amy vanderbilt ~ vin taylor ~ martin taylor ~ mark nixon ~ jan terje bjorke
DESCRIPTION
Information-Theoretic Considerations of Graph/Network Topology. Amy Vanderbilt ~ Vin Taylor ~ Martin Taylor ~ Mark Nixon ~ Jan Terje Bjorke Sven Brueckner ~ Zack Jacobson ~ Jason Moore ~ Rusty Bobrow. Amy K. C. S. Vanderbilt, Ph.D. TITLE. - PowerPoint PPT PresentationTRANSCRIPT
Visualisation Network-of-ExpertsMalvern, UK
NOV 4-6th2008
Amy Vanderbilt ~ Vin Taylor ~ Martin Taylor ~ Mark Nixon ~ Jan Terje Bjorke Sven Brueckner ~ Zack Jacobson ~ Jason Moore ~ Rusty Bobrow
Information-Theoretic Considerations of Graph/Network Topology
Amy K. C. S. Vanderbilt, Ph.D. TITLE
(USA) 571-723-5645 [email protected]
Information-Theoretic Considerations of Graph/Network Topology
2
OutlineOutline
Random Thoughts
Cognitive Model Revision
Effect of Hypernodes
Quantification of Network Visualizations
How Does Information Theory Support Visualization? – Random Philosophical Thoughts
Cognitive Model Revision
High Level Process
A First Step At Quantification
Effect Of Hypernodes
Towards Quantification Of Network Visualizations
Measuring The Information Content In An Image
Measuring The Information Content In A Visualization
User Interaction And Optimization Using Information Theory
Tuning The Sources
Amy K. C. S. Vanderbilt, Ph.D. TITLE
(USA) 571-723-5645 [email protected]
HOW DOES INFORMATION THEORY SUPPORT VISUALIZATION?
3
Random Philosophical thoughts
Use the visualization to convey syntax and semantics and let the user key off of their experience / world view (pragmatics)
Syntax ~ Semantics ~ Pragmatics : these three combine to yield a coherent understanding leading to accurate analysis
The function of visual capacity is the summation of the history plus what data you are presented and the manner in which it was presented
The measure of the information conveyed is how much you have reduced uncertainty (information entropy) from the cognitive model
Outline
Random Thoughts
Cognitive Model Revision
Effect of Hypernodes
Quantification of Network Visualizations
Amy K. C. S. Vanderbilt, Ph.D. TITLE
(USA) 571-723-5645 [email protected]
HOW DOES INFORMATION THEORY SUPPORT VISUALIZATION?
4
Random Philosophical thoughts
Information capacity is the difference between what you know and what is displayed
Can we use information theory to build visualizations tailored to what the user knows? i.e. their pragmatic history?
We have no control over what they user may get out of the visualization that is not there or that is beyond what is there
WORLD = the network the user is trying to understand + embedding fields [i.e. the user’s accumulated context/pragmatics]
Perception is an active process consisting of interaction with the environment
Outline
Random Thoughts
Cognitive Model Revision
Effect of Hypernodes
Quantification of Network Visualizations
Pragmatic Context UserSyntax & Semantics Data System
Happens in the
User’s Mind
Amy K. C. S. Vanderbilt, Ph.D. TITLE
(USA) 571-723-5645 [email protected]
Cognitive Model Revision
5VISUALIZATION IS USER-CENTRIC
High Level Process
A simple look at the process of understanding the real world using network visualizations
World
Network
Display
Visualization
Understanding
Revises User Model and Action
Utility
Outline
Random Thoughts
Cognitive Model Revision
Effect of Hypernodes
Quantification of Network Visualizations
Amy K. C. S. Vanderbilt, Ph.D. TITLE
(USA) 571-723-5645 [email protected]
Cognitive Model Revision
6REALITY DISPLAY COGNITIVE MODEL
A First Step At QuantificationCan we quantify the continual process of the human world model
converging to reality via understanding gained from visualizations?
OldCognitive Model
RealityExploratory mode:1. The viz presents some bits of
reality to the user…some correctly, some not and some inadvertently via user induction
2. The user has a set of bits that represents their belief (their model of reality)
3. The impact of the viz is the replacement/modification of some of these bits
4. Based on the revised model, the user revises the utility of the set of available actions…hopefully this optimizes
Display
New Cognitive
Model
Visualization
Mental Processes
(Perception, etc)
Outline
Random Thoughts
Cognitive Model Revision
Effect of Hypernodes
Quantification of Network Visualizations
Amy K. C. S. Vanderbilt, Ph.D. TITLE
(USA) 571-723-5645 [email protected]
Effect of Hypernodes
7
Can The User Extract Information From A Hypernode?
Not necessarily as such when the links between hypernodes are determined by the component links of the sub-nodes
BUT – we might try grouping entities into hypernodes by various measures and THEN allowing links and structure to emerge between those hypernodes
Links among independent hypernode layers indicate pragmatically identical entities
Outline
Random Thoughts
Cognitive Model Revision
Effect of Hypernodes
Quantification of Network Visualizations
Amy K. C. S. Vanderbilt, Ph.D. TITLE
(USA) 571-723-5645 [email protected]
Towards Quantification of Network Visualizations
8
NAIVE ENTROPY IS BLIND TO THE USER
Measuring The Information Content In An Image In the image processing world, task-based experiments wherein
analysis are asked to perform a detection, decision or characterization task using an image [e.g. a tank in a field, etc]
In these experiments, information content in the image is measured by the entropy in the image
This entropy is a pixel based measure
Pixel based measures are too simplistic for measuring the information content in a visualization because they ignore the user’s perception and pragmatics
However, these methods can be tailored to measure the information content in a visualization
Outline
Random Thoughts
Cognitive Model Revision
Effect of Hypernodes
Quantification of Network Visualizations
Amy K. C. S. Vanderbilt, Ph.D. TITLE
(USA) 571-723-5645 [email protected]
Towards Quantification of Network Visualizations
9
IC[V] U f(N,L,l) and/or f( U(N),U(L),U(l))
Measuring The Information Content In A VisualizationInformation_Content(Visualization)
- Entropy_Aggregation(Nodes,Links,Labels,…) A visualization at any one point in time is an image used by the analyst
to perform a task
We can calculate the entropy of the visualization, taking into account pragmatic weightings on nodes based on various factors
Node/link based measures instead of pixel based measures
Weight nodes/links based on relevancy to the query or other pragmatic measures
Calculate the entropy of the visualization image at that point in time
Let the USER dial up and down the total entropy [aka information content] of the image to their own optimal level for that query at that moment.
Outline
Random Thoughts
Cognitive Model Revision
Effect of Hypernodes
Quantification of Network Visualizations
Amy K. C. S. Vanderbilt, Ph.D. TITLE
(USA) 571-723-5645 [email protected]
Towards Quantification of Network Visualizations
10
CONTINUOUS INTERACTIVE VISUALIZATION TUNING
User Interaction and Optimization With Information Theory
USER-CENTRIC OPTIMIZATION Since visualization is a personal experience, let the user tune their
visualization in a continuous, interactive way:
Increase/decrease the relevancy/attention given to certain types of nodes or links
Dial up and down the total entropy [aka information content] of the image to their own optimal level for that query at that moment.
The software will need to iterate an optimization program to:
Predict the entropy in a given layout of the network Reduce or increase entropy accordingly Create the layout Measure again Reduce or increase as necessary and so on All of this on the fly as the user is tuning their preferences
Outline
Random Thoughts
Cognitive Model Revision
Effect of Hypernodes
Quantification of Network Visualizations
Sources
Amy K. C. S. Vanderbilt, Ph.D. TITLE
(USA) 571-723-5645 [email protected]
Towards Quantification of Network Visualizations
11
ONE EXAMPLE
Tuning The SourcesSuppose an analyst has a network visualization at hand and is searching a corpus of documents or other sources to extract additional network
information
Each document or source will return a small sub-network
Compute the entropy difference between the existing network visualization and each source’s contribution.
Allow the analyst to dial up and down the number and types of sources to be merged into the visualization
Outline
Random Thoughts
Cognitive Model Revision
Effect of Hypernodes
Quantification of Network Visualizations
Amy K. C. S. Vanderbilt, Ph.D. TITLE
(USA) 571-723-5645 [email protected]
Towards Quantification of Network Visualizations
12
…SHANNON’S LAST THEOREM?
Conclusions
IDEAL OPTIMIZATION: minimize entropy and maximize utility
The user holds the definition and measure of utility within their mind and thus must contribute this measure via interaction with the system
Information theoretic optimization of visualization requires forms of user modeling/interaction
Outline
Random Thoughts
Cognitive Model Revision
Effect of Hypernodes
Quantification of Network Visualizations