狠狠撸

Matteo Manca @mattemanca
matteomanca@gmail.com
Addressing Big Societal Challenges with Digital Behavioral Data, London, UK – November 15th, 2017
From Social Science to Computational Social Science:
Is Web Data the Key to a More Effective Analysis?

Traditional social science approaches
? Pros: accurate information about user behaviour (residence, mobility
patterns, and habits in general).
? Cons: high costs, small samples, limited in space and time, updated with
a very low frequency.
Subjects are aware of the fact that they are being studied (this
generates a bias!)
Surveys, interviews:
Addressing Big Societal Challenges with Digital Behavioral Data Workshop @mattemanca
Social Science: Study of social phenomena, i.e.how people interact
to produce collective behavior

New approaches: computational social science
? Shopping
? Economic activities
? Social media activity
? Traveling
? Professional
? Preferences and
Opinions
? Connections
? ...
Social media systems
Smart devices
Digital Behavioral data

Digital Behavioral Data
Opportunity for studying human
behavior and interactions
Challenges:
? How to access to data
? Work with huge amount of data
? How to use this data:
? Ethical concerns
? Privacy
? …

Wireless sensors :
? Pros: wrt surveys no limitation in time and high update frequency
? Cons: high costs due to the installation and management of the sensors,
and the spatial limitation.
Mobile Phone Network :
? Pros: large scale studies, good update frequency, no limitation in time
and space
? Cons: not free and public availability of the data due to privacy, security,
and proprietary reasons.
New sources of data

? Pros: covers all aspects of user behavior and life, no temporal or
spatial limitations, allows large-scale studies, accessible in
(almost) real time
? Cons: Data sampling, might be not fully accessible, etc .
Social Media Data
New sources of data

Digital Behavioral Data Case Study:
Using social media to characterize urban mobility patterns: State-of-the-art survey and case-study

Digital Behavioral Data Case Study
Research Question:
To what extent social media data can be exploited to gain knowledge about urban
dynamics and mobility patterns in a city or in a urban area in general?
Barcelona Case study:
Explore urban mobility
patterns: local citizens vs
tourists.
[Using social media to characterize urban mobility patterns: State-of-the-art survey and case-study.
Matteo Manca, Ludovico Boratto, Victor Morell Roman, Oriol Martori i Gallissà, Andreas
Kaltenbrunner – Online Social Networks and Media (OSNEM)]

Dataset: Tweet published in Barcelona from Jan 01,2015 to Dec 31, 2015
Pre-processing: Data cleaning, filtering and application of a heuristic to classify locals
and tourists;

One-hop paths performed by users in Barcelona:
Shorter paths are visualized
through warmer colors, the
longer a paths the colder
the color tone.

Flow diagrams of locals between districts comparing working days and weekends.
Locals during working days Locals during weekend
● CV: Ciutat Vella
● Ex: Eixample
● Gr: Gràcia
● HG: Horta-Guinardó
● LC: Les Corts
● NB: Nou Barris
● SA: Sant Andreu
● SM: Sant Martí
● Mj: Sants-Montjui?c
● SG: Sarrià-Sant
Gervasi.

Flow diagrams of tourists between districts comparing working days and weekends.
Tourists during working days Tourists during weekend
● CV: Ciutat Vella
● Ex: Eixample
● Gr: Gràcia
● HG: Horta-Guinardó
● LC: Les Corts
● NB: Nou Barris
● SA: Sant Andreu
● SM: Sant Martí
● Mj: Sants-Montjui?c
● SG: Sarrià-Sant
Gervasi.

Analysis of user behavior
Probability that a user visit L locations Frequency-ranking of top-L most frequent tweet locations
for users who tweet from at least for 5, 10, 30 or 50
different locations.

Path statistics for tourists and locals
Boxplots with paths distribution
Cummulative distribution function

● The most central district (like Ciutat Vella) are more visited during weekends to the
detriment of others like Nou Barris.
● Tourists have the same behavior during working days and during weekends.
● Most of the tourists paths involve the two most touristic districts of Barcelona, i.e.,
Ciutat Vella and Eixample.
● Locals are more likely to cover short or long distances, while tourists are more common
to cover intermediate distances.
● In multi-hop paths, the average distance per hop is inversely proportional to the
number of path hops.
● Independently of the number of path hops, tourists are inclined to perform on average
paths that involve longer hops in comparison to those of the locals.
● The probability to find a user in a location with ranking L can be approximated by the
function 1/L.
Case Study Conclusions

Open Issues

Digital Behavioral Data: open issues
Bias in data
Human data collected through the web are biased:
● Gender
● Age
● Technological
● Racial
● …
Example: Social media data or mobile phone data may be biased.
Biased
Input
Algorithm
Biased
Results

Activity bias
[Baza-Yates and Saez-Trumper, Hypertext'15]
Facebook Amazon Review Twitter Wikipedia
Percentage of users creating 50% of content

● Not fully accessible
○ Sampling bias: data retrieved through Twitter’s public streaming APIs cannot
exceed 1% of all tweets being tweeted at a specific moment (hard to estimate
bias).
● Bots, organizations, media are “special” users.
● Privacy issues

Andreas Kaltenbrunner, David Laniado, Ludovico Boratto, Pablo Aragón
Acknowledgement

Matteo Manca @mattemanca
matteomanca@gmail.com
Thank you!

狠狠撸

From Social Science to Computational Social Science: Is Web Data the Key to a More Effective Analysis?

More Related Content

From Social Science to Computational Social Science: Is Web Data the Key to a More Effective Analysis?