Lambiotte R, Ausloos M
SUPRATECS, Université de Liège, B5 Sart-Tilman, B-4000 Liège, Belgium.
Phys Rev E Stat Nonlin Soft Matter Phys. 2005 Dec;72(6 Pt 2):066107. doi: 10.1103/PhysRevE.72.066107. Epub 2005 Dec 7.
In this paper, we analyze web-downloaded data on people sharing their music library, that we use as their individual musical signatures. The system is represented by a bipartite network, nodes being the music groups and the listeners. Music groups' audience size behaves like a power law, but the individual music library size is an exponential with deviations at small values. In order to extract structures from the network, we focus on correlation matrices, that we filter by removing the least correlated links. This percolation idea-based method reveals the emergence of social communities and music genres, that are visualized by a branching representation. Evidence of collective listening habits that do not fit the neat usual genres defined by the music industry indicates an alternative way of classifying listeners and music groups. The structure of the network is also studied by a more refined method, based upon a random walk exploration of its properties. Finally, a personal identification-community imitation model for growing bipartite networks is outlined, following Potts ingredients. Simulation results do reproduce quite well the empirical data.
在本文中，我们分析了从网络下载的关于人们分享其音乐库的数据，我们将这些数据用作他们的个人音乐特征。该系统由一个二分网络表示，节点是音乐团体和听众。音乐团体的受众规模呈现幂律分布，但个人音乐库规模在小值时存在偏差，呈指数分布。为了从网络中提取结构，我们专注于相关矩阵，并通过去除相关性最低的链接对其进行过滤。这种基于渗流思想的方法揭示了社会群体和音乐流派的出现，这些通过分支表示进行可视化。不符合音乐行业定义的常规清晰流派的集体收听习惯的证据表明了一种对听众和音乐团体进行分类的替代方法。还通过一种更精细的方法研究了网络结构，该方法基于对其属性的随机游走探索。最后，按照Potts模型要素概述了一个用于二分网络增长的个人识别 - 社区模仿模型。模拟结果与实证数据相当吻合。
