Task Abstraction in Visualization: Actions and Targets

Posted on Mar 2, 2025 in Business Administration and Innovation Management

Task Abstraction: Why?

Chapter 3

Actions

  • Analyze:
    • Consume: Analyzing existing information without creating new data.
      • Discover: Uncover new insights or analyze information that is not readily apparent.
      • Present: Share knowledge with others.
      • Enjoy: Explore data for personal satisfaction.
    • Produce: Generating new material.
      • Annotate: Add extra information to highlight important data points.
      • Record: Capture and save data.
      • Derive: Create new data based on existing data.

A graph of trade balance and export Description automatically generated

  • Query: How to examine the target
    • Identify: Refer to a single target.
    • Compare: Refer to multiple targets.
    • Summarize: Refer to the full set of possible targets.

Targets

  • All data
    • Trend: Identify increases, decreases, and peaks.
    • Outliers: Discover data points that deviate significantly from the norm.
    • Features: Discover all elements not classified as trends or outliers.
  • Attributes
    • One:
      • Distribution: Analyze the spread of data.
      • Extremes: Identify maximum and minimum values.
    • Many:
      • Dependency: Explore relationships between attributes.
      • Correlation: Measure the strength and direction of relationships.
      • Similarity: Identify attributes with similar patterns.
  • Network data
    • Topology: Analyze the structure and connections within the network.
      • Path: Find the shortest or most efficient route between nodes.
  • Spatial data
    • Shape: Analyze the form and outline of spatial objects.

Task Abstractions: Actions

First, we need to analyze tasks abstractly. Transforming task descriptions from domain-specific language into abstract forms allows you to reason about similarities and differences between them. When analyzing, we need to consider who has a goal or makes a design choice: the designer of the visualization or the end-user.

To analyze why a visualization tool is being used, we need to break down the reasons into actions and targets.

Three levels of actions:

  • Highest-level: Analyze, consume, or produce information.
    • Consume: Present, discover, and enjoy. Discovery may involve generating or verifying a hypothesis.
    • Produce: Annotate, record, and derive.
  • Middle-level: Search. Search can be classified according to whether the identity and location of targets are known or not:
    • Lookup: Both identity and location are known.
    • Locate: The target is known, but its location is not.
    • Browse: The location is known, but the target is not.
    • Explore: Neither the target nor the location is known.
  • Low-level: Query. Queries can have three scopes:
    • Identify one target.
    • Compare some targets.
    • Summarize all targets.

Task Abstractions: Targets

These are the parts of the data that the targets find relevant.

  • All Data: Generally speaking, a user’s concerns may include identifying patterns, trends, and outliers.
  • Attributes: An individual may be interested in studying the following:
    • Single attribute (to simulate its distribution)
    • Multiple attributes (to simulate their relationships).
  • Spatial Data: An individual may be interested in learning about objects’ shapes.
  • Network Data: A user would be curious to learn about its topology, or connection patterns.

Abstractions

  • Rule of thumb: Eliminate any domain jargon in a systematic way.
  • Interplay: Data abstraction must be used within task abstraction.
  • However, task abstraction can cause you to repeatedly change the data.

Means and Ends

  • Ends, or goals, are what we aim to do, and means are what we use to achieve those goals.
  • {action, target} pairings: Find outliers, compare trends, learn about distribution, browse topology.

Why is Validation Difficult?

Chapter 4

There are different ways to get it wrong at each level:

x+ikIFmweyK5AAAAABJRU5ErkJggg==

We need to pick methods from different fields at each level. The solutions for each level are:

  • Algorithm: Doing computational benchmarks to measure system time and memory. Analyze computational complexity (technique-driven work).
  • Visual encoding (design level): Justify the design with respect to alternatives, then analyze results qualitatively (cognitive psychology) and conduct lab studies.
  • Data/task abstraction: Observing people after deployment (field study); lab studies are not sufficient.
  • Domain situation: Observing people using existing tools.

Notes:

  • Lab studies do not confirm task abstraction.
  • Computational benchmarks do not confirm idiom design.