Upload
amimul13748
View
221
Download
0
Embed Size (px)
Citation preview
8/13/2019 EyeR WWWpres
1/26
Eye Tracking Analysis of User
Behavior in WWW Search
Laura Granka
Thorsten Joachims
Geri Gay
8/13/2019 EyeR WWWpres
2/26
why use eye-tracking forinformation retrieval?
Understand how searchers
evaluate online search results
Enhanced interface design More accurate interpretation of
implicit feedback (eg, clickthrough data)
More targeted metrics for evaluating retrieval
performanceFigure: popular regions are
highlighted through shadow-intensity
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
3/26
key research questions
How long does it take searchers to select adocument?
How many abstracts do searchers look atbefore making a selection?
Do searchers look at abstracts ranked lowerthan the selected document?
Do searchers view abstracts linearly?
Which parts of the abstract are most likely to
be viewed?
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
4/26
what is eye-tracking?
Device to detect andrecord where and what
people look at
Multiple applications:reading, usability, visual
search, in both physical
and virtual contextsEye tracking device
Figure: Cornell HCI eye trackingconfigurationView of subjects pupil onmonitor; used for calibration
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
5/26
ocular indices for www tracking
Fixations: ~200-300ms; information is acquired
Saccades: extremely rapid movements between fixations
Pupil dilation: size of pupil indicates interest, arousal
Aggregate eye-tracking graphs depictviewing intensity in key regions
Scanpath output depicts pattern ofmovement throughout screen. Blackmarkers represent fixations.
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
6/26
experimental search tasks
Ten search tasksgiven to allparticipants
Search topicsincluded travel,
science, movies,local, television,college, and trivia
Searches evenly
split betweeninformational andnavigational tasks
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
7/26
experimental procedures
Users conducted liveGoogle searches
Users allowed to
search freely, with
any queries
Script removed all ad
content
Proxy stored allpages and log files
Figure: Specific zones werecreated around each result, enablingeye-movements to be analyzedspecific to the rankings
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
8/26
sample eye-tracking output
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
9/26
8/13/2019 EyeR WWWpres
10/26
sample eye-tracking output
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
11/26
sample eye-tracking output
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
12/26
sample eye-tracking output
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
13/26
overall searching behaviorHow long does it take users to select a document?
0
1
2
3
45
6
7
8
9
10
cornell
jord
an
dudera
nch
timem
achi
ne
moun
tain
hous
ing
primary
antibio
tics
em
eril
greyho
und
meant
ime(s)
Overall mean: 5.7 seconds, St.D: 5.4
Time spent before a result is clicked
less difficultmore difficult task
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
14/26
overall viewing behavior
Total number of abstracts viewed per page
0
20
40
60
80
100
120
1 2 3 4 5 6 7 8 9 10
Total number of abstracts viewed
frequency
Mean: 3.07 Median/Mode: 2.00
Most likely to view only twodocuments per results set
How many abstracts do we view, and in what order?
Notice dipafter page
break
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
15/26
overall viewing behaviorHow many abstracts do we view, and in what order?
Instance of arrival to each result
0
5
10
15
20
25
1 2 3 4 5 6 7 8 9 10
Rank of re sult
m
eanfixationvalueofarrival
Results viewed linearly
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
16/26
overall viewing behavior
Time spent viewing each abstract compared with the frequencythat each rank is selected. Error bars are 1 SE
How much time do we spend viewing each abstract?
0
20
40
60
80
100
120
140
160
180
1 2 3 4 5 6 7 8 9 10 11
Rank of result
#timesra
nkselected
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
mean
time(s)
# times result selected
time spent in abstract
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
17/26
overall viewing behavior
Number of abstracts viewed above and below selected link
How thoroughly do we view the results set?
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
18/26
overall viewing behaviorWhat information in the abstract is most useful?
Title: 30%
Snippet: 43%
Category: 0.3%
URL: 21%Other: 5%(includes,cached, similar pages,description)
Percentage of time spent viewing each part of abstract
Other: 5.3%
Title: 30.5%
Category: 0.3%Snippet: 42.8%
URL: 21.1%
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
19/26
overall searching behavior
*Difficulty and satisfaction are ranked on a 1-10 scale; 10 meaning very difficult and very satisfied, respectively
0
1
2
3
4
5
6
7
8
9
Cornell
mansion
M ichael
Jordan
Dude
ranch
Time
M achine
Tallest
mountain
CM U
housing
NY
Primary
First
antibiotic
Emeril Greyhound
search task
ranking
Difficulty
Satisfaction
Search task difficulty and satisfaction with Google
less difficultmore difficult
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
20/26
overall searching behavior
Mean rank of selected doc: 2.66 Median/ Mode: 1.00
Task difficulty
influences rank of selected document and number of abstracts viewed
less difficultmore difficult
Cornell
mansion
M ichael
Jordan
Dude
ranch
Time
M achine
Tallest
mountain
CM U
housing
NY
Primary
First
antibiotic
Emeril Greyhound
search task
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
21/26
overall searching behavior
1. Michael Jordan statistician 20
2. Thousand acres dude ranch 11
2. One thousand acres dude ranch 11
3. 1000 acres dude ranch 9
4. Time machine movie 7
4. Carnegie mellon university graduate housing 7
5. Imdb 6
5. Emeril lagasse 6
5. First modern antibiotic 6
5. Greyhound bus 6
5. Carnegie mellon graduate housing 6
Top Query Terms Frequency
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
22/26
8/13/2019 EyeR WWWpres
23/26
conclusions
Document selected in under 5 seconds Users click on the first promising link they see
Results viewed linearly
Top 2 results most likely to be viewed Users rather reformulate query than scroll
Task type and difficulty affect viewing
behavior Presentation of results affects selection
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
24/26
future researchImpact on advertising
With such fast selections being made, will searchers even view ads?
?
Ads most likely to be seen:
Difficult task
Ambiguous info needInformational query
Low searcher expertise
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
25/26
future research
Relevance judgments
Do we spent more time viewing relevant abstracts?
Do we click the first relevant abstract viewed?
Does pupil dilation increase for more relevantdocuments
If results were re-ranked, would
viewing behavior differ?
http://www.cs.cornell.edu/8/13/2019 EyeR WWWpres
26/26