Song Gao Email: [email protected] University of California, Santa Barbara Human mobility, urban structure analysis, and spatial community detection from mobile phone data http://stko.geog.ucsb.edu Big Geo-Data Age
May 24, 2015
Song Gao
Email: [email protected]
University of California, Santa Barbara
Human mobility,
urban structure analysis,
and spatial community detection
from mobile phone data
http://stko.geog.ucsb.edu
Big Geo-Data Age
Open Questions 1). Does distance still constricts human mobility in geographic
space?
2). Whether the information communication technology (ICT)
increase or decline the probability of physical movements of
urban residents in daily life?
3). What is spatio-temporal patterns of phone call activities in
urban space? How to explore?
4). Does the phone call interaction follow Tobler’s first law
(TFL) of geography?
Background
Location Awareness Devices(Mobile Phone、GPS)
Large scale spatio-temporal datasets
Ratti, 2010 MIT SENSEable City Lab
Urban Computing
sense city dynamics to
enable a city-wide
computing as so to serve
people and cities.
Yu Zheng (2012),
Microsoft Research Asia
It is emerging as a concept where sensor,
device, person, vehicle, building, and street in
the urban areas can be used as components to
Background
Individual Level
Human mobility (Nature, Science, PNAS)
Trajectory data mining(ACM,IJGIS)
Community Detection(Complex Networks)
Background
Aggregate (Regional Level)
Dynamic urban landscape
Spatial interactions between sub-regions
Transportation demands estimation
Information, Communication, Technology & Space, Place & Social
Community Networks
Human
Mobility
Urban Structure
Space is opportunity,
Place is understood reality.
• Population distribution
• Movements
• Mobile landscape
• Functional region
• Flow
• ……
Data Descriptions Mon Tue Wed Thur Fri Sat Sun
11.19 10.89 10.92 10.70 11.01 9.82 9.44
Song Gao, April, 2013
approximate 10 million records a day
Human Mobility
Spatio-temporal patterns can be found with a
large amount of trajectories (X,Y, T)
GIS visualization and analysis applied to
represent and model individual dynamics
Human
Mobility
Song Gao, April, 2013
Geo-visualizing
Space-time path Frequency of
occurrence
Kang C., Gao S. et al. Analyzing
and Geo-visualizing Individual
Human Mobility Patterns Using
Mobile Call Records. 2010
Song Gao, April, 2013
Credit: Song
The variability of mobility in space-time
Regular
Irregular
Song Gao, April, 2013
The distribution of the ROG
covered with 869,992 mobile phone users.
Radius of gyration
Song Gao, April, 2013
Urban Structure
Aggregate approach (Hourly)
--Celli (volume00, volume01, volume02,…… volume23)
The scale of the urban area, may including the
city and some inner suburbs, to highlight
interesting metropolitan dynamics
Calculate the kernel density
Urban Structure
Song Gao, April, 2013
Spatio-temporal patterns
AM 03-04 AM 06-07 AM 09-10
PM 15-16 PM 18-19 PM 21-22
Song Gao, April, 2013
Mobile Landscape
LandScan Global Population Data
-- 1KM resolution
Song Gao, April, 2013
Correlation with population distribution
AM 06-07 AM 09-10
PM 15-16 PM 18-19 PM 21-22 Song Gao, April, 2013
Correlation with population distribution
r= 0.714 r= 0.697
r= 0.632 r= 0.748 r= 0.785
Spatial Interaction Network
Spatial Networks describe
the networks in which the
nodes are embedded in a
geographical space
Goal: to explore
telecommunication flow in
geographic space and to
understand how the spatial
context affect such
interactions
Community in spatial networks
Song Gao, April, 2013
Motivation Whether interaction structure, friendship likelihoods reveal
political boundaries, physical barriers, or social divide
Song Gao, April, 2013
Spatial effects on networks
(1) Spatial constraints on the distribution of
nodes embedded in geographical locations;
(2) Physical networks like roads and railways,
which are affected by spatial topology;
(3) Restrictions on long-distance links due to
economic costs.
Community in spatial networks
Song Gao, April, 2013
Two networks of spatial interactions
G_TeleFlow (V, E) be a weighted-undirected network graph
of phone call flows where Thiessen polygons of mobile
base stations are transformed into nodes (V) while
interactions among stations are represented by weighted
edges (E).
G_MoveFlow (V, E) be a weighted undirected network graph
of human movements and let Mijt represent the total
movement flow between cell i and cell j during time interval
t, including movement flows both from i to j and from j to i.
Song Gao, April, 2013
Distance Decay of Spatial Interactions
Cumulative probability
function of distance
distributions in two
interaction networks:
89.47% phone-call
interactions and 90.98%
movements occur across
distances less than 20 km
Song Gao, April, 2013
Distance Decay of Spatial Interactions
the power-law fit with a decay
parameter β=1.45
G_TeleFlow G_MoveFlow
dP
the power-law fit with a decay
parameter β=1.60
Song Gao, April, 2013
Community Detection Algorithm
The nodes of the network can be grouped into sets of nodes
so that each community is densely connected internally.
Modularity maximization
Minimum-cut method
Hierarchical clustering
Girvan–Newman algorithm
Clique based methods
Song Gao, April, 2013
Modularity is defined as the sum of differences
between the fraction of edges falling within
communities and the expected value of the
same quantity under the random null model.
Incorporating Gravity Model
(Gao et al. 2013, Transactions in GIS)
The fraction format of gravity-modularity for detecting communities:
Community detection results of networks of
call interaction (G_TeleFlow)
Day Node Edge Number Avg
Size
Modularity
Monday 609 41960 10 61 0.528
Tuesday 608 40902 10 61 0.533
Wednesday 609 40649 10 61 0.538
Thursday 609 56070 8 76 0.405
Friday 608 54091 8 76 0.422
Saturday 605 48673 8 75 0.438
Sunday 607 46506 8 75 0.446
Song Gao, April, 2013
Urban Community detection results
MAXID-----: 616
NUMNODES--: 609
NUMEDGES--: 41960
TOTALWT---: 934561
NUMGROUPS-: 10
MINSIZE---: 27
MEANSIZE--: 60.9
MAXSIZE---: 120
MAXQ------: 0.527837
STEP------: 599
Examples of differentiated geographical context of
isolated regions in spatial communities
Examples of differentiated geographical context of
isolated regions in spatial communities
Cell A locates in the overpass
intersection of ring highway and
the airport expressway which is
near a large residential suburb
area of this city, and a high
volume of call interaction make it
merged to the northern spatial
community (yellow) of official
cells.
Examples of differentiated geographical context of
isolated regions in spatial communities
Cell B has been grouped into the same
distant community on Monday,
Thursday and Friday, whereas it
aggregates into nearby spatial adjacent
community on weekends.
It corresponds to a set of governmental
buildings which has strong connections
with eastern cells (green) of central
business district on weekdays.
Examples of differentiated geographical context of
isolated regions in spatial communities
Cell C has a strong link to the southern
cells (red) during the whole week and
they are assigned to the same community.
Cell C locates nearby a industrial place
which covers a wood processing plant,
food brewery, and wholesale market.
There may be business communications
that make these cells aggregated into the
same community.
Examples of differentiated geographical context of
isolated regions in spatial communities
Cell D covers a local famous farm and
implies a business connection to the city
community. In order to identify whether
physical movements also exist between
these spatially separated cells, we will
refer to the partition results of the
network of movements.
G_TeleFlow
G_MoveFlow
Relation between Telecommunication
and Movement
ICT & Mobility:
-替代(Substitution)
-增强(Stimulation)
-缓和(Modificaiton)
a causal relationship?
Mon Tue Wed Thur Fri Sat Sun
R2 0.857 0.852 0.852 0.848 0.852 0.857 0.865
Correlation coefficients between phone call interaction and movements
Song Gao, April, 2013
Conclusion and Discussion 1). Does distance still constricts human mobility in geographic
space? -- Yes, it is.
2). Does the information communication technology (ICT)
increase or decline the probability of physical movements of
urban residents in daily life?
– Statistically yes, but not sure whether causally
3). What is spatio-temporal patterns of phone call activities in
urban space? How to explore? -- Dynamic mobile landscape
4). Does the phone call interaction follow Tobler’s first law
(TFL) of geography? -- To some degree, Yes
5). A combined qualitative-quantitative framework to identify
phone-call interaction patterns in spatial networks
Gao et al. 2013 Discovering spatial interaction communities from mobile phone data.
Transactions in GIS 17(3)
Gao et al. 2013 Understanding urban traffic flow characteristics: A rethinking of
betweenness centrality. Environment and Planning B: Planning and Design 40(1)
Kang et al. 2013 Inferring properties and revealing geographical impacts of inter-city
mobile communication network of China using a subnet data set. International Journal
of Geographical Information Science 27(3)
Kang et al. 2012 Towards Estimating Urban Population Distributions from Mobile Call
Data. Journal of Urban Technology 19(4)
Kang et al. 2012 Intra-urban human mobility patterns: An urban morphology
perspective. Physica A: Statistical Mechanics and its Applications 391(4)
Liu Y et al. 2012 Understanding intra-urban trip patterns from taxi trajectory data.
Journal of Geographical Systems 14(4)
Liu Y et al. 2012 Urban land uses and traffic ‘source-sink areas’: Evidence from GPS-
enabled taxi data in Shanghai. Landscape and Urban Planning 106
References
Yuan et al. 2012 Correlating mobile phone usage and travel behavior – a case study of
Harbin, China. Computers, Environment and Urban Systems 36(2)
Walsh et al. 2011 Spatial structure and dynamics of urban communities.
Ratti et al. 2010 Redrawing the map of Great Britain from a network of human
interactions. Plos One 5(12)
Guo, D. 2009 Flow Mapping and Multivariate Visualization of Large Spatial
Interaction Data", IEEE Transactions on Visualization and Computer Graphics, 15(6)