没有合适的资源?快使用搜索试试~ 我知道了~
MCM美国大学生数学建模竞赛题目与答案:运动与措施,音乐的发展和音乐影响.pdf
0 下载量 50 浏览量
2023-08-21
10:45:25
上传
评论
收藏 2.09MB PDF 举报
温馨提示
试读
22页
在探讨音乐如何通过影响演变这一广泛话题时,我们的报告涵盖了通过使用多种数据驱动方法解决的大量主题。这些技术包括:开发和测试相似度指标,通过分析和使用神经网络来比较影响者和追随者;利用统计图表,如折线图和柱状图来表达跨时间和流派的总体趋势;开发和测试神经网络,以确定艺术家的音乐是否属于特定流派。我们可以确定音乐特征指标的全球趋势,识别音乐流派的特征,并深入了解音乐世界各领域的同质性。然而,我们的程序在很大程度上依赖于所提供数据集的质量以及对数据的正确处理和解释。
资源推荐
资源详情
资源评论
Problem Chosen
D
2021
MCM/ICM
Summary Sheet
Team Control Number
2125654
The music industry is an extraordinarily complex system of influence between songs, artists, and
genres that grows with each passing year. We examine this system and its components in an effort to
understand how influence shapes the landscape of the music world. Our procedures utilize data
provided by Spotify’s API, containing information about influencers and followers and various
metrics describing the musical nature of artists and their songs.
We construct a directed network based on artist influence data and use it to examine the structure and
meaning of influence within a genre. We also employ neural networks to predict artists’ genres and
influencers. Additionally, we apply descriptive procedures to various data on musical metrics for
individual songs and artists, allowing us to study inter- and intra-genre trends over time and the
defining characteristics of genres. We also introduce our own measure of similarity based on the
Euclidean distance between a subset of these metrics, useful in assessing similarity between songs,
artists and genres.
We first observe some key global trends, including a rise in the prominence of high-energy genres
over time, and a decline in global acousticness consistent with the development timeline of relevant
technology. We also examine trends in individual metrics within individual genres, which provide
insight into the global trends.
In order to investigate the nature of genres, we create a neural network that uses the provided metrics
to predict the genre of an artist given their time period. Our technique yielded an approximately 80
percent success rate in predicting whether a song is of the pop/rock genre, which may suggest that the
data provided on artists and songs may be good indicators of genre. When running our neural network
on R&B versus country songs, we found that the neural network more accurately classified the R&B
songs as non-pop/rock than country, suggesting similarities between the two genres according to the
given metrics during this time period.
We also directly examine the metrics of songs by genre to illustrate these similarities and differences.
Beginning in this portion, we apply the techniques in these broad analyses, to a closer study of the
pop/rock genre. We examine trends in pop/rock song metrics, and how these trends change when data
is weighted by artist popularity, giving insight into the unique traits of top influencers.
We also explore trends in influence, including but not limited to changes in the primary sources of
influence and the dynamic tendency of followers to deviate from their influencers’ genre. From this,
we discovered that majority of pop/rock followers were classified themselves as pop/rock, warranting
further study. Moreover, we discover that over time, the most influential artists in a given time period
become less influential overall, indicating that artists may be taking influence from a broader scope of
artists.
We then employ a subnetwork within modern pop/rock to study the degree similarity between
influencers and their followers. Using a second neural network, we attempt to predict whether an
artist is a follower of a given influencer. This may suggest that the given metrics alone are not capable
of indicating influence alone; we recommend further analysis.
8 February, 2021
To the ICM Society:
In addressing the broad topic of how music evolves through influence, our
report covers a vast array of topics tackled through the use of multiple data-
driven methods. These techniques include the following: developing and
testing similarity metrics to compare influencers and followers, both analyti-
cally and through the use of neural networks; utilizing statistical figures such
as line graphs and bar graphs to convey general trends across time and genre;
and, developing and testing a neural network to determine if an artist’s music
is a particular genre.
The value of these methods lies in their ability to make sense of large amounts
of data and recognize trends unbeknownst to the human eye. We determine
global trends in characteristic metrics of music, identify defining aspects of
genres, and offer insight into the homogeneity of domains in the music world.
However, our procedures inherently rely heavily on the quality of the data
set provided and proper manipulation and interpretation of the data.
Some limitations that were observed in the data set included the following.
First, that artist data given includes active start date, but not peak or active
end date, meaning that some artists’ careers may be represented inaccu-
rately. Given these metrics, one might better be able to establish influence
trends occurs over time; that is, how the relationships between influencers
and followers evolve. Moreover, since the genre of each song is not listed,
only the primary genre of each artist, we adopted the assumption that all
songs by a given artist are of that artist’s main genre, failing to account for
genre changes or experimentation with other genres. In both of these cases,
evaluating data incorporating these changes would eliminate some error lying
within the data.
Nonetheless, given the time allotted for this assignment, we recommend fur-
ther study into certain topics. We suggest that an investigation of how each
of the provided metrics shape listeners’ perception of song similarity would
be useful in creating more precise similarity metric. We also believe that
the creation of additional subnetworks and/or neural networks for different
genres and time periods will provide valuable insight into how music varies
within these domains.
Regards,
2125654
Movements and Measures: Developments in Music
and Musical Influence
2125654
Contents
1 Introduction 2
1.1 Metrics Present . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Assumptions 2
3 Methods 3
3.1 Subnetworks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3.2 Metrics of Similarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
4 Analysis 4
4.1 By Year . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4.1.1 General Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4.1.2 Metrics of Genres over Time . . . . . . . . . . . . . . . . . . . . . 5
4.2 By Genre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.2.1 Classifying Songs by Genre Using Neural Networks . . . . . . . . 8
4.2.2 Metrics of Songs by Genre . . . . . . . . . . . . . . . . . . . . . . 8
4.2.3 Genre Analysis: Pop/Rock . . . . . . . . . . . . . . . . . . . . . . 10
4.3 Influence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.3.1 Influence by Year . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.3.2 Genres of Influencers versus Followers . . . . . . . . . . . . . . . . 14
4.3.3 Subnetwork of Pop/Rock Artists in the 2000s-2010s . . . . . . . . 16
4.3.4 Assessing Influence through Neural Networks . . . . . . . . . . . . 17
5 Discussion 18
5.1 Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.3 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1
Team 2125654 Page 2
1 Introduction
Objective notions regarding music tend to be evasive. The language that non-experts
use to describe it tends to be broad and highly variable, and the interpretation of such
language is shaped by personal perspective. Describing the “feel” of a song can be
reminiscent of describing the taste of a wine; a conversation partner may have a general
idea of what “smoky” means, but they are likely to smile and nod rather than pursue a
more specific understanding of the term. When tracking changes in artists, genres, and
the state of music as a whole, more precise metrics provide a way forward.
Every musical artist has influences, and every musical artist has an impact on the mu-
sic world, however small. Using the data sets provided, collected via Spotify’s API, we
examine this network of influence, using metrics such as energy, danceability, and acous-
ticness (among others) to track chains of influence, as well as trends in genre substance
and popularity over time. We use this network of artists, its subnetworks, and various
analyses based on given quantitative metrics in our investigation.
1.1 Metrics Present
Some metrics present in the song and artist data are more musically precise than others.
Tempo, mode and key are common musical metrics, present in introductory musical
vocabulary [1]. Loudness, using decibels, is straightforward. However, we also consider
some less standard metrics that are present in the data set. We use danceability, a
measure of how suited a song is for dancing; energy, which refers to the intensity or
activity present in the song; and valence, which refers to the overall positive or negative
feel of the song. Acousticness refers to the presence or lack of technological enhancement.
Instrumentalness refers to the presence or lack of vocals, while speechiness refers to that
of spoken words. Liveness is a measure of whether or not an audience is detected. In
broader analyses, we also incorporate songs’ popularity. We analyze these various metrics
provided by the data in order to better understand how music changes across time, genre,
and through influence.
2 Assumptions
While our data set is not comprehensive, we assume in our analysis that it is a mostly
comprehensive view of the music world. We note that this assumption suffers for the
earliest and most recent years (such as the 2010s and the 1920s) due to less data present
for these years. For this reason, there exist cases where data from these years may be
somewhat misleading. We attempt to adjust our measures to account for this, whether
that be via omission of data from these years, standardization in some form, or simply
exercising caution when accepting conclusions based on these data. When any of these
actions are taken, we describe them.
We also assume that the number of followers an artist has is an indication of how influen-
剩余21页未读,继续阅读
资源评论
小兔子平安
- 粉丝: 215
- 资源: 1940
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功