Seeing Graphs - How to successfully visualize connected data
Post on 26-Jan-2015
105 Views
Preview:
DESCRIPTION
Transcript
Joe Parry
Seeing GraphsHow to successfully visualize connected data
Agenda
• The graph design process• Interaction models• Why animate?• Strategies for visualizing larger networks• A reference architecture
GRAPH MODELLING + Data Visualisation
= Graph Visualisation
Why use the graph data model?
Data models are generally interchangeable, but each have their own strengths in certain.
The graph data model is good for…
• Social interactions• Dependency management• Organizational structures• Connectivity of systems• Biological processes• Financial transactions• Communications• http://keylines.com/why
Good in what sense?
Graph models answer questions:
• what depends on what• how do things flow• where are the risks• impact analysis
A quick clarification…
In any data, there is not ‘one graph’. Any single data set can be modelled in many different ways:
Perhaps more suitable for understanding social networks?
Probably a more accurate picture of how email sending works
Graph Modelling
+ DATA VISUALISATION= Graph Visualisation
Data is invisible
To be seen, data must be VISUALLY ENCODED
There are many different options for encoding data
We need to understand which works best for our audience
Fortunately, people have already researched the most accurate encoding methods…
Quantitative Ordinal Nominal
More Accurate
Less Accurate
Quantitative Ordinal Nominal
Position Position Position
Length Density Hue
Angle Saturation Density
Slope Hue Saturation
Area Length Shape
Density Angle Length
Saturation Slope Angle
Hue Area Slope
Shape Shape Area
Visual encoding by data type
Data visualisation is a DESIGN PROCESS
Graph Modelling
+ Data Visualisation
= GRAPH VISUALISATION
There are several design options for graph
visualisation
Matrix
• All nodes have a column & a row• A dot is put in the middle if a link exists• Good for finding clusters• But requires a lot of cross referencing to
understand the details
Edge bundling
• Nodes are around the edge• Links between the nodes are bundled to
show patterns• Detail can become spread out• Computationally difficult to create
Hive plots
Hive Plots
• Nodes are grouped into categories• Links between nodes are added• This can help display huge networks
with a lessened ‘hairball’ effect• But generally Hive Plots are less intuitive
Classic node-link diagram
• Intuitive• Easy to understand• The most ‘successful’ option
for conveying data
INTERACTIONS
Why is interaction needed?
• The user needs to drive their own focus
• Allows the user to see details on demand
• Allows the user to add or remove bits of the graph
Interactions users expect
• Zooming• Multiple node selection• Hover and tooltip events• Context menus• Gestures & tablet support• Expand on demand• Searching• Commands
ANIMATION
Why animate?
• Users expect it• It provides smoother transitions:
– eases context switching for users– Allows them to keep a mental model
• Animations– can show data– can show dynamics
STRATEGIES FOR LARGER NETWORKS
Beyond a certain size, no automatic layout will help you decipher a graph
Your options include:
1. Use color
2. Use filters
3. Allow users to expand
4. Link & Node aggregation
REFERENCE ARCHITECTURE
Graph Database
Neo4j
Web Server
Reference Architecture
Document Store
Search Index
User Authentication(eg Active Directory)
Chart Storehttp://keylines.com/architecture
Cambridge Intelligence
Start-up, founded May 2011We make network visualisation tools
(big data projects, fraud, security services)
top related