III. Task abstraction (=WHY), visualization design and validation

1. Task Abstraction (= WHY)

Transforming task description from domain-specific language into abstract form allows to reason about similarities and differences between them for easier analysis, comparision and the application of visualization tools.

Actions = Thing to do

<aside> 🏁

Three levels of actions: analyze, search, and query.

</aside>

Analyse = the highest-level actions are to use vis to consume or produce information
1. Consume tasks are related to
  - Discover = Exploring data to have ideas, generate hypothesis
  - Presentation = Communicate results, teach
  - Enjoy = many vis application are used by general public for pure enjoyment
2. Produce of data refers to
  - Derived data = decide what the right thing to show
    
    Typically in sci vis = try to derive and visualize low dimensional parameters characterizing the event
  - Annotations = key tools to transform semantics to raw data
  - Recordings = saves or captures visualization elements as persistent artifacts
Search = at the middle level, search can be classified according to the identify and location of targets are know or not
1. Lookup (tra cứu)
  - word in dictionary
2. Locate (tìm kiếm)
  - keys in your house
  - node in the network
3. Browse (duyệt)
  - books in bookstore
4. Explore (khám phá)
  1. cool neigborhood in new city
Query = at the low level, queries can have three scopes: identify one target, compare some targets and summarize all targets

Targets = Selected data

Targets meaning some aspect of the data interesting for the user

Three high-level targets are very broadly relevent for all kinds of data
1. Trends: is a high-level characterization of a pattern in the data. Ex: increase, decrease, peaks, troughs and plateaus
2. Outliers: some data doesn’t fit well with that backdrop (bối cảnh)
3. Features: is task dependent, meanng any particular structures of interest
The lowest-level target for an attribute is
1. To find an individual value
  - Find the extremes: the minimum or maximum value across the range
  - High level scope is the distribution of all values for an attributes
2. Some targets encompass the scope of multiple attributes: dependencies, correlations, and similarities between attributes.
Some targets pertain to specific types of datasets.
1. Network data specifies relationships between nodes as links. The fundamental target is to understand the structure of these interconnections. That’s network’s topology
2. For spatial data, understanding and comparing the geometric shape is the common target

2. Vis Idioms (= HOW)

A vis idioms is a distinct approach to creating and manipulating visual representations

Two aspects

Visual encoding idiom: how to draw
Interactive idiom: how to manipulate

Relevant challenges

Size: how to draw images for large datasets?
Attributes size
Spatial/nonspatial data
Heterogeneous data encoding
Users
Evaluation

3. Design Validation

Validating the effectiveness of a visualization design (vis) is challenging because there are many possible questions to consider. The process is complicated due to the multiple levels of decision-making involved in creating a visualization tool. Each of the four levels has a different set of threats to validity

Wrong problem: The designer misunderstood the target users' needs.
Wrong abstraction: The visualization displays the wrong data or does not communicate the intended message.
Wrong idiom: The way the data is shown (visual encoding and interaction methods) doesn’t work effectively.
Wrong algorithm: The algorithm behind the visualization is inefficient or too slow.

There are two main types of validation approaches:

Immediate validation: This occurs directly after making a design decision and addresses potential issues in real-time.
Downstream validation: This comes after implementing and testing the design and addresses deeper issues, including testing how the design holds up in real-world usage.

Domain Validation (Wrong Problem)

At the domain validation level, the key issue is whether the target audience's problems have been accurately identified.

Immediate validation: A common approach for this case is a field study, where the investigator observes how people act in real-world settings. This helps ensure that the problem has been correctly identified before proceeding with tool development.
Downstream validation: Adoption rates by the target users provide a signal of success. If the tool is widely adopted by its intended users, it indicates that the tool addresses their needs effectively.

Abstraction Validation (Wrong Task/Data Abstration)

At the abstraction level, the threat is whether the abstracted tasks and data are appropriate for the target users.

Immediate validation: the system must be tested by target users doing their own work, rather than doing an abstract task specified by the designers of the vis system.
Downstream validation: is to have a member of the target user community try the tool, in hopes of collecting evidence that the tool is in fact useful.

Idiom Validation (Ineffective encoding/interaction idiom)

At the idiom level, the threat is whether the visual encoding (such as color, shape, layout) and interaction techniques effectively convey the desired abstraction.

Immediate validation: involves expert reviews or heuristic evaluations to ensure that the design complies with known best practices.
Downstream validation: This can involve lab studies or usability testing to measure user performance and gather feedback on the idiom’s effectiveness

Algorithm Validation (Slow algorithm)

At the algorithm level, the threat is whether the algorithm is efficient and correct.

Immediate validation: includes analyzing computational complexity to assess whether the algorithm will perform well under expected use cases.
Downstream validation: Measuring the wall-clock time and memory performance of the implemented algorithm provides concrete evidence of the algorithm’s efficiency.