berlin (0.20645), Los Angeles
(0.20121) and Paris (0.10437). By country, the United
States was top by some degree (0.96823 centrality, with the
second highest centrality at 0.21216 for the UK). Germany,
Canada and France were next in influence, with centrality of
0.07380, 0.05749 and 0.05193 respectively.
Genre In raw frequencies, hiphop producers were most
prevalent in our sample, with 155 users uploading tracks
9
The following is a synopsis of findings that are relevant to developing
computational analysis of cultural value. Fuller reports
of our findings are described in (Allington, Dueck, and Jordanous,
submitted) and (Allington, Jordanous, and Dueck 2014).
Proceedings of the Sixth International Conference on Computational Creativity June 2015 114
Table 2: Follow relationships by genre
Follower Following n
1 hiphop hiphop 2443
2 house house 2276
3 techno techno 1415
4 progressive house house 800
5 dubstep dubstep 679
tagged as ‘hiphop’. House music was second (90 users),
followed by rock (61), rap (59) and pop (49). However once
we start to study the inherent cultural value through interactions
between producers, we see different results as to the
influence of different genres. We used eigenvector centrality
based on follow relationships to study how producers of music
within one genre interacted with music-makers in other
genres. In our sample, house music producers were most in-
fluential, followed by hiphop, techno, and deephouse. Music
tagged as ‘electronic’ is still prevalent, though its subcategories
are widely used as tags instead. These fuller results
(Allington, Jordanous, and Dueck 2014, Table 27) also evidenced
the influence of electronic musicians (as opposed to
musicians of other genres) on SoundCloud.
Many tracks were tagged with more than one genre term,
and Figure 1 reveals patterns within genre tagging that empirically
support existing genre classifications. Clustering
together tags that frequently occurred together on tracks,
we identified three macro-genres that could be categorised
as ‘EDM’ (Electronic Dance Music), ‘urban’, and a miscellaneous
‘other’ category. The two named macro-genres
‘EDM’ and ‘urban’ align with an analysis of data from 2007
on all musical genres on MySpace by Lee & Silver (2014),
identifiably corresponding to two clusters that they tagged
as ‘Electro/Dance’ and ‘Black & Brown’ respectively.
Focusing on activity in the EDM and urban clusters (as the
‘Other’ cluster contains negligible activity) typically EDM
producers follow other EDM producers, and similarly Urban
producers follow other Urban producers. Looking at the
genre level, a similar pattern of following producers within
the same genre is noted (see Table 2).
Follower activity A common-sense hypothesis was supported
by results: users who uploaded tracks to SoundCloud
typically had more followers than those who did not (a mean
of 127 followers per account for those who uploaded public
tracks, compared to a mean of 19 per all types of users in our
150000-users sample). If we take our qualitative findings
that number of followers is generally positively associated
with value recognition, then we can underline that musicmakers
are valued in the SoundCloud community.
Commenting activity Taking the three macro-genres we
identified, EDM producers were the most prolific commenters
with 11711 comments, compared to 3673 comments
by urban producers, and 2982 comments by producers
of the ‘other’ genres. By genre, dubstep producers engaged
in commenting behaviour the most (2569 comments), then
techno (2254), hiphop (2081) and house (1725).
Figure 1: Co-occurrence of genres in track tags
From the comments we have (as described above) identi-
fied genre-specific English-language vocabularies indicating
value expressions. Keywords are presented for the top genres
in Table 3, in order of ‘keyless’ (decreasing proportional
frequency). This table shows the different types of vocabulary
prevalent per genre, for example keywords in comments
on techno tracks appear more polite than on hiphop tracks.
Evaluation of our approach
When is social network analysis appropriate as a proxy
for measuring cultural value? As shown by the lack of
useful results of SoundCloud users in cities like Nashville
and Mumbai, one needs to ensure they are analysing appropriate
social networks for their specific creative domain.
There may not be a relevant social network directly for these
acoustic-music-based communities, but general social networks
such as Twitter may prove useful.
How could the VEM findings be useful to computational
creativity researchers? Cynically, perhaps, we could set
up a London-based SoundCloud account for a hypothetical
electronic music computational creativity system we want to
promote the work of, ensuring we (or the system) upload(s)
tracks produced by our system. We could concentrate efforts
on developing our hypothetical system’s ability to interact
with other music-makers’ tracks who work in similar
Proceedings of the Sixth International Conference on Computational Creativity June 2015 115
Table 3: Genre-specific keywords for expressing value
Dubstep Techno Hiphop House
1 sick set dope nice
2 tune great shit house
3 nice tracks beat super
4 big loved leave production
5 mix fantastic song support
genres, commenting on such tracks and responding to comments
on its own tracks using keywords which have been
identified as commonly used in the genre we are working in.
We could develop our system to follow other music-makers
based in strategically important cities such as London, New
York, Los Angeles, Paris or Berlin, or who upload music
of similar genres. While this would not necessarily develop
the musicality of our hypothetical artificial electronic musician,
we argue such moves (if executed plausibly) would
help increase the cultural value attributed to our musician
(notwithstanding the debate about the effects of identifying
the account - or not - as that of an artificial musician (Moffat
and Kelly 2006; Cook and Colton 2014)).
How could social network analysis be used more broadly
within computational creativity? For this work to be
most useful to computational creativity researchers, it could
a. show how cultural value can be identified and gauged
through research and/or b. offer a way of autonomously
making value judgements about computational creativity
systems. We believe that our work above demonstrates point
a., how to tangibly identify markers that indicate cultural
value. Allington (under review) has previously used similar
network analysis to study Interactive Fiction.
What we pursue now is the afore-mentioned point b., a
methodology for using computational network analysis to
gauge the cultural value associated with a creative entity
such as a computational creativity system. For such an approach
to work, we need the system to be capable of interacting
with relevant online communities in a plausible manner,
as suggested above for our hypothetical electronic computer
musician. We also need there to exist an appropriate social
network for such interactions to take place in, or as a fascinating
alternative, a multi-agent system or similar digital
environment containing several interacting agents. For the
actual analysis, we advocate using a combination of initial
quantitative data analysis and qualitative research to identify
key indicators of cultural value that can be traced in the social
network interactions. With these conditions in place, we
can analyse interactions in the network and compare our creative
system or agent to others within the network to gauge
the value inherent in its interactive social behaviour.
Future work Our results show that electronic music subcultures
are geographically influenced and, within the UK,
heavily London-centric. Our quantitative methodology
could reveal important scenes associated with other cities,
and whether we could identify musicians that are considered
heavily influential and ‘valuable’ to the local scene(s).
Somewhat inspired by the 2014 Scotland independence referendum,
we plan to examine electronic music scenes within
Scotland to test our methodology. Our next step will be to
apply the same approach to other creative domains to see
if social network analysis can be applied more broadly for
computationally analysing cultural value. We would welcome
collaborations.
Further useful information may be gained from quantitative
analysis of comments made by users on each other’s
tracks, though this was not so straightforward to analyse during
the project’s funded time. In this work we would have
liked to explore and build networks of users based around
‘comment’ relationships between users. Such work will require
considerably more intricate and varied analysis to filter
links based around genuine comments. Ongoing work is
currently examining the links between users based on commenting
behaviours. We would also like to examine conversations;
repeated comments or comments on multiple tracks
from a user should indicate greater peer engagement. Conversations
proved rather difficult to detect quantitatively due
to the lack of a standard way to indicate who your comments
are directed towards, but their analysis would be useful.
Conclusions
Value is recognised as a key aspect of creativity. In evaluating
computational creativity, one large problem we face
is in gauging the value of the work generated by our systems.
Such evaluation is particularly problematic when we
consider that value is often a cultural and intangible resource
apportioned subjectively through the actions of peers.
To what extent can we use social network activity to identify
the cultural value of creative entities? Here we addressed
this question through a case study investigating how
electronic musicians place value in each other’s work. The
Valuing Electronic Music (VEM) project combined ethnographic
observation/interviewing with automated collection
of quantitative data from the SoundCloud music networking
site. Our approach has implications for how we could measure
cultural value in other domains, as well as contributing
to our understanding of cultural value in electronic music.
Challenges and rewards alike come from combining situated
qualitative research with quantitative analysis of large
datasets gathered online. Learning from (and feeding back
into) the findings from interviews with electronic musicians,
we used computational analysis to study interactions in social
networks.10 Through such analysis we extrapolated information
about how musicians interact with each other on
SoundCloud, and how they express appreciation of each
other’s work. Typically, it was more productive to study
clusters of strongly connected cliques within the SoundCloud
network, rather than a sample of the entire network.
The SoundCloud user community tends to cluster according
to several factors. We found empirical evidence of clusters
forming around common musical genres, and also of clusters
around certain privileged geographical locations such
10Our approach echoes (Jordanous 2012b): to better represent
creative activities using quantitative models, we need good understanding
of the creative domain as well as the models themselves.
Proceedings of the Sixth International Conference on Computational Creativity June 2015 116
as London. One key ‘take-home’ finding from this work is
that one can study cultural value computationally by studying
social activity, but often it is most useful to study interaction
between smaller sub-groups of a network, rather
than taking an overall view of the entire network as a whole.
In other words, to understand how people express value for
each other’s work, we should look for social interactions and
the building of relationships within a community.
We found that while certain kinds of activity on SoundCloud
have little apparent economic value (e.g. commenting
on each others’ tracks, publishing free downloads) these activities
seem to generate cultural value that facilitates more
economically valuable work. For the most part, musicmakers
assert their concern for all listeners, but close attention
to their activity (and how they describe it) suggests that
interactions with peers (i.e. fellow music makers, preferably
within similar genres, areas or with other links) are especially
important for the production of value for their work.
Our computational analysis of SoundCloud data allowed
us to approximate the value placed in electronic musicians’
work, showing that we can use social network analysis as
a proxy for measuring certain types of musical and cultural
value in a creative domain. We hypothesise that our methodology
can be extended to analyse quantitatively the value
inherent in other social networks centred around creative activity.
We believe that this work contributes towards a significant
type of tool in our ‘computational creativity toolkit’:
an automatable method for evaluating social/cultural value.
Acknowledgments This work was funded by an AHRC Cultural
Value grant “Online networks and the production of value in
electronic music”. We thank our interviewees, workshop and public
engagement event participants and advisory board members.
<references_biblio/>
References
Allington, D. Producing the Cultural Value of Interactive Fiction:
Field Theory, Social Network Analysis, and the Text Adventure
Game.
Allington, D.; Dueck, B.; and Jordanous, A. 2015. Networks of
Value in Electronic Dance Music: SoundCloud, London, and the
Importance of Place.
Allington, D.; Jordanous, A.; and Dueck, B. 2014. Online Networks
and the Production of Value in Electronic Music.
Blacking, J. 1973. How musical is man? Univ. Washington Press.
Boden, M. A. 2004. The creative mind: Myths and mechanisms.
London, UK: Routledge, 2nd edition.
Bourdieu, P. 1993. The Field of Cultural Production: Essays on
Art and Literature. Cambridge, UK: Polity Press.
Colton, S. 2008. Creativity versus the Perception of Creativity in
Computational Systems. In Proceedings of AAAI Symposium on
Creative Systems, 14–20.
Cook, M., and Colton, S. 2014. Ludus Ex Machina: Building A 3D
Game Designer That Competes Alongside Humans. In Proceedings
of 5th International Conference on Computational Creativity.
Csikszentmihalyi, M. 1988. Society, culture, and person: a systems
view of creativity. In Sternberg, R. J., ed., The Nature of Creativity.
Cambridge, UK: Cambridge University Press. Ch. 13, 325–339.
De Nooy, W. 2003. Fields and Networks: Correspondence Analysis
and Social Network Analysis in the Framework of Field Theory.
Poetics 31(5-6):305–327.
DHA Communications. 2012. The Working Musician. Report for
the Musicians’ Union.
Dueck, B. 2013. Jazz Endings, Aesthetic Discourse, and Musical
Publics. Black Music Research Journal 33(1):91–115.
Elafros, A. 2013. Greek Hip Hop: Local and Translocal Authentication
in the Restricted Field of Production. Poetics 41(1):75–95.
Finnegan, R. 2007. The Hidden Musicians: Music-Making in an
English Town. Middletown, CT: Wesleyan Univ. Press, 2nd ed.
Jordanous, A.; Allington, D.; and Dueck, B. 2014. Using online
networks to analyse the value of electronic music. In Proceedings
of 5th International Conference on Computational Creativity.
Jordanous, A., and Keller, B. 2012. What makes musical improvisation
creative? Journal of Interdisciplinary Music Studies
6(2):151–175.
Jordanous, A. 2011. Evaluating Evaluation: Assessing Progress in
Computational Creativity Research. In Proceedings of the Second
International Conference on Computational Creativity (ICCC-11).
Jordanous, A. 2012a. A Standardised Procedure for Evaluating
Creative Systems: Computational Creativity Evaluation Based on
What it is to be Creative. Cognitive Computation 4(3):246–279.
Jordanous, A. 2012b. Evaluating Computational Creativity: A
Standardised Procedure for Evaluating Creative Systems and its
Application. Ph.D. Dissertation, University of Sussex.
Lee, M., and Silver, D. 2014. Testing the Springsteen Conjecture:
“Exploring the Post-Authentic Musical World” with
big, messy internet data. http://badhessian.org/2014/07/testingthe-springsteen-conjecture-exploring-the-post-authentic-musicalworld-with-big-messy-internet-data/
last accessed Feb 2015.
Lena, J. C., and Pachucki, M. C. 2013. The Sincerest Form of
Flattery: Innovation, Repetition, and Status in an Art Movement.
Poetics 41(3):236–264.
Lopes, P. 2000. Pierre Bourdieu’s Fields of Cultural Production:
a Case Study of Modern Jazz. In Brown, N., and Szeman, I., eds.,
Pierre Bourdieu: Fieldwork in Culture. New York: Rowman and
Littlefield. 165–185.
McCormack, J. 2007. Creative Ecosystems. In Proceedings of
the 4th International Joint Workshop on Computational Creativity,
129–136.
Moffat, D. C., and Kelly, M. 2006. An investigation into people’s
bias against computational creativity in music composition. In Proceedings
of the 3rd International Joint Workshop on Computational
Creativity ECAI06 Workshop.
Pease, A.; Guhe, M.; and Smaill, A. 2010. Some Aspects of Analogical
Reasoning in Mathematical Creativity. In Proceedings of
the International Conference on Computational Creativity, 60–64.
Perez y P ´ erez, R.; Aguilar, A.; and Negrete, S. 2010. The ERI- ´
Designer: A Computer Model for the Arrangement of Furniture.
Minds and Machines 20(4):533–564.
Pihlaja, S. 2012. The development of ‘drama’ in YouTube discource.
Ph.D. Dissertation, Open University.
Ritchie, G. 2007. Some Empirical Criteria for Attributing Creativity
to a Computer Program. Minds and Machines 17:67–99.
Turino, T. 2008. Music as Social Life: The Politics of Participation.
Chicago: University of Chicago Press.
Proceedings of the Sixth International Conference on Computational Creativity June 2015 117