摘要
真实世界中的网络大多都是带有时变属性的动态网络。动态网络可视化的目标是更好地帮助用户分析网络数据,发现数据特征。针对现有动态网络状态演化可视化方法存在的状态聚类和演化轨迹不明显的问题,提出一种引入特征分量相似度的动态网络状态演化可视化方法。该方法通过预先计算降维数据的本征维数,避免过度降维造成的数据缺失,最大程度保留时间步数据原有特征;再求得时间步特征分量相似度,将相似度融入力导引布局,加入相似力和万有引力实现更明显的状态聚类和演化轨迹。通过与Elzen等人的方法进行对比实验证明,本文提出的可视化方法能够直观地展示更多的动态网络演化状态。
Most of the networks in real world are dynamic networks with time-varying attributes. The goal of dynamic network visualization is assisting users in analyzing network data,discovering data features in a better way. Aiming at the problem that the existing dynamic network state evolution visualization methods cannot express the state cluster and evolution clearly,a method that introduces eigenvector similarity to visualization of state evolution of dynamic networks is proposed. To avoid the data loss caused by excessive dimension reduction,keep the original characteristics of the time steps,this method reduces the data to its intrinsic dimension by using the dimensionality reduction technique and maximum likelihood estimation. It calculates the eigenvector similarity and introduces the force-directed layout algorithm. By adding the similarity-force and gravity,the layout of state clusters and evolution is much clearer. The experiments which compare with Elzen’s method show that the proposed visualization method can present evolution states of dynamic network more clearly and directly.
作者
刘超
万莹
LIU Chao;WAN Ying(State Grid Jibei Information ＆ Telecommunication Company, Beijing 100053,China)
出处
《信息技术》
2018年第5期115-120,共6页
Information Technology
关键词
动态网络
状态演化
网络可视化
特征分量相似度
力导引算法
dynamic networks
state evolution
networks visualization
eigenvector similarity
forcedirected algorithm
