Visualization M. Behrisch , M. Blumenschein, N. W. Kim, L. Shao, M. - - PowerPoint PPT Presentation

Visual Computing Group, Harvard

for Information

• M. Behrisch, M. Blumenschein, N. W. Kim, L. Shao,
• M. El-Assady, J. Fuchs, D. Seebacher, A. Diehl, U. Brandes,
• H. Pfister, T. Schreck, and D. Weiskopf, D. A. Keim

Visualization

Visual Computing Group, Harvard

Appropriateness Effectiveness Efficiency Expressiveness Expressiveness Usefulness Usability Interpretability

Visual Computing Group, Harvard

Patterns in Information Visualizations

3

NO NO OK OK

Conceptual/ Pattern Space

Task Task

(NO) (NO)

Visual Computing Group, Harvard

Patterns in Information Visualizations

4

NO NO OK OK (NO) (NO)

Conceptual/ Pattern Space

Task Task

Visual Computing Group, Harvard

Patterns in Information Visualizations

5

Log Log Lin Lin

Visualization Parameter Dataset Characteristics Visualization Parameter User Understandability

Visual Computing Group, Harvard

Optimization Algorithm

Quality

Criterion

Quality Metric

arg

q(𝜚 | D, U, T)

𝜚 ∈ Φ

min max

D U T 𝜚

Visual Computing Group, Harvard

Structure and Goals of the Survey

7

Multi- and High-Dimensional

Relational Data Geo-Spatial Data Sequential/Temporal Text Data

Research Goals Reference Manual for QM Establish Common Vocabulary Open Challenges and Future Research Directions

Visual Computing Group, Harvard

Structure and Goals of the Survey

8

Multi- and High-Dimensional

Relational Data Geo-Spatial Data Sequential/Temporal Text Data

Research Goals Reference Manual for QM Establish Common Vocabulary Open Challenges and Future Research Directions

In total ~300 Paper

B A C

…

134 Paper Selection

per Vis. Type

14 Vis Types Categorization

per Vis. Type

Insight Generalization

Independent of

• Vis. Types

B A C B A C B A C B A C

For each Vis Type

• 1. Visualization Description
• 2. Why do we need QMs?
• 3. Typical Analysis Tasks
• 4. Summary of Approaches
• 5. Evaluations Methods
• 6. Open Research Questions

Visual Computing Group, Harvard

Clutter Removal vs Pattern Retrieval

Quality Metric

arg

q(𝜚 | D, U, T)

min max

Visual Computing Group, Harvard

Auto-Sampling – Clutter Removal

Grouping Outlier Correlation Trend

Parallel Coordinates Patterns and Tasks

Optimization

[Ellis2006] Overplotted% Percentage of pixels containing more than one plotted point Overcrowded% Percentage of plotted points hidden behind a pixel Hidden% Percentage of plotted points that are hidden due to being overplotted

Visual Computing Group, Harvard

Scatter Plots

Scagnostics – Pattern Retrieval

11 a b

Grouping Correlation Outlier Trend Patterns and Tasks

Optimization

[Wilkinson2006] Convex: Area of Alpha Shape divided by area of Convex Hull Skinny: Ratio of perimeter to area of the Alpha Shape Stringy: Ratio of 2-degree V in MST to # of V > 1-degree

v v v v

Convex Hull Alpha Shape Minimum Spanning Tree

v v

[Dang2014]

Visual Computing Group, Harvard

Quality Metric

arg

q(𝜚 | D, U, T)

min max

Reduces Cognitive Overload Focuses on Analysis Task

Pattern-Driven Analysis Clutter Removal Analysis Scenarios/Tasks for QM Search for Data Groups Search for Outliers Search for Dimension Relations Preservation Task Data- and Visualization Specific Tasks

Visual Computing Group, Harvard

Explicit and Implicit QM

Quality Metric

arg

q(𝜚 | D, U, T)

min max

Visual Computing Group, Harvard 14

Grouping Correlation Trend Outlier

Line Spiral Hilbert

Patterns and Tasks Pixel-based Techniques

Optimization

Noise Dissimilarity – Explicit QM

[Albuquerque2011]

Visual Computing Group, Harvard

Force Directed Layout – Implicit QM

15

Patterns and Tasks Node-Link Diagrams

Optimization

Connectivity

Grouping Item Path Group

Connectivity Connectivity

[Purchase2002]

• M. Kaufmann und D. Wagner, editors. Drawing Graphs —

Bends Edge crossings Minimum Angles Orthogonality Symmetry

Input: Graph G = (V,E) Start with random placement of vertices Repeat k times (k constant){ 1. for all v in V do Calculate the repelling forces on v that are excerted by V \ v 2. for all e = (u, v) in E do Calculate the attracting forces between u and v 3. for all v in V do Add repelling and attracting forces Move v in direction F(v) }

[Wang2018]

Cluster Overlap Cluster Overlap + Circle Constraint Cluster Overlap + Star Constraint Cluster Overlap + Circle + Star Constraint

Visual Computing Group, Harvard

Data Space vs Image Space

Quality Metric

arg

q(𝜚 | D, U, T)

min max

Visual Computing Group, Harvard

Quality Metric

arg

q(𝜚 | D, U, T)

min max

Transformed Data Visual Structures Views

Data Transformation Visual Mapping Rendering View Transformation

User

Data Space

Quality Metrics

Image Space

Quality Metrics Adaption

Evaluation

Quality-Metrics-Driven Automation

User / Task Concept Intuition Computation Aspects Data Specification

QM Influence

Biases

Source Data

[adapted from Bertini2011]

Visual Computing Group, Harvard

TextFlow – Data Space QM

18

Split / Merge Trend Parallel

[Cui2011]

Patterns and Tasks

Optimization

Stacked Charts Topic Flow Topic Bundles Thread

Visual Computing Group, Harvard

Magnostics – Image Space QM

19 Optimization

Pattern Response

C1 C1

Pattern Variability

C2 C2

Pattern Sensitivity

C3 C3

Pattern Discrimination

C4 C4

Low High

[Behrisch2016]

Matrix Patterns and Tasks Grouping Item

Connectivity

Path Group

Connectivity Connectivity

Visual Computing Group, Harvard

Quality Metrics Landscape

20

Cognitive Process / Complexity

High-Level Meta-Perception / User Mid-Level Perception / Task Low-Level Perception

Memorability, Understanding, Confidence, Faithfulness, Trustworthiness, Cognitive Biases, Engagement, User Level, Conventions, Aesthetics, Joyfulness Patterns versus Anti-Patterns, Clutter-reduction, Task-effectiveness “Overview-First & Details-on-Demand”, “Search, show context, expand on demand” Preattentive Processing, Gestalt Laws, Visual Variables, Change Blindness, Just-Noticeable-Differences

Data Space vs Image Space Explicit vs Implicit QM Clutter Removal vs Pattern Retrieval

Visual Computing Group, Harvard

Quality Metrics Landscape

21

Cognitive Process / Complexity

High-Level Meta-Perception / User Mid-Level Perception / Task Low-Level Perception

Memorability, Understanding, Confidence, Faithfulness, Trustworthiness, Cognitive Biases, Engagement, User Level, Conventions, Aesthetics, Joyfulness Patterns versus Anti-Patterns, Clutter-reduction, Task-effectiveness “Overview-First & Details-on-Demand”, “Search, show context, expand on demand” Preattentive Processing, Gestalt Laws, Visual Variables, Change Blindness, Just-Noticeable-Differences

Color Research

[Gramazio2016] [Mittelstädt2015]

39 studies about human perception in 30 minutes

[Elliott2016]

Visual Computing Group, Harvard

Quality Metrics Landscape

22

Cognitive Process / Complexity

High-Level Meta-Perception / User Mid-Level Perception / Task Low-Level Perception

Memorability, Understanding, Confidence, Faithfulness, Trustworthiness, Cognitive Biases, Engagement, User Level, Conventions, Aesthetics, Joyfulness Patterns versus Anti-Patterns, Clutter-reduction, Task-effectiveness “Overview-First & Details-on-Demand”, “Search, show context, expand on demand” Preattentive Processing, Gestalt Laws, Visual Variables, Change Blindness, Just-Noticeable-Differences

Memorability

[Skau2017] [Borkin2015]

Aesthetics Joyfulness

Visual Computing Group, Harvard

Quality Metrics Landscape

23

Mid-Level Perceptual Quality Metrics

User / Task Concept Intuition Computation Aspects Quality Metric Influence Clutter Removal Pattern-Driven Analysis

Focuses on Analysis Task Reduces Cognitive Overload

Data Spec.

Cognitive Process / Complexity

High-Level Meta-Perception / User Mid-Level Perception / Task Low-Level Perception

Memorability, Understanding, Confidence, Faithfulness, Trustworthiness, Cognitive Biases, Engagement, User Level, Conventions, Aesthetics, Joyfulness Patterns versus Anti-Patterns, Clutter-reduction, Task-effectiveness “Overview-First & Details-on-Demand”, “Search, show context, expand on demand” Preattentive Processing, Gestalt Laws, Visual Variables, Change Blindness, Just-Noticeable-Differences

Visual Computing Group, Harvard

Discussion and Findings

24

Quality Metric

arg

q(𝜚 | D, U, T)

min max

T

> Implicit, domain-inspired, pot. subjective expectation > What if pattern is not known apriori? Which QM? > Majority of QMs do not describe visual pattern

Which QM favors which visual pattern?

Visual Computing Group, Harvard

Discussion and Findings

25

> Noise (in-)variances and robustness toward skewed distributions > Bad QM must mean no pattern Quality Metric

arg

q(𝜚 | D, U, T)

min max

What are extreme cases that a QM can deal with?

Visual Computing Group, Harvard

Discussion and Findings

26

Quality Metric

arg

q(𝜚 | D, U, T)

min max

> Some vis subdomains share similar concepts > Set of base patterns in both visualizations

[Fink2013] [Heer2006]

Is QM research transferable among visualization types?

Visual Computing Group, Harvard

Discussion and Findings

27

Quality Metric

arg

q(𝜚 | D, U, T)

min max

Are QMs equally descriptive?

[Wang2018]

> QM for recommendation of visualization technique > But, only standard patterns (not domain dependent)

Visual Computing Group, Harvard

Discussion and Findings

28

Quality Metric

arg

q(𝜚 | D, U, T)

min max

Evaluation of Quality Metrics

> Heuristic, quantitative, pattern-focused QM research not backed up (enough) by perception QM studies > Design recommendations solely base their recommendations on studies

[Pandey2016] [KimBylinskii 2018]

Visual Computing Group, Harvard

Multi-Criteria & Task-Adapted QM.

Mixture of patterns; Tasks change in exploratory settings

Research Directions

29

Interactive & Human-Supported Quality Steering.

Algorithms can benefit from the user’s input to produce high quality results.

Machine Learning.

Deep learning possible, iff (1) suff. large training set; (2) appropriate architecture

Closing the Gap to Higher-level Perceptual QM.

Central goal is still reduce cognitive overload.

Intelligently navigate pattern space. Visual Support and Visual Analytics is needed. Deep-Learning based QM. QM can help to build understanding and trust.

f(t) =

Visual Computing Group, Harvard

Take Home Messages

30

visualquality.dbvis.de

Task and Pattern-based Quality Metrics.

Choose the right QM! More evaluation is necessary.

Visual Exploration Interfaces.

Needed to make use of QMs in the wild.

Visual Analytics will change the field (once again).

Opening the Black-Box will lead to novel algorithms.

Visual Computing Group, Harvard

Thank YOU

31

visualquality.dbvis.de

visualquality.dbvis.de