Team 2125654 Page 3
tial that artist was. This assumption may be problematic when analyzing the influence
of more recent artists. Moreover, this assumption assumes that influencing one artist
counts the same amount as influencing another artist, although intuitively we know that
artists can have a different amount of influence on one artist than another. It also fails
to account for chains of influence; however, we consider this factor at several points in
our analysis.
Additionally, the data set contains some artists of unknown genre. We include these data
points in our procedures, but note that they exist in a very small number, and therefore
do not give any significant weight to properties and trends they exhibit.
3 Methods
3.1 Subnetworks
In order to observe the influence of the influencers on followers, we require a method of
easily collecting the followers of a given influencer. Hence, using the artist influence data,
we generate a directed graph in R using an adjacency matrix. With this adjacency matrix,
we can perform operations such as applying similarity comparisons between influencers
and followers, and generating subnetworks.
In order to generate relevant subnetworks of the given data, we require an algorithm to
generate all neighbors (or all subsequent “followers”) of a given node (“influencer”) and
vice versa. For this, we implemented a breadth-first search traversal algorithm in R.
3.2 Metrics of Similarity
We primarily analyze similarity based on seven metrics, all of which are provided on scales
from 0 to 1. These metrics are danceability, energy, valence, acousticness, instrumental-
ness, liveness, and speechiness. Throughout this paper, we refer to our seven chosen
metrics collectively as “similarity metrics.”
We choose to eliminate some metrics from our similarity analyses because they are cate-
gorical, and others because of scaling. Because of this, our similarity analyses may fail to
recognize the influence of the other metrics recorded in the data set on similarity between
artists.
Once deciding to consider seven metrics ranging from 0 to 1, we considered the intuition
behind genre and song similarity. Consider genres as regions in space, with songs of a
given genre falling in its region. Each of our seven metrics represents a dimension, with
some genres or songs being close with regard to some metrics and far with regard to
others. Thus, the vector of the means of a genre’s metrics would define a single point
in space, as would the vector of a song’s metrics. From this, we define our measure of
similarity between genres, and by extension, songs, as the Euclidean distance between
剩余21页未读，继续阅读
小兔子平安
- 粉丝: 243
- 资源: 1940
我的内容管理
展开
最新资源
- 多模态联合稀疏表示在视频目标跟踪中的应用
- Kubernetes资源管控与Gardener开源软件实践解析
- MPI集群监控与负载平衡策略
- 自动化PHP安全漏洞检测：静态代码分析与数据流方法
- 青苔数据CEO程永：技术生态与阿里云开放创新
- 制造业转型： HyperX引领企业上云策略
- 赵维五分享：航空工业电子采购上云实战与运维策略
- 单片机控制的LED点阵显示屏设计及其实现
- 驻云科技李俊涛：AI驱动的云上服务新趋势与挑战
- 6LoWPAN物联网边界路由器：设计与实现
- 猩便利工程师仲小玉：Terraform云资源管理最佳实践与团队协作
- 类差分度改进的互信息特征选择提升文本分类性能
- VERITAS与阿里云合作的混合云转型与数据保护方案
- 云制造中的生产线仿真模型设计与虚拟化研究
- 汪洋在PostgresChina2018分享：高可用 PostgreSQL 工具与架构设计
- 2018 PostgresChina大会：阿里云时空引擎Ganos在PostgreSQL中的创新应用与多模型存储
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
