Winglets: Visualizing Association with Uncertainty in Multi-class Scatterplots

IEEE Transactions on Visualization and Computer Graphics (Proceedings of InfoVis 2019)


Min Lu1    Shuaiqi Wang1    Joel Lanir2    Noa Fish3    Yang Yue1    Daniel Cohen-Or1,3    Hui Huang1*

1Shenzhen University    2The University of Haifa    3Tel Aviv Univeristy



Fig. 1. (Left) A common visualization of a multi-class scatterplot uses colors to associate the points to their clusters. The uncertainty is visually encoded by darkening the colors. (Right) Little wings attached to the points have a stronger descriptive power of association based on the Gestalt principles. The wings length expresses the associating uncertainty of a point.



Abstract

This work proposes Winglets, an enhancement to the classic scatterplot to better perceptually pronounce multiple classes by improving the perception of association and uncertainty of points to their related cluster. Designed as a pair of dual-sided strokes belonging to a data point, Winglets leverage the Gestalt principle of Closure to shape the perception of the form of the clusters, rather than use an explicit divisive encoding. Through a subtle design of two dominant attributes, length and orientation, Winglets enable viewers to perform a mental completion of the clusters. A controlled user study was conducted to examine the efficiency of Winglets in perceiving the cluster association and the uncertainty of certain points. The results show Winglets form a more prominent association of points into clusters and improve the perception of associating uncertainty.





Fig. 6. Winglets encode the association and its uncertainty via its orientation and length.

Fig. 7. Orienting Procedure: (a) Gaussian kernel density map is calculated for the plot. (b) isocontours of sampled densities are extracted by Marching Squares algorithm. (c) a global reference contour is picked for the coherent perception of grouping, before splitting into multiple contour siblings (annotated in (b)). (d) contours are interpolated from outside to inside. (e) points grow their wings along the orientation as their nearest points on the contours.



Fig. 8. Orientation Choice of Winglets: there are two main types of wings’ orientation according to whether the global form is open or closed. Compared with Contour, orienting towards Centroid, Line or Boundary Circle may fail in some cases (marked by dashed line boxes).





Fig. 10. Winglets in MNIST: 1200 samples are embedded in a 2D plot using t-SNE projection. Each class is assigned a unique color; with Winglets, associations are perceptually more pronounced when clusters are broken into parts (A), or are overlapping with one another (B, C). Winglets convey the association uncertainty, such as high association certainty points situated far away from the majority of their cluster (D).



Fig. 12. Cluster and Overlap conditions: three options for the amount of clusters with and without Winglets; three levels of overlap with and without Winglets. Note that in the figures the scatterplots are aligned. In the experiment, a random rotation was performed to vary each of the scatterplots.



Acknowledgement

We thank the reviewers for their valuable comments. This work is supported in parts by NSFC (61802265, 41671387, 61761146002, 61861130365), LHTD (20170003), Guangdong Provincial Natural Science Foundation (2018A030310426, 2015A030312015), and the National Engineering Laboratory for Big Data System Computing Technology.



Bibtex
@article{Winglets19,
title = {
Winglets: Visualizing Association with Uncertainty in Multi-class Scatterplots},
author = {Min Lu, Shuaiqi Wang, Joel Lanir, Noa Fish, Yang Yue, Daniel Cohen-Or, Hui Huang},
journal = {},
volume = {},
number = {}, 

pages = {}, 

year = {2019},


Downloads(faster for people in China)

Downloads(faster for people in other places)