2. Look Before You Link: Eye Tracking in
Multiple Coordinated View Visualization
Chris Weaver
School of Computer Science and the Center for Spatial Analysis
University of Oklahoma
weaver@cs.ou.edu
3. Pre-filtering Grouping Cross-filtering
γ G’
Name
mr >= k minrating mn G mn
φmn mn
πmn V
mn
σmn
coordinated multiple views
Movies
T φm T’ γ G φmd G’ πmd V σmd
Date
m m md md md md
box office
average rating
Rating
φmr T’ πmr V σmr
number of ratings
are common in
id?
mr mr
Genres
γ G’
Name
Tg
φg T’ g gn G gn
φgn gn
πgn V
gn
σgn
visual analysis tools
Oscars
T φo T’ γ G φot G’ πot V σot
Type
o o ot ot ot ot
γ G’
Name
Tp
φp T’ p pn G pn
φpn pn
πpn V
pn
σpn
People
|γpn(pn)| >= k γ G φpr G’ πpr V σpr
Role
minroles pr pr pr pr
elemental forms of coordination are established
compound forms of coordination are emerging
cross-filtering node, edge, pack ∞
matrix matrices Λ λ? Layout
Grouping
1 2 7
Nodes
Tα’ γα Gα φα G’α N
φα Nα Σ T
N
input
ψN T
N
graph πN T
N
glyph
φN T
view
N
σN?
3 4 8
Edges
E E E E
Tα’ σα E
φαβ Eαβ Σ Tinput
ψE T
graph πE T glyph
φE T
view σE?
’ P 5 P 6 P 9 P
α β
id id Cαβ φαβ Cαβ P
φαβ Pαβ Σ Tinput
ψP T
graph πP T glyph
φP T
view σP?
Packs
Tβ’ σβ P
φβα Pβα
Cliquing Drilling Slicing Collecting Forming Encoding Filtering Brushing
3
4. Cross-filtered views
analytic utility arises from navigation and selection
in individual views
and
in compositions of views
by chaining together sequences of interactions
Jigsaw list view
4
5. we’re still looking mostly at tool designs in terms of
representation
interaction
process
how does representation shape interaction?
how does interaction reflect analytic process?
6. coordination is we act here
a special kind while looking there
of interaction ...on purpose!
pixels/points
here and there can be shapes/regions
entire views
(is coordinated interaction like juggling? or more like a sobriety test?)
6
7. supplant (not replace) input tracking with
eye tracking
exploit dual spatial modalities of gaze and motion to
analyze interaction patterns
are entire views more suitable targets for current
hardware capabilities?
SMIvision RED 250
temporal
rate (250Hz)
latency (10ms)
spatial
resolution (0.5°, ~10 pixels)
7
8. Cinegraph (visualization)
High-dimensional drill-down into people, genres, awards, release
dates, and box office characteristics of mainstream movies
Data Sources: www.imdb.com and InfoVis 2007 Contest Co-Chairs
8
11. so what are we planning to do?
beat the hardware into submission (sigh...)
implement a Java API for calibration and data collection
splice gaze data into the input event stream
consumed by views
expose gaze data to the Improvise transformation
pipeline/query language
metavisualize aggregated gazes in the multiview context
precompute query ensembles for likely future paths of
interaction across coordinations?
think about head-to-head collaborative coordination
(we have two trackers)
how far can we go looking at the view level?
11