march12 chatterjee
TRANSCRIPT
MOUSTRESS: DETECTING STRESS FROM MOUSE MOTION (CHI, 2014)
Presenter: Soujanya Chatterjee
David Sun, Pablo Paredes and John Canny
Introduction• Chronic stress has negative effects in our lives.• Physiological markers discovered linked with stress are
HRV, muscle tension, GSR, pulse oximetry.• Explores the use of computer mouse operations for
measuring stress.• Increased arm muscle activity is a prominent stress marker.• Utilizes Mass-Spring-Damper (MSD) system for hand-arm
dynamics to capture muscle stiffness.
IntroductionMass Spring Damper model• MSD system consist of,
• Mass (m), spring component (k), viscous damper (c).• k determines stiffness of the string.• Increased k suggest increased muscle stiffness hence increase in
stress.
• MSD parameters,• Damped frequency (ω), where ω ∝ √k, with m constant.• Damping ratio (ζ ), where ζ ∝ √(1/k), with m constant.
• MSD model best suited for modelling muscle stiffness.
Research QuestionHypothesesObservation: Muscle stiffness increases with increased stress• Due to high stress, the damped frequency ω will be higher.• Due to high stress, the damping ratio ζ will be lower.• Due to high stress, time to complete different mouse tasks will
be shorter.
Methodology
Mouse task designThree abstract mouse operations selected: Point and click
Drag and drop
Steering
Methodology
Model computation methodology 49 participants completed 51, 1 hour session of mouse tasks. Experiment had 4 main phases
Calm-phase - de-stressor task assigned. mCalm-phase - Immediately after Calm-phase subject performs mouse tasks. Stressor-phase - Stress-inducing activity. mStressor-phase - Immediately after Stressor-phase subject performs mouse
tasks. The ordering was randomized between subjects to control order effect.
Data Processing• Three independent measures of stress recorded
• Subjective stress rating – Self-reported• ECG data – Continuous data collected• Mouse activity data.
• Three mouse tasks and mouse motion was recorded in a logger.• Event tuple (dx,dy,t), where,
• dx – displacement along x axis.• dy – displacement along y axis.• t – time stamp.
Key Results
Subjective stress rating• One-tailed, paired t-test applied between Calm-Stressor and mclam-mstressor phase.• SSR indicator was significantly higher during Stressor and mStressor phase as compared
to Calm and mCalm phase respectively.
Heart Rate Variability• HRV data collected during the four phases.• MeanRR and HHF were significantly lower and HLF was significantly higher in Stressor
phase as compared to Calm phase.• No difference in mCalm and mStressor phase.
Key ResultsMouse Motion
• Following are the mouse motion stress indicators.
Key Results
Mixed Model Task• Model independent of specific mouse task. • Involves all random task types and configurations.• Each subject average of the indicators obtained for mStressor and mCalm phases.
Key Results
Task Specific Model• Model dependent on the three specific mouse tasks. • Separate averages of the indicators computed for each task. Clicking task
Damped frequency (ωx and ωy) significantly higher for mStressor than in mCalm.
Damping ratio along y axis (ζy) significantly lower for mStressor than in mCalm.
Drag and drop task – No Effect Steering task
Damped frequency (ωx and ωy) significantly higher for mStressor than in mCalm.
No effect on damping ratio.
Summary and Contribution• Hypotheses 1 is well supported by both the models.
• Effects of stress was strong and consistent for damped frequency, ωx and ωy.
Summary and Contribution
• Hypotheses 2 is not very well supported.
• Effects of stress was not strong for damping ratio on ζx.
• No effect at all on ζy.
• Intuitively, stress reflected more on dominant left-to-right mouse movement.
• Hypotheses 3 supported only by clicking task.• Task completion time(t) significantly lower in mStress than in mCalm
Summary and Contribution
Contribution• Probable alternative to classical stress detection from physiological
measurement.• Less prone to motion artifacts.• Opens a new field of research on stress detection while interacting with
computer.
Critique of work• Bounded within closed environments, where people can use
mouse.• Model accuracy depends on dominant mouse direction.• Difficult to deploy in natural field environment.
Thanks.
Questions?