Set vs Graph-based visualisations of multiple trees
1. A Comparison of Set-based and Graph-based
Visualisations of Overlapping Classification
Hierarchies
Martin Graham, Jessie Kennedy &
Chris Hand
Napier University, Edinburgh
2. Overview
Introduction
Problem Domain
Similar Approaches
Prototype 1 - Graph
Prototype 2 - Set (+ example)
Testing
Conclusion
3. Introduction
Hierarchical information structures are a
major topic within Information
Visualisation (IV)
Cone Trees - Robertson, Mackinlay & Card
Tree Maps - Johnson & Shneiderman
Multiple hierarchies relatively unexplored
Relations (intra- and inter- tree)
5. Tasks for Multiple
Taxonomies
Taxonomist Requirements
Track a particular genus’s siblings and
parents
Track a higher level (internal) node’s children
Compare depths
Compare global structures
6. Previous techniques
Animation
Strong pre-attentive cues
Suited for gradual, not abrupt, changes
Shows change between only 2 states at any
one point
Small Multiples (cf Tufte)
Display all structures
‘N’ small multiples = 1/’N’th screen area each
7. Previous Techniques
Wittenburg et al’s Group Asynchronous
Browsing on the WWW
Combines multiple TreeMaps on top of a
MultiTree structure
Shows correlations (cross-references) using
colour
Static colouring only
Can show top-level correlations but not for
individual parts of hierarchies
8. Graph-based prototype
Combine multiple hierarchies into one
graph structure
Visualised using spring-mass metaphor
Different coloured links & glyphs indicate
relationships in different hierarchies
Interaction via panning/zooming and
filtering of a particular node’s relations
12. Set-based prototype
Sets reflect mental model of taxonomic
reasoning
Differentiate between categories and the
objects we are organising
Use colour via linking to indicate
correlations
Linking applied over a set of hierarchies
13.
14. Set-based Prototype
Advantage of
Reflects taxonomist’s mental model
This is at the price of
Lose a level of detail (unnamed leaves and
only one relation set shown at a time)
Have to accept restrictions of small multiples
However...
Clearer visualisation
15.
16.
17. Prototype Effectiveness
Metrics v Empirical Testing
Brath’s metrics
Suitable for static visualisations
Interactive visualisations cause problems
Zooming and filtering reduce on-screen data
density (but make for a more relevant picture)
18. Prototype Effectiveness
Informal User Testing (cf Nielsen)
A handful of typical would-be users
Used early in development
Task-based observations generate solutions
(Monk & Wright)
19. Prototype Effectiveness
Observations
Overlapping (occluding) nodes caused
confusion in graph prototype
Non-leaf (category) nodes in set based
prototype indicated by highlighting their
descendent leaf nodes, confusing users as to
extent of the non-leaf node
20. Prototype Effectiveness
Solutions
Ensure that highlighted nodes always
displayed on top of unselected nodes
Allow user to fine-tune positioning.
Non-leaf nodes and leaf nodes to be
highlighted in similar fashion
21. Prototype Effectiveness
Outcome
Definite preference for set-based visualisation
prototype
Graph-based visualisation prototype too
cluttered, even with zooming/filtering
mechanisms
More tasks envisaged
22.
23. Conclusions
Constructed 2 prototypes for visualising
multiple hierarchies and their correlations
Users stated both prototypes enabled
them to see information that is not
possible with their current system
Future work
Test prototypes on other information sets