Visual exploration of multiway dependencies in multivariate data

Abstract

Analyzing dependencies among variables within multivariate data is an important and challenging problem, especially when the number of data points is large, the number of variables is high, or multiway dependencies are of interest. Several visualization methods have been proposed to aid in the exploration of such information through the direct visualization of the summary statistics. These methods are typically limited to the study of all possible pairwise relationship but in a manner that does not scale to large multidimensional data. In cases where 3-way relationships are investigated, only subsets of dimensions are considered. In this paper, we propose a novel technique for analyzing multiway dependencies through an overview+detail visualization. In this approach, the overview represents all pairwise, 3-, and 4-way dependencies in the data using glyphs that provide a global visual exploration interface for selecting candidate relationships. Exploration is supported through interactive filtering, sorting, zooming, and selection operations. Once selected, the detailed view helps in developing an inference by providing specific information about those selected variables. Various use cases demonstrate how our approach helps to explore multiway dependencies efficiently in large datasets.

Video


Downloads

Download the Paper Download the Video Download the BiBTeX

Citation

Hoa Nguyen, Paul Rosen, and Bei Wang. Visual exploration of multiway dependencies in multivariate data. In SIGGRAPH ASIA 2016 Symposium on Visualization, page 2. ACM, 2016.

Bibtex


@inproceedings{Nguyen.2016.SASV,
  title = {Visual exploration of multiway dependencies in multivariate data},
  author = {Hoa Nguyen and Paul Rosen and Bei Wang},
  booktitle = {SIGGRAPH ASIA 2016 Symposium on Visualization},
  pages = {2},
  year = {2016},
  organization = {ACM},
  abstract = {Analyzing dependencies among variables within multivariate data is an important 
    and challenging problem, especially when the number of data points is large, the number of 
    variables is high, or multiway dependencies are of interest. Several visualization methods have 
    been proposed to aid in the exploration of such information through the direct visualization of 
    the summary statistics. These methods are typically limited to the study of all possible pairwise 
    relationship but in a manner that does not scale to large multidimensional data. In cases where 
    $3$-way relationships are investigated, only subsets of dimensions are considered. In this paper, 
    we propose a novel technique for analyzing multiway dependencies through an overview+detail 
    visualization. In this approach, the overview represents all pairwise, $3$-, and $4$-way 
    dependencies in the data using glyphs that provide a global visual exploration interface for 
    selecting candidate relationships. Exploration is supported through interactive filtering, 
    sorting, zooming, and selection operations. Once selected, the detailed view helps in developing 
    an inference by providing specific information about those selected variables. Various use cases 
    demonstrate how our approach helps to explore multiway dependencies efficiently in large datasets. }
}