static spatial graph features

29
Improving Revisitation in Graphs through Static Spatial Features Sohaib Ghani Purdue University West Lafayette, IN, USA Graphics Interface 2011 May 25-27, 2011 ▪ St. John’s Newfoundland, Canada Niklas Elmqvist Purdue University West Lafayette, IN, USA Presented by Pourang Irani University of Manitoba

Upload: niklas-elmqvist

Post on 22-Apr-2015

602 views

Category:

Technology


0 download

DESCRIPTION

Presentation from GI 2011 for our paper on "Improving Graph Revisitation Using Static Spatial Features".

TRANSCRIPT

Page 1: Static Spatial Graph Features

Improving Revisitation in

Graphs through Static Spatial

Features

Sohaib GhaniPurdue University

West Lafayette, IN, USA

Graphics Interface 2011May 25-27, 2011 ▪  St. John’s Newfoundland, Canada

Niklas Elmqvist

Purdue UniversityWest Lafayette, IN, USA

Presented by

Pourang IraniUniversity of Manitoba

Page 2: Static Spatial Graph Features

+

Basic Idea

Page 3: Static Spatial Graph Features

Overview

• Motivation• Static Spatial Graph Features• User Studies• Results• Summary• Conclusion

Page 4: Static Spatial Graph Features

Memorability & Revisitation

MemorabilityThe memorability of a visual space is a measure of a user’s ability to remember information about the space 

RevisitationRevisitation is the task of remembering where objects in the visual space are located and how they can be reached

Page 5: Static Spatial Graph Features

Motivation• Graphs prevalent in many information tasks– Social network analysis (Facebook, LinkedIn, Myspace)– Road networks and migration patterns– Network topology design

• Graphs often visualized as node-link diagrams• Node-link diagrams have few spatial features– Low memorability– Difficult to remember for revisitation

• Research questions– How to improve graph memorability?– How to improve graph revisitation performance?

Page 6: Static Spatial Graph Features

Example: Social Network Analysis

• Interviewed two social scientists who use graphs for Social Network Analysis (SNA)

• Often experience trouble in orienting themselves in a social network when returning to previously studied network

• At least 50% of all navigation in SNA  in previously visited parts of a graph

Page 7: Static Spatial Graph Features

+

• People remember locations in visual spaces using spatial features and landmarks

• Geographical maps have many spatial features and are easy to remember

• Evaluate whether static spatial features to node-link diagrams help in graph revisitation– Inspired by geographic maps

Idea: Spatial Features in NL Diagrams?

Page 8: Static Spatial Graph Features

Design Space:Static Spatial Graph Features

• Three different techniques of adding static spatial features to graphs– Substrate Encoding (SE)– Node Encoding (NE)– Virtual Landmarks (LM)

• But which technique is optimal?

Page 9: Static Spatial Graph Features

Substrate Encoding• Idea: Add visual features to substrate (canvas)• Partitioning of the space into regions– Space-driven: split into regions of equal size– Detail-driven: split into regions with equal numbers of items

• Encoding identity into each region– Color– Textures

Figure 1 Figure 2

Page 10: Static Spatial Graph Features

Node Encoding

• Idea: Encode spatial position into the nodes (and potentially the edges) of a graph

• Available graphical variables:– Node Size– Node Shape– Node Color

Page 11: Static Spatial Graph Features

Virtual Landmarks

• Idea: Add visual landmarks as static reference points that can be used for orientation

• Landmarks– Discrete objects– Evenly distributed invisual space

Page 12: Static Spatial Graph Features

User Studies 

• Experimental Platform– Node-link graph viewer in Java– Overview and detail windows

• Participants: 16 paid participants per study

• Task: Graph revisitation–Phase I: Learning–Phase II: Revisitation

Page 13: Static Spatial Graph Features

Phase I: Learning• N blinking nodes  shown in sequence, Participants visit and learn their positions.

Page 14: Static Spatial Graph Features

Phase I: Learning (cont’d)

Page 15: Static Spatial Graph Features

Phase II: Revisitation• Participants revisit the nodes whose location they had learned, in the same order

Page 16: Static Spatial Graph Features

Phase II: Revisitation

Page 17: Static Spatial Graph Features

Study 1: Substrate Encoding

• Study Design:– Partitioning: Grid and Voronoi Diagram.– Identity Encoding: Color and Texture– Layout: Uniform and Clustered

• Hypotheses:– Voronoi diagram will be faster and more accurate than grid for spatial partitioning

– Texture will be more accurate than color for identity encoding

Page 18: Static Spatial Graph Features
Page 19: Static Spatial Graph Features

Study 1: Results

Page 20: Static Spatial Graph Features

Study 2: Node Encoding

• Study Design:– 3 Node Encoding techniques: Size, Color and Size+Color 

• Hypothesis:– Size and color combined will be the best node encoding technique in terms of both time and accuracy

Page 21: Static Spatial Graph Features
Page 22: Static Spatial Graph Features

Study 2: Results

Page 23: Static Spatial Graph Features

Study 3: Combinations

• Best techniques from Study 1 (Grid with Color) and Study 2 (Size+Color) as well as virtual landmarks

• Study Design:– Eight different techniques: SE, NE, LM,SE+NE, SE+LM, NE+LM, SE+NE+LM, and simple graph (SG)

• Hypotheses:– Techniques utilizing substrate encoding will be faster and more accurate than node encoding and landmarks

– The combination of all three spatial graph feature techniques will be fastest and most accurate

Page 24: Static Spatial Graph Features
Page 25: Static Spatial Graph Features

Study 3: Results

Page 26: Static Spatial Graph Features

Study 3: Results (cont’d)

• Techniques with substrate encoding significantly faster and not less accurate.

• SE+NE+LM not significantly faster and more accurate than all other techniques

• Virtual landmarks promising strategy, performing second only to substrate encoding

Page 27: Static Spatial Graph Features

Summary

• Substrate encoding (SE) is dominant strategy– Space-driven partitioning– Solid color encoding

• Virtual landmarks (LM) help significantly• Node encoding (NE) not as good other two• Combination of virtual landmarks (LM) and substrate encoding (SE) is optimal

Page 28: Static Spatial Graph Features

Conclusion

• Explored design space of adding static spatial features to graphs

• Performed three user studies– Study 1: grid with color is optimal substrate encoding

– Study 2: node size and color is optimal node encoding

– Study 3: substrate encoding, landmarks, and their combination are optimal techniques

Page 29: Static Spatial Graph Features

Thank You!

Contact Information:Sohaib GhaniSchool of Electrical & Computer EngineeringPurdue UniversityE-mail: [email protected]

http://engineering.purdue.edu/pivot/