4. Task Abstraction

---

Why abstract Task?

domain-specific 한 방법보다는 abstract 하게 task를 고려해야한다.

→ 그렇지 않으면, 도메인마다 너무 task가 달라짐.

Actions

• High-level: Analyze
• Mid-level: Search
• Low-level: Query

hierarchical하게 나눠짐. 각 수준에 따른 선택은 독립적이며, 모든 action은 각 3개의 수준에서 설명가능하다.

---

High-level: Analyze

Consume

(1) Discover

• Find new knowledge that is not previously known
  • generate new hypothesis, or verify existing hypothesis.
• For Scientific inquiry

(2) Present (=explain)

• The communication of information that is already understood (보고용)
• e.g. Infographic
• Output of discover → input of present

(3) Enjoy

• Motivated by user enjoyment
• Casual

Produce

(1) Annotate (=tag)

• attaches temporary info

(2) Record

• save or capture as persistent

(3) Derive (=transform)

• To produce new data elements based on existing elements
• Changing type of data (aggregation)
• transform with additional info
• using arithmetic/statistic/logical operations

→ Do not draw what you are given

---

Mid-level (Search)

high level을 수행하려면 먼저 search를 수행해야함

	Target Known	Target Unknown
Location Known	Lookup	Browse
Location Unknown	Locate	Explore

---

Low-level (Query)

target을 search를 통해 찾았다면, QUERY해야힘.

ex. 선거 결과

(1) Identify

• 서울 관악구의 선거 결과 확인

(2) Compare

• 서울 관악구와 동작구의 결과 비교

(3) Summarize / Overview

• 전체적인 투표 결과 분포 확인

---

Targets

All-data level

(1) Trends (=patterns)

• high-level pattern
• increase, decrease…

(2) Outliers

(3) Features

• Particular structures of interest
• Task-dependent definition
  • e.g. clusters

Attribute Level

(1) One attribute

• Individual values
• extremes

(2) Multiple attributes

• Dependency (인과관계)
  • association
• Correlation (상관관계)
• Similarity

For specific dataset

• Network data
  • paths
• Spatial data
  • shape

---

Practice Analysis!