the night is young: urban crowdsourcing of nightlife patterns
TRANSCRIPT
The Night is Young: Urban Crowdsourcing of Nightlife Patterns
Darshan SantaniIdiap Research Institute and EPFL Switzerland
Joint work with Joan Isaac-Biel, Florian Labhart, Jasmine Truong, Sara Landolt, Emmanuel Kuntsche and Daniel Gatica-Perez
A large-scale mobile crowdsourcing study to capture, examine, and provide insights on nightlife patterns of
youth in Switzerland.
Youth@Night
Alcohol Epidemiologists
Urban Geographers
Computer scientists
Study Design
Lausanne
ZurichLocation
2 major Swiss nightlife hubs
Excellent public transportation
Drinking in public places is legal
High cost of a night out
Study Design
Lausanne
Zurich
Location Recruitment Mobile Apps
Weekend Nights between 8PM to 4AM
Two major nightlife hubs in Switzerland
On street recruitment of youth aged 16-25 years
Mob
ile D
ata
Col
lect
ion
Fram
ewor
k Place Functional Attributes Ambiance
Drink Attributes Social
Place Survey
Drink Survey
Video Survey
\
Sur
vey
Dat
a (E
MA
)
Sensor Data\
Mob
ile S
enso
r an
d Lo
gsIn
terv
iew
s an
d S
urve
ys
Collected Dataset
Study Duration: September – November 2014
Fridays and Saturdays between 8PM and 4AM
Each person was requested to contribute for 10 nights
Each check-in contains place, drink and video responses
Spatial Coverage
Participants Home Locations: 128 unique postal codes 11 states (cantons) of Switzerland
Lausanne
Zurich
Research Questions
RQ1: What are the places and social contexts in which young people hang out during night?
RQ2: What are the connections between automatically extracted physical ambiance and in-situ vs. external
observations?
RQ1: What are the places and social contexts in which young people hang out during night?
Participants Demographics
Balanced gender ratio (48% female)
Over 83% of participants reported to be living with their parents
63% go out at least once per weekend
40% reported living within the city limits of either Zurich or Lausanne
Population is different than those reported in previous UbiComp research
62% are below the age of 20
[1] Yohan Chon et al. “Understanding the Coverage and Scalability of Place-centric Crowdsensing”, UbiComp' 13[2] Rui Wang et al. “SmartGPA: How Smartphones Can Assess and Predict Academic Performance of College Students”, UbiComp' 15
User Contributions
894 videosAvg. check-ins per participant: 6.5
Place Analysis
Private Places
Public Places
47.3% check-ins
52.7% check-ins
PBS: Public Parks, Plazas, Squares
Public Place Categories
Place Analysis
83% live with their parents
Going out is relatively costly
Videos recorded at homes clearly have an intimate, unfiltered flavor
Youth@Night Foursquare
Representativeness of Social Media
Place Analysis
83% live with their parents
Going out is relatively costly
Videos recorded at homes clearly have an intimate, unfiltered flavor
Youth@Night Foursquare
Representativeness of Social Media
Place Analysis
83% live with their parents
Going out is relatively costly
Videos recorded at homes clearly have an intimate, unfiltered flavor
Youth@Night Foursquare
Representativeness of Social Media
Comparing Spatial Coverage
Youth@Night
FoursquareMix
Social Context and Activities
For most of the public places, the majority of check-ins were reported to be with
fewer than three people
Alcohol Consumption
Home Drinking: 48% of private check-ins with alcohol consumption
Street Square Drinking: 84% of check-ins with alcohol consumption in the PBS
category
Findings from RQ1
Youth spend a considerable amount of time hanging out away from mainstream nightlife areas
Crowdsensing with young workers can capture places along the full spectrum of social and ambiance context
RQ2: What are the connections between automatically extracted physical ambiance and in-situ vs. external
observations?
Conceptualize this problem in terms of Brunswik’s lens model
Video Content Analysis
● Video Corpus● 843 videos; Mean duration: 9.4 seconds● Public Places: 69% vs. Public Places: 66%
● Video Crowdsourcing Compliance – 32% of check-ins with no video● Safety, Ethics and Social reasons – ~30% each
● Automatic Feature Extraction ● Loudness (AEL) as audio power using the audio channel of videos ● Brightness (AEB) as average brightness of videos in the YUV color space
● Manual Coding of Videos● Rate the loudness and brightness after watching the videos
In-situ Ambiance
Loudness Brightness
Bars and clubs were reported to more crowded, louder, and darker
Automatically Extracted Ambiance
Private places are quieter than public places
Clubs, events, and bars are found to be the loudest places
Clubs and bars to be the darkest placestogether with PBS across all place types
Travel category have the highest median brightness
Loudness Brightness
Diversity of ambiance in home environments may reflect different social settings
Physical Ambiance Feature Comparison
To examine the reliability of different crowd-workers (in-situ and ex-situ) in comparison with automatic feature extraction
Are in-situ self-reports reliable?
What can be considered the “ground-truth”?
Brunswik Lens Model
In-situ PerceivedAmbiance
Manually PerceivedAmbiance
Aut
omat
ical
ly E
xtra
cted
Am
bian
ce
Cue Validity (rv) Cue Utilization (ru)
Higher rv and ru is obtained if what is automatically perceived was equally perceived by both in-situ and external observers respectively
Feature Reliability
Cue utilization effect sizes are overall higher than for cue validity for both
loudness and brightness
Effect sizes of public and private places are comparable
In-situ External
Findings from RQ2
Basic automatic audio-visual features are informative of ambiance
External observers consistently assess loudness and brightness
Automatic ambiance described more accurately the perception of external online observers than that of participants in-situ
With the help of young workers, mobile crowdsourcing allows to understand of patterns of physical mobility, activities, and social
context of youth population
The resulting data is diverse across place types, social, ambiance context, and video content.
Taken together, the data and analysis will enable new research directions in human geography and alcohol epidemiology
Overall Conclusions
Q & A
Email: [email protected]: @SabMayaHai