toward interaction models derived from eye-tracking data presented at polish ia summit 2012

Upload: jaceksg

Post on 05-Apr-2018

218 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    1/22

    Interaction Models DerivedFrom Eye-tracking Data .

    Jacek Gwizdka & Michael ColeRutgers University, USA

    [email protected]

    http://jsg.tel

    April 19, 2012

    Towards

    http://eye.html/http://eye.html/
  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    2/22

    Eye-tracking?

    2

    Eye-trackers Tobii

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    3/22

    Eye-movement in UX Research

    3

    http://images/KPE1/KPE1_fromS004/05-1/Composer_Answ_Lookup_KPE1_fromS004_05-1Segment%201.avi
  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    4/22

    There is a Lot More Eye-tracking Datacan offer UX / HCI / IA

    4

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    5/22

    Eye-tracking Data

    5

    State2State1

    State3

    Patterns

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    6/22

    Eye-movement Patterns

    New methodology to analyze eye-movement patterns

    Model readingand Measure cognitive effort Correlate with higher-level constructs

    user task characteristics,

    user knowledge, etc.

    6

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    7/22

    Eye-tracking Fundamentals

    7

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    8/22

    Reading Model Origins

    8

    Based on E-Z Reader modelRayner , Pollatsek, Reichle

    Serial reading

    Words can be identified in parafovialregion

    Early lexical access (word familiarity) + Complete lexical processing (word identification)

    2o (70px) foveal region parafoveal region

    a bit MORE

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    9/22

    Two-State Reading Model

    9

    Filter fixations < 150ms (min time required for lexical processing)

    Model states characterized by:

    probability of transitions; number of lexical fixations; duration

    length of eye-movement trajectory, amount of text covered

    ScanRead

    1-q

    p

    1-p

    q

    a bit MORE

    isolatedfixationsfixation

    sequences

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    10/22

    Example Reading Sequence

    10

    Fixation sequence: (F F F) F (F F F) F F F F (F F F F F F) FReading modelstates: R S R S S S S R S

    Reading state R | Scanning state S

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    11/22

    Cognitive Effort Measures of Reading

    Reading Speed

    Fixation Regression

    Perceptual Span

    Fixation Duration(lexical processing excess)

    11

    foveal region

    a b c d

    Perceptual span = Mean(a,b,c,d)

    regression

    excess

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    12/22

    User Study 1: Cognitive Effort and Tasks

    12

    OBI: advanced obituary

    INT: interview preparation

    CPE: copy editing

    BIC: background information

    N=32

    MORE

    JournalistsInformation Search

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    13/22

    Eye-data and Cognitive Effort Measures

    Cognitive effort measuresreading speed

    mean fixation durationperceptual spantotal fixation regressions

    Task complexity

    by designCopy Editing (CPE)Advance Obituary (OBI)

    Search efforttask timepages visitedqueries entered

    Subjective TaskDifficulty

    CPE INT BIC OBI

    As expected:

    Copy Editing CPE easiestAdvance Obituary OBI most difficultSig: Kruskal-Wallis 2 =46.1, p

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    14/22

    Eye-data and Task Characteristics

    14

    Measure Related Task CharacteristicsFrequency

    of reading

    state

    transitions

    SRbias to readAdvanced obituary andInterview preparationtasks: search

    for document; task goal not specific

    RS bias to scanCopy Editingtask: search for segment and task goal specific

    ScanRead

    1-q

    p

    1-p

    q

    MORE

    Copy EditingInterview preparation

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    15/22

    User Study 2: Assessing Users Knowledge

    Search in Genomics Domain

    15

    N=40

    MORE

    Rate own domainknowledge

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    16/22

    Results: Modeling Domain Knowledge

    16

    Reading Model features &cognitive effort measures

    Eye-tracking Data

    Domain knowledgeMeSH-based self-ratings

    predicted

    self-rated

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    17/22

    Results: Modeling Domain Knowledge

    17

    predicted

    self-rated

    Reading Model features &

    cognitive effort measuresreading seq length and

    total durationperceptual spanfixation durationregressions

    Reading Model

    Eye-tracking Data Random

    Forest Model

    Domain knowledgeMeSH-based self-ratings

    m

    tk

    =PDK

    m

    iii

    *5

    )*(1

    Foreachuser predict

    build model

    agglomerativehierarchical

    clustering(Wards)

    PDK: Participants

    domain knowledge

    MORE

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    18/22

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    19/22

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    20/22

    From Real-time Interactions to Applications

    CognitiveLoadDomainKnowledge InformationRelevance

    Adaptpresentation& content

    EnableInteraction(e.g., disabilities)

    TaskAspects

    Eye-TrackingData

    Standardinput devices(mouse, keyboard)

    other psycho-physiological devices(EEG, SCR, HRV)

    Betterunderstandinteraction

    Micro-level

    Macro-level

    Applications

    CognitiveLoad Model

    ReadingModel

    TaskModel

    Models

    20

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    21/22

    Thank You! Dziekuje!

    21

    Funding: Google, HP, IMLS (now funded by IMLS CAREER)

    Collaborators: Drs. Nicholas Belkin, Art Chaovalitwongse (U Wash), Xiangmin Zhang,

    Ralf Bierig (Post Doc); PhD students: Michael Cole (co-author), Chang Liu, Jingjing Liu, Irene Lopatovska

    + many Master and undergraduate students

    Acknowledgements:

  • 7/31/2019 Toward Interaction Models Derived From Eye-tracking Data presented at Polish IA Summit 2012

    22/22

    Pytania?

    22

    More info & contact http://jsg.tel