lecture 2: design studies - computer science at...

$: Lecture 2: Design Studies - Computer Science at UBCtmm/courses/cpsc533c-06-fall/slides/design.pdfapplication domain background information can be useful to ... (wavelet, fourier, fractal)$
Lecture 2: Design StudiesInformation VisualizationCPSC 533C, Fall 2006

Tamara Munzner

UBC Computer Science

14 September 2006

Questions

I impressed by quality so far!I want at least one question per reading

I if <5 texts you can pick which to use for multiple Qs

I plain (ASCII) text not Word/PDF/etc

Papers Covered

Cluster and Calendar based Visualization of Time Series Data.Jarke J. van Wijk and Edward R. van SelowProc. InfoVis 99, pp 4-9http://www.win.tue.nl/˜vanwijk/clv.pdfUsing Multilevel Call Matrices in Large Software Projects.Frank van HamProc. InfoVis 2003, pp 227-232http://www.win.tue.nl/˜fvham/DL/callmatrix.pdfConstellation: Linguistic Semantic NetworksTamara MunznerInteractive Visualization of Large Graphs and Networks (PhDthesis) Chapter 5, Stanford University, 2000, pp 87-122http://graphics.stanford.edu/papers/munzner thesis

Design Study

I describe taskI justify solutionI refine until satisfied

Design Study Definition

Design study papers explore the choices made when applyinginfovis techniques in an application area, for example relatingthe visual encodings and interaction techniques to therequirements of the target task. Although a limited amount ofapplication domain background information can be useful toprovide a framing context in which to discuss the specifics ofthe target task, the primary focus of the case study must be theinfovis content. Describing new techniques and algorithmsdeveloped to solve the target problem will strengthen a designstudy paper, but the requirements for novelty are less stringentthan in a Technique paper.

[InfoVis03 CFP, infovis.org/infovis2003/CFP]

Cluster-Calendar, van Wijk

I data: N pairs of (value, time)I N large: 50K

I tasksI find standard day patternsI find how patterns distributed over year, week, seasonI find outliers from standard daily patternsI want overview first, then detail on demand

I possibilitiesI predictive mathematical models

I details lost, multiscale not addressedI scale-space approaches (wavelet, fourier, fractal)

I hard to interpret, known scales lostI 3D mountain: x hours, y value, z days

I excellent example, emulate for project writeups!

3D Time-series DataI 3D extrusion pretty but not useful

I daily, weekly patterns hard to see

[van Wijk and van Selow, Cluster and Calender based Visualization of Time SeriesData, InfoVis99, http://www.win.tue.nl/˜vanwijk/clv.pdf]

Hierarchical ClusteringI start with all M day patterns

I compute mutual differences, merge most similar: M-1I continue up to 1 root cluster

I result: binary hierarchy of clustersI choice of distance metricsI dendrogram display common

I but shows structure of hierarchy, not time distribution


Link Clusters and CalendarI 2D linked clusters-calendars shows patterns

I number of employees:I office hours, fridays in/and summer, school breakI weekend/holidays, post-holiday, santa claus


Power Consumption


Lessons

I derived space: clustersI visual representation of time: calendar

I linked displayI interactive exploration

I clear task analysis guided choicesI reject standard 3D extrusionI reject standard dendrogram

I critique

I color choice not so discriminableI especially legend

Lessons

I derived space: clustersI visual representation of time: calendar

I linked displayI interactive exploration

I clear task analysis guided choicesI reject standard 3D extrusionI reject standard dendrogram

I critiqueI color choice not so discriminableI especially legend

Multilevel Call Matrices, van Ham

I large software project, implementation vs. specI link matrix vs. node network

matrix force-directed layered subset

[van Ham, Using Multilevel Call Matrices in Large Software Projects. InfoVis03http://www.win.tue.nl/˜fvham/DL/callmatrix.pdf]

MatricesI uniform, recursive, stableI subdivide by

total component count visible subcomponent count


Zooming

I abstraction levels

I linear interpolation plus crossfadeI trajectories: will read van Wijk 03 in week 6


Additional Encoding

color:call allowedby spec

color:local regionclosest red

transparency:call density

I histograms: size distribution


Tasks Succesfully Supported

I visual categorizationI i.e. libraries with mostly incoming calls

I previous summary shown to be incompleteI spotting unwanted callsI determining component dependencies

Linguistic Networks, Munzner

I data: MindNet query resultsI definition graph

I dictionary entry sentenceI nodes: word sensesI links: relation types

[Munzner, Interactive Visualization of Large Graphs and Networks (PhD thesis),Stanford University, 2000, http://graphics.stanford.edu/papers/munzner thesis]

Semantic Network

I definition graphs used as building blocksI unify shared wordsI large network

I millions of nodesI grammar checking now, translation futureI global structure known: dense

I probes return local info

Path Query

I best N paths between two wordsI words on path itself

I definition graphs used in computation

Task: Plausibility Checking

I paths ordered by computed plausibilityI researcher hand-checks results

I high-ranking paths believable?I believable paths high-ranked?I are stop words all filtered out?

Top 10 Paths Kangaroo→ Tail

Goal

I create a unified view of relationships between pathsand definition graphs

I shared words are keyI thousands of words (not millions)

I special purpose algorithm debugging toolsI not understand structure of English

Constellation Video

Traditional Layout

I avoid crossingsI reason: avoid false attachments

ambiguity artifact salience

Information Visualization Approach

I spatial position is strongest perceptual cueI encode domain specific attributeI plausibility gradient

Constellation Semantic Layout

I novel layout algorithmI paths as backbone, definition graphs attachedI curvilinear gridI iterative design for maximum semantics with

reasonable information density

I allow crossings for long-distance proxy links

Selective Emphasis

I highlight sets of boxes and edgesI interactionI additional perceptual channels

I avoid perception of false attachments

Hidden State

I avoid hidden stateI change salience instead of toggle drawing

I why? closed world assumptionI implicit assumption: if not visible, doesn’t existI easy to forget previous actions

I draw false negative conclusions

Single vs. Multiple Word Instances

Information Density

I early prototype: poor

Information Density

I design tradeoff with visual salience

Information DensityI grid adjustment

Task-oriented design

I task-specific methods

lecture 2: design studies - computer science at...

Documents