Interactive Navigation of Large Graphs and Networks Tamara

Interactive Navigation of Large Graphs and Networks Tamara Munzner Stanford University Graphics Group, CS Dept

Overview z. Cognitive psychology wish list z. Info. Vis problems z. Graph drawing examples ygraph == network ynode-link as in graph theory ynot bar charts

Interaction z. Fluid interaction is hallmark of modern CG z. Can create non-real environments z. Example: navigation y. RW: rigid motion across fixed terrain y. CG: distortion, warping of structure/space xwhen (if ever) is this useful?

Wanted: Prescriptive Advice z. Have for static 2 D y. Gestalt, Bertin, Mackinlay, etc z. My wish list: ydynamic, spatial xwhen are distortion (focus+context) systems useful?

Process z[Distill domain knowledge into explicit problem statement] z. Find visual technique(s) to help solve problem with preattentive processing z. Feedback: is system solving problem? y. Are they using it? y. Does it help?

CP wish list, cont. z. Analyze what’s good when: y. Tease apart complex visual metaphors into constituent low-level components y. Conceptual framework, user studies z. Backmapping: once know what it’s good for, what other domains can be abstracted into this problem?

Problems z. Scalability yreal-world datasets consistently outstrip tools z. Efficacy ylack of rigor in evaluating viz systems z. Novelty ycreating new visual metaphors is difficult z. Adoption yend user buy-in

Graph drawing: traditional z. Static++ yinteractivity not inherent ypan/scroll/zoom substitute for really big paper z. Does not scale ysmall (dozens, hundreds, maybe thousands) yaggregation/abstraction of large datasets xgreat for expository, poor for exploratory zspring-force, circular, hierarchical, etc

Three Interactive Systems z. Geographic: Planet Multicast z. Hyperbolic: H 3/H 3 Viewer z. Importance gradient: Constellation

Geographic: Planet Multicast z. Target users: Mbone maintainers z. Technique: arcs on globe [Eick 95] z. Task: identify long-haul misconfigurations

Planet Multicast excerpt z. Interaction techniques: yclick on arc for tunnel info yrotate globe around center x selective occlusion yrotate around point on surface xhorizon view disambiguates

PM analysis z. Visual popout: ylong-distance links ycluttered tunnel topology z. Literal & natural: no explanations needed z. Filtering: intercity not intracity y 4000 -> 700 yhemisphere occlusion

PM efficacy zgeog distance only approximates bottleneck sources yideal data uncollectable yfew false positives, many false negatives z. IP address -> lat/lon mapping infeasible z. Adoption: ymaintainer coauthor during development z. Does it help? y. Anecdotal

Hyperbolic: H 3 Viewer z. Target users: webmasters, gawkers z. Technique: yuses 3 D hyperbolic space for recursive hemispherical layout and navigation yspanning tree backbone with nontree links drawn on demand yguaranteed frame rate drawing algorithm z. Task: show context of surfing choices

H 3 Viewer video

H 3 analysis z. Visual popout: ysubtree clusters z. Filtering: yspanning tree in default case z. Scalability: y>100, 000 nodes

H 3 Efficacy z. Adoption: y. Ships with SGI Irix 6. 2 -6. 5 xminor user testing xuser feedback in the wild extremely minimal y. PD code, other developers xnetworking, genetic algorithms z. Does it help? y. Real user study prelim results promising

Importance Gradient: Constellation z. Target users: two linguists at MSR z. Techniques: ycustom spatial layout, horiz gradient ycareful use of HSV to minimize clutter impact z. Task: debug semantic network creation yfind implausible computed paths

Constellation video

Constellation analysis z. Visual popout: yhotspots, highlighted structures z. Filtering: none ydesign principle of avoiding hidden state z. Adoption: TBD ystill under development ypros and cons of tiny user community

Conclusion z. Key problem in the field: y. Evaluating efficacy y. Scalability z. Exploration of the design space ythree quite different GD systems

Acknowledgements z. Planet Multicast: y. Eric Hoffman, Kim Claffy, Bill Fenner z. Site Manager: y. Greg Ferguson, Alan Braverman, Ken Kershner z. Constellation: y. Francois Guimbretiere, George Robertson z. Advisor: Pat Hanrahan

More info zhttp: //graphics. stanford. edu/~munzner ypapers ytalks ysoftware