Big Data Analysis of Music Influencing Factors Based on Complex

Network

Qirui Yu

, Wenshuo Zhai

, Rui Zhu

, Yuanyuan Jiao

, Liang Bai

and Nanjun Li

College of Systems Engineering, National University of Defense Technology, Changsha, China

College of Electronic Science and Technology, National University of Defense Technology, Changsha, China

These authors contributed equally to this work *Corresponding author

Keywords: Complex Network, Node Aggregation, Cosine Distance.

Abstract: Music plays an important role in human society. This paper attempts to explore the law of music development

through the music style of artists over the years, the works created and the mutual influence between artists.

Firstly, this paper adopts node aggregation algorithm to simplify the complex network of artists of different

ages and genres into a simple sub-network. Then, based on the cosine distance, this paper analyzed musical

characteristics of artists so as to work out the similarity degree between them. Finally, this paper combines

the development of society and technology to comprehensively analyze the law of the development of music,

and draws the conclusion that different music genres have different influences in different ages. Major social

changes and major scientific and technological development related to music will also have a great impact on

the development of music.

1 INTRODUCTION

As an important part of art, music plays an irreplace-

able role in the development of human society. In or-

der to better explore the development of music, we

wanted to develop a method to quantify the develop-

ment of music. There are many factors that influence

a artist's work, such as personal experiences, the in-

fluence of other artists, social events, technological

developments, and so on. And the song has rhythm,

loudness, dance and other characteristics. In order to

discover the law of the development of music, we

should pay attention to the emergence of music gen-

res and the development of different music genres in

different periods from the perspective of the whole.

The Integrative Collective Music (ICM) Society

aim to define a model for measuring musical influ-

ence. They launched a global challenge in 2021, hop-

ing that teams would come up with a better solution

to the problem. These data were scraped from AllMu-

sic.com. This paper analyzed the music-related data

given in the competition and the data collected from

the Internet. In order to better explore the process of

music evolution, looking for the law of music schools

and branches, as well as the interaction between art-

ists, we use a variety of methods to analyze the simi-

larity, characteristics, influence, genre evolution and

other relevant aspects of music.

We built up the model of Node aggregation, fus-

ing all artists of the same genre from the same era. We

will get a new compressed network weighted by the

sum of the out-degrees of the same kind of nodes in

the original network, with the value of weight reflect-

ing the influence of a particular genre in a certain

year. We showed the measurement of the interaction

between genres in some eras in the main body of our

paper. And the influence of each artist is reflected by

his genre and era. We defined seven intervals as the

characteristic value of each genre according to the

principle of Box Plot. The similarity between genres

is determined by the coincidence degree of the inter-

vals by contrast while the similarity within genres is

ascertained by the variance degree of the intervals.

We analyzed the degree of similarity between influ-

encer and follower by straightly dividing the seven

characteristic values of a follower by those of his in-

fluencers correspondingly and defining the similarity

by the deviation degree between the ratio and 1.

Through the fluctuation of each characteristic with

time, we found that valence and energy functioned as

the indicators that reveal the dynamic influencers.

Combined with the figure of the amount of works

290

Yu, Q., Zhai, W., Zhu, R., Jiao, Y., Bai, L. and Li, N.

Big Data Analysis of Music Inﬂuencing Factors Based on Complex Network.

DOI: 10.5220/0011910700003613

In Proceedings of the 2nd International Conference on New Media Development and Modernized Education (NMDME 2022), pages 290-294

ISBN: 978-989-758-630-9

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

changing with time, we can explain the regulation of

genres and artists changing with time.

In the end, the paper mentioned the method of

combining musical revolutions with historical and

cultural events, thus applying the idea of deep learn-

ing in order to dig out the effect laws of social, polit-

ical or technological changes on music.

2 CONSTRUCTION AND

OPTIMIZATION OF COMPLEX

NETWORKS AMONG ARTISTS

2.1 The Construction of Complex

Networks among Artists

We regard that only the amount of a specific artist’s

immediate followers counts when measuring his per-

sonal musical influence since if the followers of his

followers were influenced by him, they must also be

his followers according to the given information. To

quantify the musical impact of every single genre, we

consider conducting dynamic fusion of related nodes

and measuring the influence between every two art-

ists by the interacts of the genres they belong to.

Eventually, we are going to extract a sub-network

from the original directed network, thus conducting

micro analysis on three randomly-chosen nodes, mak-

ing a comparison to find out whether the genre of a

certain artist’s main influencers is concurrently the

main influential genre of the genre he belongs to. If

so, we can conclude that our ‘music influence’

measures are revealed in this sub-network.

Figure 1. The directed network of musical influence drawn

in python.

2.2 Simplify Complex Networks Based

on Node Aggregation

Under the obtained network, we discovered that the

in-degree of a certain node is exactly the number of

its corresponding artist's influencers while its out-de-

gree equivalent to the amount of his followers. Then,

we fused the same type of music based on the year,

that's, fusing the nodes according to both influencers'

active starts and their main genres, at the same time

accumulating both in-degree and out-degree of every

single fused node in order to form the new in-degrees

and out-degrees of the points after aggregation.

Eventually, extracted a sub-network from the

original directed network, making a comparison to

find out whether the genre of a certain artist's main

influencers is concurrently the main influential genre

of the genre he belongs to. If so, we can conclude that

our 'music influence' measures are revealed in this

sub-network.

2.3 Data Mining on the Network

Randomly choosing three nodes to test the accuracy

of our results, corresponding to artists belonging to

respectively 1980 Pop/Rock, 2000 Pop/Rock and

1960 Pop/Rock, we found that the largest proportion

of their influencers are respectively 1960 Pop/Rock

with the proportion 0.488372, 1960 Pop/Rock,1980

Pop/Rock, 1950 R&B and 1940 Vocal sharing the

same proportion and 1970 Pop/Rock with the propor-

tion 0.34.Comparing the results with the aggregated

directed network, we found that the results are highly

consistent, indicating that measuring the mutual in-

fluence of artists in different genres by the cross-in-

fluence of the genres they belong to embodies the idea

does make sense.

Figure 2. The directed network of musical influence.

Big Data Analysis of Music Inﬂuencing Factors Based on Complex Network

291

Our model is based on the aggregation of net-

works. The fusion of genres of the same type by time

scale reflects the influence of diverse genres in differ-

ent periods. The subtlety of this model is that it pos-

sesses enough robustness to reflect the influence of a

certain musical genre in each period reasonably and

ingeniously since the musical influence of diverse

genres differs greatly in distinct decades while a spe-

cific genre's influence on music of different ages in a

particular year is also different. Our model realized

the goal of conducting concrete analysis to specific

problems. Therefore, it can be regarded more targeted

in time, clearer and more quantifiable. Compared

with the establishment of fuzzy comprehensive eval-

uation system, this model evaded the unnecessary

trouble when conducting empowerment due to the ab-

stractness of the research object and the lack of rele-

vant data. However, we should admit that our model

is limited in some respects. For example, the accuracy

of the results largely depends on the soundness of the

samples. If the number of chosen samples of artists in

a certain genre or a certain era is too small, then the

result of our solution wouldn't be representative

enough, far from the actual situation.

3 ANALYSIS OF MUSICAL

SIMILARITY

3.1 Similarity Analysis Based on

Cosine Distance

We selected the corresponding data of seven music

characteristics as valid data, thus generating a seven-

dimensional characteristic vector for each song, each

dimension corresponding to a characteristic index.

Next, we identified the correspondence between art-

ists and genres and classified the artists by the genre,

that is, putting artists belonging to the same genre into

a collection, while scouring off unclassified artist

data. For artists of the same genre, we calculated the

Cosine Distance between every two seven-dimen-

sional characteristic vectors in the set. In conclusion,

the similarity of a particular genre is the mean value

of all these Cosine Distances.

cos





(1)

And the Cosine Distance is defined as ω1

cos θ.

Set the Cosine Distance between the 𝑖



artist and

the 𝑗



artist of the same genre as 𝑎



, and the sample

size of the genre as n, thus the similarity of a certain

genre can be expressed as the equation below:

ijn

≤< ≤



(2)

For artists belonging to different genres, we thus

calculated the mean value of cosine distance from all

the seven-dimensional vectors in one genre to those in

the other genre to work out the similarity between two

genres. Set the Cosine Distance between the 𝑖



artist

in one genre and the 𝑗



artist in the other genre as

𝑏



,and the sample size of the two genres as m and n,

thus the similarity of two genres can be expressed as

the equation below:



(3)

The former figure is the original similarity data of

20 genres and the latter one is a more intuitive thermal

graph presentation after grading and quantifying the

similarity data.

Figure 3. Thermal graph based on similarity

3.2 Comparing Similarities and

Influences between Different

Genres

For the sake of distinguishing a genre, we firstly put

our hand to measuring the similarity between differ-

ent genres. Due to the consideration of determining a

parameter to characterize each characteristic, we de-

fined the interval from one fourth to three fourth of

all the songs' characteristic values in a certain genre

(the characteristic values had been strictly sorted) as

the parameter according to the data analysis method

based on Box Plot.

NMDME 2022 - The International Conference on New Media Development and Modernized Education

292

As for comparing the similarity among different

genres about a specific characteristic, we took turns

contrasting the same music characteristic intervals of

every two genres. And in order to unify the standard

of measurement, we further defined the degree of

similarity as the ratio of the obtained span and the

span of the smaller parent interval.

Figure 4. Line Chart of 4 Characters Changing with Time

(data from http://AllMusic.com)

The area of each matching result was the total

similarity between every single song and a particular

genre. Thus, we defined the largest area of the match-

ing result as the genre this song belongs to. In order

to seek out the law of the change of genres over time,

we started from both horizontal and vertical perspec-

tives. From the angle of breadth, we calculated the

variance of all elements in the song set corresponding

to each characteristic for all the 20 genres. We found

that most characteristics of the songs in each genre

are not that similar, with only several of the seven

ones having small variances. From the portrait angle,

we classified and aggregated songs on a time scale,

thus analyzing the changing trend of each character-

istic with time.

As for judging whether some genres are related to

others, we took both music influence and the similar-

ity into consideration. If the similarity scaling be-

tween two genres belongs to close resemblance and

their influencing scaling are over great impact, we

then regard that these two genres are related to each

other. For the sake of obtaining the result of how gen-

res change over time, we took the genre Pop/Rock as

an example.

4 ANALYSIS AND

EXPLORATION OF MUSIC

INFLUENCE

4.1 Analyzing the Influence Process of

Musical Evolution

To determine whether the influencers actually affect

the music created by the followers, we still started

from analyzing the similarity of seven characteristic

values. For each follower, we divided his values by

those of the influencers correspondingly. Observing

the data of proportion, we can conclude that the pro-

portion of songs with extremely similar characteris-

tics is not very large, indicating that the music styles

of followers are not so affected by their influencers.

Next, we considered from the following two per-

spectives. On the one hand, we explored the influence

of the same influencer on various followers. On the

other hand, we took the distinct effects from various

influencers on a specific follower into consideration

as well. According to the similarity division basis

above, we first calculated the average proportion of

followers sharing high similarity for a particular char-

acteristic with every single influencer. Subsequently,

we worked out the average proportion of influencers

sharing high similarity for a particular characteristic

with every single follower utilizing the same method.

We concluded that the influencers actually affect

the music created by the followers and there do exist

some music characteristics more ‘contagious’ than

others, they are Danceability and Tempo. In order to

measure musical evolution of a genre over time, we

transformed our focus on the works belonging to that

genre. We took the average value of each characteris-

tic of R&B songs in each year to reflect the overall

situation of a specific characteristic of the genre in

that year.

4.2 Discussing the Role of Cultural

Factors Acting on the Field of

Music

Our work expressed information about cultural influ-

ence of music in time or circumstances by building up

the bridge between the alteration of characteristics

and the time corresponding to its occurrence.

As for identifying the effects of social, political or

technological changes on the music evolution, we

come up with the idea of deep learning. After the pre-

vious data processing and analysis, we've already

been equipped with the ability of quantifying both

Big Data Analysis of Music Inﬂuencing Factors Based on Complex Network

293

music styles and genres by conducting data analysis

of all the characteristic values.

In order to microcosmically measure a musical

revolution during a historical period, we considered

the musical characteristic values before the revolution

as the input and those after the revolution as output.

Through literature search, we were going to find out

all the social, political and technological events dur-

ing the period. Classifying these events strictly ac-

cording to their categories, we would get three event

sets covering Social Category, Political Category and

Technological Category. Defining the law of action

of the above three kinds of events on music respec-

tively as equation f(x), g(x) and h(x), we were enabled

to build up the model of input, output and effect about

deep learning. Through a large number of data train-

ing with events of music revolutions scattered in

every year and every month, the effect laws of f(x),

g(x) and h(x) would be eventually dug out.

5 CONCLUSION AND

DISCUSSION

Music plays a vital role in expressing people's

thoughts and reflecting social life. In order to explore

the law of music development and uncover the inter-

nal logic of music development, this paper constructs

a complex network among artists, which includes the

characteristics of artists, the characteristics of artists'

works, social factors and technological factors, re-

flecting not only the cross-influence between differ-

ent genres of music in different ages, but also the ef-

fect of a genre on itself during various periods. Then,

we simplify the complex network based on node ag-

gregation through taking genre of a specific era as the

unit. Meanwhile, we macroscopically measure the

mutual influence of artists of different genres by the

mutual influence of their respective genres. In the

analysis of the influence degree, this paper bases on

the cosine distance for the correlation analysis.

Through our research, we find that the influence

among artists, the major social events related to music

creation and the development of technology all have

a huge impact but with various degree of influence on

the development of music. We have only made a pre-

liminary study of the factors that influence the devel-

opment of music. In order to better reveal the law of

the development of music and make contributions to

human art, let us all work together.

REFERENCES

Anna J. Reisenweaver. Guido of Arezzo and His Influence

on Music Learning [J]. Musical Offerings, 2012,3(1).

Alberto Arroyo. The meta-instrument as compositional

technique: Strategies of the current musical creation[J].

Emille: The Journal of the Korean Electro-Acoustic

Music Society,2018,16.

James Leonard, Jérôme Villeneuve, Alexandros Kontoge-

orgakopoulos. Multisensory instrumental dynamics as

an emergent paradigm for digital musical creation A

retrospective and prospective of haptic-audio creation

with physical models [J]. Journal on Multimodal User

Interfaces, 2020,14(3).

Jon M Sweeney. Review: Richard Wagner's immense influ-

ence on music (and history) [J]. America,2020,223(5).

Kaskon W. Mindoti, Hellen Agak. Political Influence on

Music Performance in Kenya between 1963-2002 [J].

Bulletin of the Council for Research in Music Educa-

tion, 2004(161/162).

Kong Wenting. Analysis of the Influence of Digital Media

Technology on Music Communication. [J]. Journal of

Physics: Conference Series, Volume 1453, 2019 2nd

International Conference on Computer Information Sci-

ence and Artificial Intelligence (CISAI 2019).

Sunil Aryal, Kai Ming Ting, Takashi Washio, Gholamreza

Haffari. Data-dependent dissimilarity measure: an ef-

fective alternative to geometric distance measures [J].

Knowledge and Information Systems,2017,53(2).

Tao Wang, Hongjue Wang, Xiaoxia Wang. A novel cosine

distance for detecting communities in complex net-

works [J]. Physica A: Statistical Mechanics and its Ap-

plications,2015,437.

Trainor Laurel J. The origins of music in auditory scene

analysis and the roles of evolution and culture in musi-

cal creation. [J]. Philosophical transactions of the Royal

Society of London. Series B, Biological sciences,

2015,370(1664).

Trainor Laurel J. The origins of music in auditory scene

analysis and the roles of evolution and culture in musi-

cal creation. [J]. Philosophical transactions of the Royal

Society of London. Series B, Biological sci-

ences,2015,370(1664).

W. Xiong, X. Yu, C. Liu, G. Wen and S. Wen, "Simplifying

Complex Network Stability Analysis via Hierarchical

Node Aggregation and Optimal Periodic Control," in

IEEE Transactions on Neural Networks and Learning

Systems.

Xiaoyan Gao. P2P Network Node Aggregation Algorithm

Research Based on Community [J]. International Jour-

nal of Engineering Practical Research, 2015,4(1).

Yucheng Lin, Zhigang Chen, Jia Wu, Leilei Wang. An Op-

portunistic Network Routing Algorithm Based on Co-

sine Similarity of Data Packets between Nodes [J]. Al-

gorithms, 2018,11(8).

NMDME 2022 - The International Conference on New Media Development and Modernized Education

294