平行坐标系聚类数据绑定绘制评价方法  

Evaluation of a Cluster Bundling Technique for Parallel Coordinates Plot Based on Distribution Feature

作　　者：巫滨 曹卫群[1] 杨波[1] Wu Bin;Cao Weiqun;Yang Bo(School of Information Science and Technology, Beijing Forestry University, Beijing 100083;School of Art and Design, Henan University of Science and Technology, Luoyang 471003)

机构地区：[1]北京林业大学信息学院,北京100083 [2]河南科技大学艺术与设计学院,洛阳471003

出　　处：《计算机辅助设计与图形学学报》2017年第11期2047-2056,共10页Journal of Computer-Aided Design & Computer Graphics

基　　金：中央高校基本科研业务费专项资金(2015ZCQ-XX);河南省科技厅软科学研究基金(142400410036)

摘　　要：针对聚类绑定簇间分离时产生的聚类中心偏移和线束布局等问题,从数据分布认知的角度提出可视化评估方法.首先构建数据相关性指数、聚类中心准确性指数、离散度指数等5个指标作为数据分布绘制效果的评价量规;然后设计各指标的视觉映射方式,与对应的分布属性实际值作相关性分析,以实现分布特征的量化视觉评价;最后对3种不同的平行坐标视图进行实例分析和比对.基于评估实验,验证了数据分布特征映射准确性的改善方法,以及维度重排和绑定参数等因素对评价结果的影响.In order to solve problems caused by cluster separation during bundling,such as layout and cluster center deviation,we proposed an evaluation method on cluster bundling from distribution attributes.Firstly,we developed the measure gauges consisted of5elements,such as correlation,cluster center,and dispersion.Secondly,we designed the visual mapping method of each distribution attribute and analysed the correlation between the assessed value and actual value to support the gauge.Finally,we evaluated the cluster bundling presentation in several formats.Through several evaluation experiments,we verified the better way to mapping the data distribution feature in parallel coordinates,and the influence of dimension reordering and parameter selection in visualization.

关 键 词：评价量规 聚类绑定 分布特征 维度重排 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]