Visual dimension analysis based on dimension subdivision
- PDF / 1,894,380 Bytes
- 15 Pages / 595.276 x 790.866 pts Page_size
- 9 Downloads / 264 Views
R E G UL A R P A P E R
Yi Zhang • Chenxi Yu • Ruoqi Wang • Xunhan Liu
Visual dimension analysis based on dimension subdivision
Received: 1 July 2020 / Revised: 1 July 2020 / Accepted: 9 August 2020 The Visualization Society of Japan 2020
Abstract Visualization of multidimensional data has always been a research hotspot. Dimensional analysis is an efficient way to solve multidimensional problems. The current dimensional analysis methods mostly consider that all dimension correlations are at the same granularity, but actually the correlation between dimensions may be multi-scale. Multi-scale dimensions can also reflect the multi-scale data association mode, which is of certain value for analyzing the hidden information of multidimensional data. In this paper, we propose a method of dimension subdivision to resolve the multi-scale correlations between dimensions. To explore the multi-scale complex relationship between dimensions, we subdivide the original dimensions into finer sub-dimensions and build a graph-based data structure of the correlations to partition strongly relevant and irrelevant dimensions. We also proposed D-div, a visual dimension analysis system to support our method. In D-div, we provide visualization and interaction techniques to explore subdivided dimensions. Via case studies with two datasets, we demonstrate the effectiveness of our method of dimension subdivision. Keywords Multidimensional data Dimensional analysis Correlation analysis Multidimensional visualization
1 Introduction Multidimensional data are prevalent in daily life and scientific research. The complexity of multidimensional data is not only reflected in a large number of dimensions and large data scale, but also in the complex relationships between dimensions. Therefore, the analysis of relationships between dimensions is also of great significance to the study of multidimensional data. Visual dimensional analysis of multidimensional data has been a research hotspot in recent years. Currently, there are two major categories of multidimensional data visualization methods. The first group of techniques is to only display important dimensions, including dimension reduction, principal component analysis (PCA), dimension analysis and so on. However, some information in the original dimensions may
Y. Zhang (&) C. Yu R. Wang X. Liu College of Intelligence and Computing, Tianjin University, Tianjin, China E-mail: [email protected] C. Yu E-mail: [email protected] R. Wang E-mail: [email protected] X. Liu E-mail: [email protected]
Y. Zhang et al.
be lost, and in this case the results from analyzing the relationship between the dimensions may be biased. The second group of techniques, including parallel coordinates (Inselberg and Dimsdale 1990) and scatter plot matrix (Becker et al. 1987), contributes to dimensional analysis by showing more dimensions in limited screen space. However, as the dimension increases, the overlapping problem will make it difficult to obtain effective information. And it analyzes one di
Data Loading...
