Project A03

A03 | Quantification of Visual Explainability

Prof. Daniel A. Keim, University of Konstanz
Email | Website

Prof. Michael Sedlmair, University of Stuttgart
Email | Website

Frederik Dennig, University of Konstanz – Email | Website

Lucas Joos, University of Konstanz – Email | Website

About this project
Results

High-dimensional data analysis requires dealing with numerous challenges, such as selecting meaningful dimensions, finding relevant projections, and removing noise. As a result, the extraction of relevant and meaningful information from high-dimensional data is a difficult problem. This project aims at advancing the field of quality-metric-driven data visualization with the central research question of how to quantify the quality of transformations and mappings of high-dimensional data for visual analytics.

Research Questions

How can we measure and quantify the quality of a visualization? In which way do methods in the data space differ from methods in the image space?

How can we compare the measured quality of a visualization with the perception of a human?

How can the user be involved into a quality-metric-driven process of visual mappings and transformations?

What is the influence of perceptual effects on quality measures? Can we enhance the visual representation of information by introducing perceptual effects into visualizations?

Comparison of regular parallel coordinates with our slope-dependent polyline rendering. Parallel coordinates face two problems: Diagonal lines are rendered more closely (a), and zig-zag patterns are perceived as clusters, although there are no such clusters in the data (c). We render each line segment based on its slope between two axes. Cluster density is not distorted (b), and the zig-zag pattern effect is reduced (d).

Fig. 1: Comparison of regular parallel coordinates with our slope-dependent polyline rendering. Parallel coordinates face two problems: Diagonal lines are rendered more closely (a), and zig-zag patterns are perceived as clusters, although there are no such clusters in the data (c). We render each line segment based on its slope between two axes. Cluster density is not distorted (b), and the zig-zag pattern effect is reduced (d).

Comparison of similarity (SIM) and dissimilarity (DIS) based ordering using identical data records. Our results show that users perform better when the glyphs represent salient shapes and spikes, which is achieved by a dissimilarity-based ordering of the dimensions.

Fig. 2: Comparison of similarity (SIM) and dissimilarity (DIS) based ordering using identical data records. Our results show that users perform better when the glyphs represent salient shapes and spikes, which is achieved by a dissimilarity-based ordering of the dimensions.