Winglets: Visualizing Association with Uncertainty in Multi-class Scatterplots

论文传送门

作者

深圳大学

• Min Lu
• Shuaiqi Wang
• Yang Yue
• Daniel Cohen-Or
• Hui Huang

海法大学

• Joel Lanir

以色列特拉维夫大学

• Noa Fish

摘要

这项工作提出了“小翼”，它是对经典散点图的增强，可以通过改善关联点的感知和与其相关聚类的不确定性来更好地感知多个类别。小翼被设计为属于数据点的一对双向笔划，它利用格式塔的闭合原理来塑造对簇形式的感知，而不是使用显式的除法编码。通过对长度和方向这两个主要属性的微妙设计，小翼使观看者能够对群集进行心理上的补全。通过进行一项受控用户研究，检查了小翼感知聚类关联的效率和某些点的不确定性。结果表明，小翼将点更紧密地关联到聚类中，并改善了关联不确定性的感知。

Introduction

Visualize Multiclass in Scatterplot

• To support better perception of groups
• To convey assignment uncertainty

Related work

Perceptual Grouping

• Gestalt Principles
  • a set of principles in psychology
  • about how objects are visually perceived as groups by human

Example - MNIST Dataset

Proximity Principle

• Layout Methods
  • PCA, t-SNE, MDS, etc.

Quality of Layouts: DSC

Similarity Principle

• Coloring Strategies
• Coloring Mixing Techniques

Continuity Principle

• Enclose with a Continuous Boundary

Winged Scatterplot

Disign choices and winglets construction

Orientation
实验比较

构造过程：

Length:
Mapping from uncertainty to winglets’ length L
$l(i) a * s_{i}^{n}+b$
where $s_{i}$ is the Silhouette Index of point $i$

Evaluation

44 participants, 4 tasks, time cost and accuracy

Conclusions

• Winglets: multi-class scatterplot visualization
  • Gestalt Principle of Closure
  • Association and assignment uncertainty

Key Message: Winglets are a visual encoding technique

• Provides a strong perception of grouping
• Significantly enhances multi-class scatterplot visualization