Browsing Hierarchical Data with Multilevel Dynamic Queries and Pruning

1
Browsing Hierarchical Data withMulti-level
Dynamic Queries and Pruning

• By H. P. Kumar, C. Plaisant B. Schneidernam
• Presented by Dawit Yimam Seid

2
Tree Representations

• Node-link visualization
• 2D (outline, tree diagrams)
• 3D (Cone Trees)
• Tree-map visualization (nested non-nested)
• Graph structures (hypertext, flow charts, etc.)

3
HDS/Tree Browsing

• Hierarchical browsing
• interaction matches structure
• When do we do hierarchical browsing ?
• hierarchical decision making, iterative
  refinement/progressive querying of datasets
• e.g. sales budget data, catalog of products,
  computer file systems
• most AND queries can be considered hierarchical
• e.g. SEXmale AND M_STATUSsingle AND
  INCOMEgt200,000

4
Tree Browsing - requirements

• Somewhat contradictory
• Retain context - not getting lost deep in the
  tree
• vs.
• Dont display non-relevant information
• How to achieve both at the same time ?

5
PDQ Tree-Browsers Approach

• Enable viewing tree at different levels
• Selection (filtering) criteria at various levels
  in the tree
• attribute based specification of the foci of
  interest
• Hide uninteresting nodes and branches rapidly

6
PDQ Tree-Browser Techniques

• Dynamic Queries tight coupling
• let users dynamically query nodes at each level
• tightly couple query widgets to tree
  visualization
• tightly couple the overview and detail views
• Automatic Pruning
• (in addition to zooming and panning)
• automatically prune out uninteresting
  nodes/sub-trees to get more compact views

7
Dynamic Queries for Tree Browsing

• Applying DQ to trees is challenging !
• Predefined relationships between data points
  (unlike the case of flat/independent data sets).
  A query at one node affects other nodes
• Dynamic generation of widgets since
• nodes at different levels can have different
  attributes, and
• since these nodes are dynamic

8
The PDQ Tree-Browser Interface

• Data display - two resizable windows
• Overview (left)
• Detail view (right)
• Dynamic Query Panel
• Attribute list (shows attributes of the current
  lowest level)
• Widgets panel - widgets created by users
• range-sliders (for numeric)
• menu (for text)

9
The PDQ Tree-Browser Interface
Detailed view
Overview
Field-of-view
Level name button
Level increment/ decrement
Widgets Panel
Attributes List
Feedback Indicator
10
Query Manipulation

• Query using Widgets
• Queries AND-ed together and matching nodes
  colored yellow in real time
• Manipulating a widget at an internal node level
  makes that node the current lowest level
• Tree structure changed through
• level buttons
• widgets
• or - buttons
• pan detail view dragging the field-of-view
• Two types of coupling !

11
(No Transcript)
12
Pruning

• Done when depth of the tree is increased.
• Children of non-matching nodes are pruned
• Orange shows non-pruned nodes

13
(No Transcript)
14
(No Transcript)
15
Example Application

• The UniversityFinder

16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
PDQ Tree-Browser Limitations

• Max depth of 5
• only up to 3 attributes to query queries only
  AND-ed
• scrolling the overview needed when tree is large
• Clicking nodes in detail view do not open their
  sub-trees

20
Issues/Alternatives

• Pruning Vs. graying out ?
• To change tree structure as widgets are created
  or not ?
• How to handle large trees ?
• Maintain overview to detail view coupling ?
• Intermediate views, let user restrict before
  display
• How to generalize the PDQ Tree-Browser to other
  tree representations (e.g. treemap)
• where to place widgets ? How to prune ?
• How to add OR and NOT queries ?

21
Usability Testing

• 8 subjects (2 pilot, 6 test)
• 7 tasks (feature based task-based)
• Results (on a scale of 1-9)
• 8.7 - pruning (Hide Grey Leaves) is useful
• 8.2 - dynamic query is useful
• 4.8 - easiness of range-slider manipulation
• Some users did not like the dynamic widget
  creation !

22
Controlled Experiment

• Goal - compare Full-tree, Partially-pruned tree
  and Fully-pruned tree
• 25 subjects 7 tasks
• Results
• Task completion time
• 1. Fully-pruned 2. Partially-pruned 3. Full-tree
• Subjective satisfaction rate
• 1. Partially-pruned 2. Fully-pruned 3.
  Full-tree
• No clear cut finding on graying Vs. hiding pruned
  nodes