Brushing, Linking & Interactive Querying

1
Brushing, Linking Interactive Querying

• Information Visualization
• February 15, 2002
• Sarah Waterson

2
Interaction

• Interaction involves the transformations that
  map the data to visual form.
• More than just the controls? Integrate controls
  into the visualization.
• Allow for direct manipulation of the graphical
  representation of the data.

3
Exploratory Data Analysis

• Beyond the small multiples - the next generation
  of Exploratory Data Analysis!
• Detective work spot trends, patterns, errors,
  features in the data.
• Unless exploratory data analysis uncovers
  indications, usually quantitative ones, there is
  likely to be nothing for confirmatory data
  analysis to consider.

4
Time

• Response times of computer must be tuned to human
  response times
• Psychological Moment (0.1 sec.)Fusion into
  single precept motion, animation, cause effect
• Unprepared Response (1 sec.)dialogue, driving,
  updating user
• Unit Task (10 sec.)elementary interaction
  cycles, pace of routine cognitive skills

5
Overview of Papers

• High Interaction GraphicsStephen G. Eick
  Graham J. Wills, ATT Bell Labs 1994
• Dynamic Queries for Visual Information
  SeekingBen Shneiderman, U. of Maryland 1994
• Visual Information Seeking Tight Coupling of
  Dynamic Query Filters with Starfield
  DisplaysChristopher Ahlberg Ben Shneiderman,
  U. of Maryland 1994
• Data Visualization SlidersStephen G. Eick,
  ATT Bell Labs 1994
• Interactive Data Analysis The Control
  ProjectJoseph Hellerstein Co., U.C. Berkeley
  IBM Almaden 1999
• Enhanced Dynamic Queries via Movable
  FiltersKen Fishkin Maureen C. Stone, Xerox
  PARC 1995

6
High Interaction Graphics

• ClarityInformation only on demand, cleaner
  more focused displays, allow a range of options
• RobustnessAvoid drawing inferences from only one
  view
• PowerCombine views, leverage exploration
• Possibility3 dimensional data, animation

7
Principles

• Simple, easy to interpret views
• Information hiding, details on demand
• Direct Manipulation

8
Linking Brushing

• LinkingVisually indicating which parts of one
  data display correspond to that of another
• BrushingAllowing the user to move a region
  (brush) around the data display to highlight
  groups of data points. Generally used on scatter
  plots.
• Usability issues selection, de-selection,
  setting values, appropriate widgets

9
Examples
Linking altitude to grass and grain types in
Scottish Districts

• Districts of the city of Dublin showing areas
  with high levels of average income

10
Another Example

• Point Visualization Tool (PVT) of data related by
  postal codes

11
Application Domains

• Spatial Data VisualizationIn general, there are
  more assumptions made about spatial data than
  about non-spatial data and thus more diagnostic
  plots are required.
• Software VisualizationVery difficult problem
  with many dimensions and possible visualizations
  the code, data structures, communication,
  execution threads, debugging, memory management,
  etc.

SeeSoft
12
Comments

• Great introduction of purpose, general
  techniques.
• Some mention of usability, though more would be
  appreciated.
• Examples were somewhat simple, despite mentioning
  complex application domains.
• Easy to read. Seems like the beginnings of a book
  or survey paper.

13
Dynamic Queries

• Selecting value ranges of variables via controls
  with real time feedback in the display

• Principles
• Visual presentation of querys components
• Visual presentation of results
• Rapid, incremental, and reversible control

• Selection by pointing, not typing
• Immediate and continuous feedback
• Support browsing
• Details on demand

14
Examples

• Periodic Table of the ElementsAdjust properties
  with sliders on the bottom to highlight matching
  elements.

15
More Examples
Unix Directory Exploration

• DynaMapCervical cancer rates from 1950-1970 -
  modify year, state, demographics

16
Even More Examples
17
Yet More Examples
Devise
Information Visualization and Exploration
Environment (IVEE) Job to Skills matching
18
Coupling Starfield Displays

• Tight coupling
• Query components are interrelated in ways that
  preserve display invariants, reveal state of
  system
• Output of queries can be easily used as input to
  produce other queries. Eliminate distinction
  between commands/queries/input and
  results/tables/output
• Starfields
• For data without natural mapping
• Glorified scatter plots?

19
Home Finder Map
20
Home Finder Text
21
Film Finder
22
(No Transcript)
23
Pros Cons

• Quick, easy, safe, playful
• Good for novices experts
• Excellent for exploration of very large data sets

• Database management systems cant handle the
  queries
• Slow hardware
• Application specific programming
• Simple queries only
• So many controls

24
Research Directions

• Widgets for multiple ranges
• Boolean combinations for sliders
• Zooming
• Selecting controls from large sets of attributes
• Grand tours of the data
• New interaction devices

25
Comments

• Good paper for overview, purpose and research
  directions for dynamic queries.
• Particularly for research directions.
• Compelling examples for need.
• Usability study showed dynamic queries faster
  than Symantec's QA, though other measures might
  be more important than speed.
• Well written.
• Big impact contribution to the field.

26
Data Visualization Sliders

• Use the sliders themselves as data displays
• Painting metaphor for specifying disconnected
  intervals

27
The Control ProjectContinuous Output and
Navigation Technology with Refinement Online

• Of all mens miseries, the bitterest is this to
  know so much and have control over nothing.
  Herodotus
• Full scale data analysis will always be slow.
• Goal Build a system that iteratively refines
  answers to queries and give users online control
  of processing.
• Aggregation, Enumeration, Visualization, Mining

28
The Crystal Ball

• Anytime Algorithms produce a meaningful
  approximate result at any time during their
  execution
• Trade quality and accuracy for interactive
  response times
• Continuously fetch new data at random users
  prefer a to see a representative sample of the
  data at any time
• Preferential re-ordering
• Ripple joins

29
Online Aggregation
30
Online Enumeration UI

• Database analysts vs. Domain experts
• Eyeballing in Databases and lists
• Using fuzzy techniques, such as the scrollbar

31
Online Data Visualization

• CloudsRender records as they are fetched but
  also generate overlay of shaded regions
  estimating missing data. Cloud color chosen to
  minimize expected error.

32
Comments

• Great work. Really cool. Big impact.
• Very necessary technology, intelligent solution,
  and very compelling.
• More analysis of the visualization would be nice
  and perhaps more on usability (Katie Everitt and
  Ka-Ping Yee)
• Overall, quite impressive.

33
Movable Filters

• Movable Magic LensTM filters over starfield
  displays for multiple simultaneous visual
  transformations and queries
• Enhanced brushing with sliders?

34
Queries Filters

• Boolean Composition

Real-valued Queries
Semantic Filters
Missing Values
35
Comments

• Interesting idea, but I would like to see it in
  action
• The UI looks a bit horrid and no usability
  studies
• Only seems appropriate for scatter plots, and
  selection is limited by shape
• Good that it can do some more complex queries,
  but are they understandable?
• Where else could one use these lenses?

36
Thoughts

• More than MiceInteraction techniques beyond
  point and click
• Understanding the DataUnderstanding the data and
  model How to create the interface appropriate
  for investigation.