Transcript Slide 1

On Using Existing Time-Use Study Data
for Ubiquitous Computing Applications
UbiComp ’08
Kurt Partridge and Philippe Golle
Palo Alto Research Center
SangJeong
Time-Use Studies
• American Time-Use Survey (ATUS)
▫ USA, Bureau of Labor Statistics
▫ To estimate work not included in economics measures (e.g.,
home childcare)
• Many versions
▫ Korean 1999
▫ Japanese over 200,000
▫ American Heritage Time Use Study (AHTUS)
 ATUS + 4 older studies
▫ Harmonized European Time Use Study (HETUS)
▫ Multinational Time Use Study (MTUS)
▫ www.timeuse.org
Excerpt of Time-Use Data
Activity by Time of Day
ATUS Activity Classification
Suitability of Time-Use Data for Ubicomp
• Duration difference
▫ A couple hours vs. an instant to tens of minutes
▫ “eating breakfast” vs. “scooping granola, pouring milk, lifting
spoon, …”
▫ Cannot predict activities at detail, but can bias predictions toward
more likely activities
• Domain specificity difference
▫ All activities in an entire day vs. a limited domain (physical motion,
in-home activities of daily living, mechanical repair)
▫ Can benefit from cross-domain inferences
• Cognitive interpretation difference
▫ Participant and interviewer vs. sensor
▫ Can collect data of privacy-sensitive activities, e.g. bathroom use
Inferring Activity from Context
Using maximum likelihood classifier / tenfold cross-validation
Activity Inference Accuracy, by Location
Further Research Questions
• How much do time-use activity and location taxonomies
vary?
• What issues arise when adopting an activity taxonomy for
a ubicomp application?
• What methodologies used by time-use studies can be
adopted in ubicomp systems?
• How can ubicomp contribute to time-use study research?