Post

DAVI Visualization Design redo

Posted Updated
By Wei Xiong
8 min read
DAVI Visualization Design redo

Topic: Transitioning from Interaction Techniques to Data Visualization Design Processes, Data & Task Abstractions, and the Design Activity Framework

Introduction & Recap

  • The professor starts by greeting the class and notes that today’s lecture is structured into two parts:
    1. Briefly finish the few leftover slides from the “Interaction” lecture (previous week).
    2. Dive into the design process for data visualization.
  • As usual, there is a short quiz at the beginning of the lecture to “wake everyone up.”

Public Service Announcement: Danish National Championships in AI

  • The annual Danish National Championships in AI are taking place this fall.
  • There is a prize of around 40,000 kroner.
  • Students can participate individually or in teams.
  • The professor suggests leveraging visualization plus machine learning or deep learning skills from current courses.

Quiz Review

Context: The quiz focuses on concepts from the previous lecture related to interaction techniques.

Questions and Answers:

  1. Question: How do we call the diagrams that show a zoom level on the y-axis and a simplified screen space on the x-axis?
    • Possible Answers: Angular plots, Two and a half (2.5D) representations, Sunburst, or Space-scale diagrams.
    • Correct Answer: Space-scale diagrams.
    • Explanation: These diagrams show different zoom levels over space. Sunburst diagrams are related to tree data, not space-scale. Angular plots was a made-up term. 2.5D representation is not correct here.
  2. Question: Interactive lenses are an example of which principle?
    • Possible Answers: Overview and detail, Pan and zoom, Drag and drop, Focus and context.
    • Correct Answer: Focus and context.
    • Explanation: Interactive lenses embed a high-detail focus region within a contextual overview. Overview+detail separates the two views; focus+context integrates them into one.
  3. Question: The combination of panning and zooming, giving both origin and destination in view, is called what?
    • Possible Answers: View switching, Smooth navigation, Interactive lens, Overview and detail.
    • Correct Answer: Smooth navigation.
    • Explanation: Smooth navigation (Nui & FatBike concept) provides a fluid transition between two viewpoints, ensuring both start and end points remain visible during the navigation.
  4. Question: Which uses temporal separation for overview and detail?
    • Possible Answers: Interactive lenses, Zooming, Brushing and linking, Minimap.
    • Correct Answer: Zooming.
    • Explanation: Zooming requires time passing to move from an overview to a detailed view. Interactive lenses are focus+context, brushing and linking is for multiple coordinated views, and a minimap is spatial separation, not temporal.
  5. Question (Technical Issues): Another question about lenses and parallel pipelines.
    • The professor clarifies that interactive lenses are realized using a parallel visualization pipeline, not overview+detail, not brushing and linking, etc.

Summary of Quiz Concepts:

  • Space-scale diagrams represent zoom and space in a 2D plot.
  • Interactive lenses are a focus+context technique, not overview+detail.
  • Smooth navigation combines panning and zooming to keep origin and destination visible.
  • Zooming is a temporal approach to overview+detail.
  • Interactive lenses are realized using a parallel visualization pipeline approach for flexibility and stackability.

Tangible Interaction and Pseudo-Haptics

Tangible Interaction

  • Example: A multi-touch table with physical rings (3D printed) placed on the table surface.
  • The table detects each ring (via markers).
  • Each ring acts as a “lens”: placing a ring over a map switches the underlying layer or view (e.g., from satellite to street map), turning it like a dial adjusts blend parameters.
  • Multiple rings can be stacked to combine lens functions (smart lenses).
  • Tangible props can facilitate exploration in ways a mouse cannot, lowering barriers to entry for non-technical users.

Another Example: Paper Prop for 3D Interaction

  • Using a depth-sensing camera and a coded piece of paper rolled up into a cylinder.
  • Users manipulate 3D microscopy data (fibers in tissues) by moving and rotating a paper prop.
  • This tangible user interface helps domain experts intuitively explore complicated 3D data.
  • Although advanced, such interfaces show how tangible interaction can simplify complex navigation tasks.

Pseudo-Haptics (Lightweight Haptics)

  • Without special hardware, modify cursor speed or cursor size to “feel” data density.
  • Example: Slowing cursor movement over dense data regions (like a “bump”).
  • Changing cursor size for fine-grained navigation.
  • Similar to “snap to grid” in PowerPoint—an existing example of pseudo-haptics.
  • Useful for subtle guidance and improving data exploration without extra hardware.

Transition to Data Visualization Design Process

  • Brief mention of software engineering slides (postponed).
  • Main focus now: The design process for data visualization.

Understanding the Visualization Problem

Data-Information-Knowledge-Wisdom (DIKW) Hierarchy

  • Data: Raw facts, measurements, or observations.
  • Information: Derived from data (distributions, relations, patterns).
  • Knowledge: Understanding causal relations, building models, functional dependencies.
  • Wisdom: Using knowledge for predictions, decision-making.

In visualization, we mostly deal with turning data into information that supports gaining knowledge and possibly wisdom.

Data Context vs. Data Content

  • Data Context (Domain/Reference Space/Independent Variables): Frame of reference (e.g., time, space, patient ID) that relates observations.
  • Data Content (Attribute Space/Dependent Variables): Actual measured values (e.g., temperature, CO2 concentration).

Data often modeled via graph topologies:

  • Regular/Irregular grids
  • Networks, Hierarchies

For continuous fields from sparse points, use interpolation (e.g., Inverse Distance Weighting) to estimate intermediate values.

Task Abstractions

  • A Task is what the user wants to do with a visualization.
  • Task lies between:
    • User’s intent (“explore data”) and
    • Actual interactions (“click”, “drag”).
  • Mid-level tasks: like Shneiderman’s mantra (“Overview first, zoom & filter, details on demand”) or simpler tasks like identify, locate, compare.

Types of Tasks

  • Elementary tasks: Identifying a value, locating items by value, comparing two points.
  • Synoptic tasks: Observing patterns, trends, clusters over sets of data, not just single points.
  • Andre & Andrew’s Framework:
    • Lookup & Inverse Lookup
    • Comparison & Inverse Comparison
    • Relation Seeking (given relations, find data points)
    • Synoptic equivalents: Pattern definition, pattern search, etc.

Why tasks matter:
They help ensure expressiveness (matching data types to visual encodings) and effectiveness (visualization supports intended user tasks).

Design Processes for Visualization

Design Activity Framework

  1. Understand: Determine the visualization problem, user needs, data/task abstractions.
    • Outcomes: Visualization requirements.
  2. Ideate: Generate visualization ideas.
    • Use sketching, low-fidelity mockups.
    • Outcomes: A set of possible solution ideas.
  3. Make: Create prototypes (interactive hi-fi prototypes).
    • Outcome: Testable visualization prototypes.
  4. Deploy: Implement final, robust visualization systems.

We focus on the Understand/Ideate/Make steps. Deploy is often a separate engineering challenge.

The Five Design Sheet Method (for Ideation)

  1. Brainstorm Sheet:
    • Generate ~15-20 rough idea sketches.
    • Filter out non-viable ones.
    • Categorize similar ideas.
    • Combine and refine to pick 3 main ideas.
  2. 3 Separate Idea Sheets (2nd-4th sheets):
    • For each chosen idea:
      • Show big picture layout.
      • Highlight key operations, interactions.
      • Show look & feel of crucial elements.
      • Discuss pros & cons.
  3. Realization Sheet (5th sheet):
    • Chosen final design.
    • More detailed plan: algorithms, data handling, milestones, feasibility.

Benefits of 5 Design Sheets:

  • Structured creativity for non-artists.
  • Documented design rationale (proof of work).
  • Facilitates client communication and possible billing for partial work.

Guiding Design Principles (Andy Kirk)

  1. Functional Design: Function before form. Must work correctly first.
  2. Deliberate Design: Every element must have a justified purpose.
  3. Intuitive Design: Should not be more complex than the data demands.
  4. Ethical Design: Do not mislead the viewer.

Evaluation (Munzner’s Nested Model)

  • Layers: Domain situation → Data/task abstraction → Visual encoding → Algorithms.
  • Threats and Validation: Each layer has potential pitfalls; validate accordingly.
    • Algorithms: Test complexity, runtime.
    • Visual encoding: Test for overplotting, clarity.
    • Tasks: User studies, feedback loops.

Benchmarking

  • Standard datasets, pixel-based metrics (e.g., overplot percentage).
  • Graph metrics (edge crossing counts).
  • User Studies: Measure task completion time, error rates, user experience.
  • Field Studies: Demonstrate how new visualizations lead to novel discoveries.

Visualization Critiques (Examples Shown by the Professor)

Example 1: Parallel Sets Visualization

  • Data: People displaced by disasters and conflicts.
  • Chart type: Sankey-like parallel sets.
  • Good aspects: Color choice works in grayscale, corrected shape to reduce perceptual illusions.
  • Critique:
    • The chosen chart type (parallel sets) is meant for showing cross-correlations, but this chart does not show meaningful cross-correlations.
    • The data categories and “other” grouping are inconsistent.
    • Ethical choice of showing absolute numbers vs. normalization is context-dependent; here absolute counts emphasize humanitarian impact.

Suggested Improvement: Use simpler stacked bar charts for clarity if no cross-correlations are needed.

Example 2: Complex Connected Scatterplot for COVID Data

  • Data: COVID deaths over time, multiple countries.
  • Chart complexity confused readers; considered too complicated.
  • Issues:
    • Negative values shown in parentheses (hard to interpret).
    • Non-colorblind-safe color encoding.
    • Overly smoothed data can be misleading.
    • Connected scatterplots are among the most complex chart types; must provide reading guides.
  • Recommended approach: Simpler line charts, small multiples, or clear normalization.

Conclusion

  • We finalized discussion on interaction techniques and introduced tangible and pseudo-haptic interactions.
  • Main focus: Understanding visualization problems via data and task abstractions.
  • Learned about the design activity framework and how to ideate systematically using the Five Design Sheet method.
  • Discussed guiding principles for good visualization design and evaluation strategies.
  • Practiced critique on complex real-world visualizations, emphasizing the importance of clarity, appropriateness of chart types, and ethical considerations.

Next Lecture: Data Pre-processing.
Preparation: Read Chapter on “Preparing Data Tables” from “Making Sense of Data, Part 1.”

Reminders for the Next Session (Wednesday):

  • Bring your data and task abstraction for your project.
  • No need to bring the dataset itself, but know your data’s attributes, types (quantitative, categorical, etc.), and tasks.
  • You will sketch visualization ideas using the Five Design Sheet method.
Data Visualization, Redo
DIKW Hierachy Data Content Data Context Task Abstraction Andrienko
This post is licensed under CC BY 4.0 by the author.