1. A tale of many cities: universal patterns in human urban mobility
Paper Presentation
Namesh Kher
INST 750 – Big Data Insights

2. Introduction
Paper focuses on determining a relationship between mobility and distance
Talks about 2 classical studies of the past
Gravity Models: There is a direct relation between mobility and distance.
No direct relation between mobility and distance and distance is a surrogate for intervening opportunities.
Data Source
For studies conducted in past major data sources were: surveys, small scale observations
Recent studies make use of
Locations based Social Services accessed via GPS enabled smartphones
Exploiting data from check-ins

3. Introduction Research questions and facts that paper talks about
Focus on human mobility patterns in a large number of cities across the world
Do people move in a different way in different cities or do they have universal traits across disparate urban centers ?
Try and fit the results for some of the findings into a power law distribution
Analysis Insights
Analysis in favor of concept of intervening opportunities
Probability of transiting from one place to another is inversely proportional to a power of their rank (Intervening opportunities theory)
Can be analogous to finding friendship between two individuals

4. Results – Urban movements and power laws
Data set obtained from four square
Over 35 million movements of about 925K users across 5 million places (2010)
Analysis was performed on distribution of human displacements in two different ways
First dataset was planetary (contained movements up to 20,000kms)
Second dataset was measuring human displacements within cities (dataset of about approximately 10 million users)
Observations
The first distribution is well approximated by a power law and has an exponent value of (beta = 1.5)
Power law does not accurately fit the distribution (beta = 4.67)

5. Results – Urban movements and power laws
Disp. across planet
Disp. across cities

6. Results – Fitting movement across cities
For cities power law distribution fails. How to represent ?
Compare human movements across cities to analyze
PDF of human displacements of 3 cities:
Observation:
Probability of a jump decreases after a threshold of 100 meters, but the shapes of distributions vary from city to city
The threshold could be because a city border might be reached

7. Results – The importance of place density
Study how the density of a city or a place effects human displacements within it
We first plot the density of a city (from check-ins) VS average distance of displacements observed in a number of cities
Observations
Average distance of human movements is inversely proportional to the city’s density
Example, a dense city like New York would have more shorter movements
Not only distance matters but the availability of places at a distance matters ! (Correlation = 0.59)

8. Results – The importance of place density
Density of a city VS Mean human transition

9. Results – The importance of place density
Next Step : Study how geographic size of a city effects human mobility
Plot average transition in a city versus the area size
Observations
Very low Correlation ( = 0.19 )
Density is more informative than area
Insights
Calculating rank value for each movement between a pair of places
The rank densities follow the power law plot
This suggests that probability of moving to a place goes down when the number of places nearer than a potential destination increases

10. Results – The importance of place density
Related Work: A similar finding showing the importance of rank models were observed in a research paper (2005) titled “Geographic Writing in social networks”. They suggested that the probability of observing a user’s friend at a certain distance in a social network is inversely proportional to the number of people geographically closer to the user.
Area of a city VS Mean human transition

11. Results – The importance of place density
Studying density more closely
Related Work: A similar finding showing the importance of rank models were observed in a research paper (2005) titled “Geographic Writing in social networks”. They suggested that the probability of observing a user’s friend at a certain distance in a social network is inversely proportional to the number of people geographically closer to the user.
Plot shows that probability of moving to a place decays
when number of places near the potential destination
Increases.

12. Results – Modeling Urban Mobility
Run agent based simulation experiments using the rank based modeling technique
Observations -
Captures the movements in the city pretty well in spite of not taking in factors like heterogeneity, temporal factors etc.
Rank formula

13. Results – Modeling Urban Mobility

14. Results – The importance of place density
To highlight the importance of density, another study performed
Disp. Differ from city to city
Majorly because spatial distribution of places is different
Run agent based simulators as before but
Set of places are not as previous
They are taken from an empirical data set

15. Results – Modeling Urban Mobility
Spatial Distribution across 3 cities

16. Further Discussion & Similar Work
The paper has performed well in performing empirical validation of past theories on human movements
Stouffer’s theory of intervening opportunities serves as a plausible explanation
Movements analyzed through respective rank values
Similar study on urban dynamics was done in the papers “A unified theory of urban living” and “Growth, innovation, scaling, and the pace of life in cities”
Cities are scaled versions of each other
According to observations density and spatial distribution are vital for mobility

17. Q n A