NYC Data Science Meetup: Computational Social Science

Computational Social Science
Jake Hofman
Microsoft Research
November 6, 2014
@jakehofman (Microsoft Research) Computational Social Science November 6, 2014 1 / 62

MSR NYC
http://research.microsoft.com/en-us/labs/newyork/

Questions
Many long-standing questions in the social sciences are notoriously
difficult to answer, e.g.:
• “Who says what to whom in what channel with what e↵ect”?
(Laswell, 1948)
• How do ideas and technology spread through cultures?
(Rogers, 1962)
• How do new forms of communication a↵ect society?
(Singer, 1970)
• . . .

Conventional methods
Typically difficult to observe the relevant information via
conventional methods
(Katz & Lazarsfeld, 1955)

Large-scale data
Recently available electronic data provide an unprecedented
opportunity to address these questions at scale
Demographic Behavioral Network

Computational social science
An emerging discipline at the intersection of the social sciences,
statistics, and computer science

(motivating questions)

(fitting large, potentially sparse models)

(parallel processing for filtering and aggregating data)

possible to estimate accurately the age of ori-gin
of almost all extant genera. It is then possi-ble
to plot a backward survivorship curve (8)
for each of the 27 global bivalve provinces (9).
On the basis of these curves, Krug et al. find
that origination rates of marine bivalves in-creased
biodiversification event in the Paleogene (65 to
23 million years ago) that is perhaps not yet
captured in Alroy et al.’s database (5, 7). The
jury is still out on what may have caused this
event. But we should not lose sight of the fact
that the steep rise to prominence of many mod-ern
8. M. Foote, in Evolutionary Patterns, J. B. C. Jackson et al.,
Eds. (Univ. of Chicago Press, Chicago, IL, 2001), vol. 245,
pp. 245–295.
9. M. D. Spalding et al., Bioscience 57, 573 (2007).
10. S. M. Stanley, Paleobiology 33, 1 (2007).
11. M. J. Benton, B. C. Emerson, Palaeontology 50, 23 (2007).
10.1126/science.1169410
SOCIAL SCIENCE
We live life in the network. We check
our e-mails regularly, make mobile
phone calls from almost any loca-tion,
swipe transit cards to use public trans-portation,
and make purchases with credit
cards. Our movements in public places may be
captured by video cameras, and our medical
records stored as digital files. We may post blog
entries accessible to anyone, or maintain friend-ships
through online social networks. Each of
these transactions leaves digital traces that can
be compiled into comprehensive pictures of
both individual and group behavior, with the
potential to transform our understanding of our
lives, organizations, and societies.
The capacity to collect and analyze massive
amounts of data has transformed such fields as
biology and physics. But the emergence of a
data-driven “computational social science” has
been much slower. Leading journals in eco-nomics,
sociology, and political science show
little evidence of this field. But computational
social science is occurring—in Internet compa-nies
such as Google and Yahoo, and in govern-ment
agencies such as the U.S. National Secur-ity
Agency. Computational social science could
become the exclusive domain of private com-panies
and government agencies. Alternatively,
there might emerge a privileged set of aca-demic
researchers presiding over private data
from which they produce papers that cannot be
A field is emerging that leverages the
capacity to collect and analyze data at a
scale that may reveal patterns of individual
and group behaviors.
critiqued or replicated. Neither scenario will
serve the long-term public interest of accumu-lating,
verifying, and disseminating knowledge.
What value might a computational social
science—based in an open academic environ-ment—
offer society, by enhancing understand-ing
of individuals and collectives? What are the
David Lazer,1 Alex Pentland,2 Lada Adamic,3 Sinan Aral,2,4 Albert-László Barabási,5
Devon Brewer,6 Nicholas Christakis,1 Noshir Contractor,7 James Fowler,8 Myron Gutmann,3
Tony Jebara,9 Gary King,1 Michael Macy,10 Deb Roy,2 Marshall Van Alstyne2,11
“... a computational social science is emerging that
leverages the capacity to collect and analyze data with an
unprecedented breadth and depth and scale ...”
http://sciencemag.org/content/323/5915/721

possible to estimate accurately the age of ori-gin
of almost all extant genera. It is then possi-ble
to plot a backward survivorship curve (8)
for each of the 27 global bivalve provinces (9).
On the basis of these curves, Krug et al. find
that origination rates of marine bivalves in-creased
biodiversification event in the Paleogene (65 to
23 million years ago) that is perhaps not yet
captured in Alroy et al.’s database (5, 7). The
jury is still out on what may have caused this
event. But we should not lose sight of the fact
that the steep rise to prominence of many mod-ern
8. M. Foote, in Evolutionary Patterns, J. B. C. Jackson et al.,
Eds. (Univ. of Chicago Press, Chicago, IL, 2001), vol. 245,
pp. 245–295.
9. M. D. Spalding et al., Bioscience 57, 573 (2007).
10. S. M. Stanley, Paleobiology 33, 1 (2007).
11. M. J. Benton, B. C. Emerson, Palaeontology 50, 23 (2007).
10.1126/science.1169410
SOCIAL SCIENCE
We live life in the network. We check
our e-mails regularly, make mobile
phone calls from almost any loca-tion,
swipe transit cards to use public trans-portation,
and make purchases with credit
cards. Our movements in public places may be
captured by video cameras, and our medical
records stored as digital files. We may post blog
entries accessible to anyone, or maintain friend-ships
through online social networks. Each of
these transactions leaves digital traces that can
be compiled into comprehensive pictures of
both individual and group behavior, with the
potential to transform our understanding of our
lives, organizations, and societies.
The capacity to collect and analyze massive
amounts of data has transformed such fields as
biology and physics. But the emergence of a
data-driven “computational social science” has
been much slower. Leading journals in eco-nomics,
sociology, and political science show
little evidence of this field. But computational
social science is occurring—in Internet compa-nies
such as Google and Yahoo, and in govern-ment
agencies such as the U.S. National Secur-ity
Agency. Computational social science could
become the exclusive domain of private com-panies
and government agencies. Alternatively,
there might emerge a privileged set of aca-demic
researchers presiding over private data
from which they produce papers that cannot be
A field is emerging that leverages the
capacity to collect and analyze data at a
scale that may reveal patterns of individual
and group behaviors.
critiqued or replicated. Neither scenario will
serve the long-term public interest of accumu-lating,
verifying, and disseminating knowledge.
What value might a computational social
science—based in an open academic environ-ment—
offer society, by enhancing understand-ing
of individuals and collectives? What are the
David Lazer,1 Alex Pentland,2 Lada Adamic,3 Sinan Aral,2,4 Albert-László Barabási,5
Devon Brewer,6 Nicholas Christakis,1 Noshir Contractor,7 James Fowler,8 Myron Gutmann,3
Tony Jebara,9 Gary King,1 Michael Macy,10 Deb Roy,2 Marshall Van Alstyne2,11
“... shares with other nascent interdisciplinary fields
(e.g., sustainability science) the need to develop a
paradigm for training new scholars ...”
http://sciencemag.org/content/323/5915/721

The clean real story
“We have a habit in writing articles published in
scientific journals to make the work as finished as
possible, to cover all the tracks, to not worry about the
blind alleys or to describe how you had the wrong idea
first, and so on. So there isn’t any place to publish, in
a dignified manner, what you actually did in order to
get to do the work ...”
-Richard Feynman
Nobel Lecture1, 1965
1http://bit.ly/feynmannobel

Outline
Search predictions "Right Round"
Week
Rank
10
20
30
40
c
Billboard
Search
Mar−09 Apr−09 May−09 Jun−09 Jul−09 Aug−09
Web diversity
Daily Per−Capita Pageviews
70
60
50
40
30
20
10
0
Under $25k
●
White
●
Some College
●
Under 65
●
●
Over $25k
Black
&
Hispanic
No College
Over 65
Female
Male
Income Race Education Age Sex
Information di↵usion

Predicting consumer activity with Web search
with Sharad Goel, S´ebastien Lahaie, David Pennock, Duncan Watts
"Right Round"
Week
Rank
10
20
30
40
c
Billboard
Search

Search predictions
Motivation
Does collective search activity
provide useful predictive signal
about real-world outcomes?
"Right Round"
Week
Rank
10
20
30
40
c
Billboard
Search

Search predictions
Motivation
Past work mainly focuses on predicting the present2 and ignores
baseline models trained on publicly available data
Date
Flu Level (Percent)
8
7
6
5
4
3
2
1
Actual
Search
Autoregressive
2004 2005 2006 2007 2008 2009 2010
2Varian, 2009

Search predictions
Motivation
We predict future sales for movies, video games, and music
"Transformers 2"
Time to Release (Days)
Search Volume
a
−30 −20 −10 0 10 20 30
"Tom Clancy's HAWX"
Time to Release (Days)
Search Volume
b
−30 −20 −10 0 10 20 30
"Right Round"
Week
Rank
10
20
30
40
c
Billboard
Search

Search predictions
Search models
For movies and video games, predict opening weekend box office
and first month sales, respectively:
log(revenue) = !0 + !1 log(search) + ✏
For music, predict following week’s Billboard Hot 100 rank:
billboardt+1 = !0 + !1searcht + !2searcht−1 + ✏

Search predictions
Search volume

Search predictions
Search models
Search activity is predictive for movies, video games, and music
weeks to months in advance
Movies
●
●
● ●
●
●
●
●
●
●
Predicted Revenue (Dollars)
Actual Revenue (Dollars)
109
108
107
106
105
104
103
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
a
103 104 105 106 107 108 109
Video Games
●
● Non−Sequel
Sequel
107
106
105
104
103
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
b
103 104 105 106 107
Music
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Predicted Billboard Rank
Actual Billboard Rank
100
80
60
40
20
0
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
c
0 20 40 60 80 100
Movies
Time to Release (Weeks)
Model Fit
0.9 d
0.8
0.7
0.6
0.5
0.4
−6 −5 −4 −3 −2 −1 0
Video Games
Model Fit
0.9 e
0.8
0.7
0.6
0.5
0.4
−6 −5 −4 −3 −2 −1 0
Music
Model Fit
0.9 f
0.8
0.7
0.6
0.5
0.4
−6 −5 −4 −3 −2 −1 0

Search predictions
Baseline models
For movies, use budget, number of opening screens and Hollywood
Stock Exchange:
log(revenue) = !0 + !1 log(budget) + !2 log(screens) +
!3 log(hsx) + ✏

Search predictions
Baseline models
For video games, use critic ratings and predecessor sales (sequels
only):
log(revenue) = !0 + !1rating + !2 log(predecessor) + ✏

Search predictions
Baseline models
For music, use an autoregressive model with the previously
available rank:
billboardt+1 = !0 + !1billboardt−1 + ✏

Search predictions
Baseline + combined models
Baseline models are often surprisingly good
Movies (Baseline)
●
●
●
●
●
●
●
●
●
109
108
107
106
105
104
103
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
a
103 104 105 106 107 108 109
Video Games (Baseline)
●
●
●
●
● Non−Sequel
Sequel
107
106
105
104
103
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
b
103 104 105 106 107
Music (Baseline)
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
100
80
60
40
20
0
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
c
0 20 40 60 80 100
Movies (Combined)
●
●
●
109
108
107
106
105
104
103
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
● ●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
d
103 104 105 106 107 108 109
Video Games (Combined)
●
●
●
●
●●
● Non−Sequel
Sequel
107
106
105
104
103
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
e
103 104 105 106 107
Music (Combined)
●
●
●
●
●
●
●
●
●
●
●
●
100
80
60
40
20
0
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
f
0 20 40 60 80 100

Search predictions
Model comparison
For movies, search is outperformed by the baseline and of little
marginal value
Model Fit
1.0
0.9
0.8
0.7
0.6
0.5
0.4
Combined
Search
Baseline
Nonsequel Games
Sequel Games
Music
Movies
Flu

Search predictions
Model comparison
For video games, search helps substantially for non-sequels, less so
for sequels
Model Fit
1.0
0.9
0.8
0.7
0.6
0.5
0.4
Combined
Search
Baseline
Nonsequel Games
Sequel Games
Music
Movies
Flu

Search predictions
Model comparison
For music, the addition of search yields a substantially better
combined model
Model Fit
1.0
0.9
0.8
0.7
0.6
0.5
0.4
Combined
Search
Baseline
Nonsequel Games
Sequel Games
Music
Movies
Flu

Search predictions
Summary
• Relative performance and value of search varies across
domains
• Search provides a fast, convenient, and flexible signal across
domains
• “Predicting consumer activity with Web search”
Goel, Hofman, Lahaie, Pennock & Watts, PNAS 2010

Outline
Search predictions "Right Round"
Week
Rank
10
20
30
40
c
Billboard
Search
Web diversity
70
60
50
40
30
20
10
0
Under $25k
●
White
●
Some College
●
Under 65
●
●
Over $25k
Black
&
Hispanic
No College
Over 65
Female
Male

Demographic diversity on the Web
with Irmak Sirer and Sharad Goel (ICWSM 2012)
70
60
50
40
30
20
10
0
Under $25k
●
White
●
Some College
●
Under 65
●
●
Over $25k
Black
&
Hispanic
No College
Over 65
Female
Male

Motivation
Previous work is largely survey-based and focuses and group-level
di↵erences in online access

Motivation
“As of January 1997, we estimate that 5.2 million
African Americans and 40.8 million whites have ever used
the Web, and that 1.4 million African Americans and
20.3 million whites used the Web in the past week.”
-Ho↵man & Novak (1998)

Motivation
Focus on activity instead of access
How diverse is the Web?
To what extent do online experiences vary across demographic
groups?

Data
• Representative sample of 265,000 individuals in the US, paid
via the Nielsen MegaPanel3
• Log of anonymized, complete browsing activity from June
2009 through May 2010 (URLs viewed, timestamps, etc.)
• Detailed individual and household demographic information
(age, education, income, race, sex, etc.)
3Special thanks to Mainak Mazumdar

Data
# ls -alh nielsen_megapanel.tar
-rw-r--r-- 100G Jul 17 13:00 nielsen_megapanel.tar

Data
• Normalize pageviews to at most three domain levels, sans www
e.g. www.yahoo.com ! yahoo.com,
us.mg2.mail.yahoo.com/neo/launch ! mail.yahoo.com

Data
• Restrict to top 100k (out of 9M+ total) most popular sites
(by unique visitors)

Data
• Restrict to top 100k (out of 9M+ total) most popular sites
(by unique visitors)
• Aggregate activity at the site, group, and user levels

Aggregate usage patterns
How do users distribute their time across di↵erent categories?
Fraction of total pageviews
0.25
0.20
0.15
0.10
0.05
●
●
●
● ●
Social Media
E−mail
Games
Portals
Search
All groups spend the majority of their time in the top five most
popular categories

Aggregate usage patterns
How do users distribute their time across di↵erent categories?
● Social Media
E−mail
Games
Portals
Search
User Rank by Daily Activity
Fraction of Pageviews in Category
0.30
0.25
0.20
0.15
0.10
0.05
●
● ● ● ●
●
●
●
●
●
10% 30% 50% 70% 90%
Highly active users devote nearly twice as much of their time to
social media relative to typical individuals

Group-level activity
How does browsing activity vary at the group level?
70
60
50
40
30
20
10
0
Under $25k
●
White
●
Some College
●
Under 65
●
●
Over $25k
Black
&
Hispanic
No College
Over 65
Female
Male
Large di↵erences exist even at the aggregate level
(e.g. women on average generate 40% more pageviews than men)

How does browsing activity vary at the group level?
70
60
50
40
30
20
10
0
Under $25k
●
White
●
Some College
●
Under 65
●
●
Over $25k
Black
&
Hispanic
No College
Over 65
Female
Male
Younger and more educated individuals are both more likely to
access the Web and more active once they do

All demographic groups spend the majority of their time in the
same categories
Age
0.5
0.4
0.3
0.2
0.1
0.0
●
●
●
●
●
●
● ●
●
●
●
●
●
●
● Social Media
E−mail
Games
Portals
Search
● ●
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80
0.4
0.3
0.2
0.1
0.0
Education
● ●
●
●
●
●
●
High School Graduate
Some High School
Grammar School
Some College
Associate Degree
Post Graduate Degree
Bachelor's Degree
Sex
●
●
Female
Male
Income
●
● ●
●
●
●
$0−25k
$75−100k
$25−50k
$50−75k
$100−150k
$150k+
Race
● ●
● ●
●
Other
Hispanic
Black
White
Asian

Older, more educated, male, wealthier, and Asian Internet users
spend a smaller fraction of their time on social media
Age
0.5
0.4
0.3
0.2
0.1
0.0
●
●
●
●
●
●
● ●
●
●
●
●
●
●
● Social Media
E−mail
Games
Portals
Search
● ●
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80
0.4
0.3
0.2
0.1
0.0
Education
● ●
●
●
●
●
●
Some High School
Grammar School
Some College
Associate Degree
Bachelor's Degree
Sex
●
●
Female
Male
Income
●
● ●
●
●
●
$0−25k
$75−100k
$25−50k
$50−75k
$100−150k
$150k+
Race
● ●
● ●
●
Other
Hispanic
Black
White
Asian

Lower social media use by these groups is often accompanied by
higher e-mail volume
Age
0.5
0.4
0.3
0.2
0.1
0.0
●
●
●
●
●
●
● ●
●
●
●
●
●
●
● Social Media
E−mail
Games
Portals
Search
● ●
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80
0.4
0.3
0.2
0.1
0.0
Education
● ●
●
●
●
●
●
Some High School
Grammar School
Some College
Associate Degree
Bachelor's Degree
Sex
●
●
Female
Male
Income
●
● ●
●
●
●
$0−25k
$75−100k
$25−50k
$50−75k
$100−150k
$150k+
Race
● ●
● ●
●
Other
Hispanic
Black
White
Asian

Female−to−male pageview ratio
2
1
0.5
● ●
●
●
●
● ● ● ● ●
● ● ●●● ●
● ● ● ●
● ● ●●●●●● ● ● ● ● ● ●● ● ● ● ● ●●● ● ● ●●●
● ● ● ● ●
●●● ●
● ● ● ● ● ● ● ● ● ●● ●
● ●
● ● ● ●●
●
● ● ●
●
●
●
●
●
●
●
Pets
Home & Garden
Food & Cooking
Apparel/Beauty
Family Resources
Holidays & Special Events
Health, Fitness & Nutrition
Multi−category Home & Fashion
Photography
Non−Profit
Genealogy
Universities
Greeting Cards
Cruise Lines
Government
Online Games
Directories/Local Guides
Gifts & Flowers
Corporate Information
Real Estate/Apartments
Mass Merchandiser
Multi−category Family & Lifestyles
Multi−category Special Occasions
Books
Member Communities
Educational Resources
Shopping Directories & Guides
E−mail
Insurance
Cellular/Paging
Coupons/Rewards
Kids, Games, Toys
Loans
Arts/Graphics
Destinations
Multi−category Travel
Broadcast Media
Religion & Spirituality
Full Service Banks & Credit Unions
Credit Card
Software Manufacturers
General Interest Portals & Communities
Multi−category Telecom/Internet Services
Full Service Commercial Banks & Credit Unions
Financial Tools
Classifieds/Auctions
Maps/Travel Info
Delivery/Stamps
Search
Career Development
Hotels/Hotel Directories
Airlines
Instant Messaging
Ground Transportation
Free Merchandise
Multi−category Entertainment
Long Distance/Local Carrier
Events
ISP
Music
Research Tools
Gambling/Sweepstakes
Special Interest News
Current Events & Global News
Multi−category News & Information
Multi−category Finance/Insurance/Investments
Weather
Web Hosting
Videos/Movies
Parts & Accessories
Financial News & Information
Automotive Manufacturer
Internet Tools/Web Services
Military
Multi−category Automotive
Automotive Information
Hardware Manufacturers
Targeted Portals & Communities
Multi−Category Education & Careers
Computer & Consumer Electronics News
Multi−category Computers & Consumer Electronics
Sports
Adult
Humor
Personals
Online Trading

Revisiting the digital divide
How does usage of news, health, and reference vary with
demographics?
Average pageviews per month
12
10
8
6
4
2
0
Education
●
●
●
● ●
●
●
Some High School
Grammar School
Some College
Associate Degree
Bachelor's Degree
Sex
●
●
Female
Male
Income
● ● ●
●
●
●
$0−25k
$75−100k
$25−50k
$50−75k
$100−150k
$150k+
Race
● ●
●
●
●
Other
Hispanic
Black
White
Asian
● News
Health
Reference
Post-graduates spend three times as much time on health sites
than adults with only some high school education

demographics?
12
10
8
6
4
2
0
Education
●
●
●
● ●
●
●
Some High School
Grammar School
Some College
Associate Degree
Bachelor's Degree
Sex
●
●
Female
Male
Income
● ● ●
●
●
●
$0−25k
$75−100k
$25−50k
$50−75k
$100−150k
$150k+
Race
● ●
●
●
●
Other
Hispanic
Black
White
Asian
● News
Health
Reference
Asians spend more than 50% more time browsing online news than
do other race groups

demographics?
12
10
8
6
4
2
0
Education
●
●
●
● ●
●
●
Some High School
Grammar School
Some College
Associate Degree
Bachelor's Degree
Sex
●
●
Female
Male
Income
● ● ●
●
●
●
$0−25k
$75−100k
$25−50k
$50−75k
$100−150k
$150k+
Race
● ●
●
●
●
Other
Hispanic
Black
White
Asian
● News
Health
Reference
Even when less educated and less wealthy groups gain access to
the Web, they utilize these resources relatively infrequently

demographics?
12
10
8
6
4
2
0
News
●
● ●
●
●
Some College
Associate Degree
Bachelor's Degree
Health
●
● ●
● ●
Some College
Associate Degree
Bachelor's Degree
Reference
●
● ●
● ●
Some College
Associate Degree
Bachelor's Degree
Asian
Black
Hispanic
White
Controlling for other variables, e↵ects of race and gender largely
disappear, while education continues to have large e↵ect
pi =
X
j
↵jxij +
X
j
X
k
!jkxijxik +
X
j
$jx2
ij + ✏i

demographics?
12
10
8
6
4
2
0
Health
●
● ●
● ●
Associate Degree
Some College
Bachelor's Degree
Female
Male
However, women spend considerably more time on health sites
compared to men

demographics?
Female
Male
20 40 60 80 100
Monthly pageviews on health sites
However, women spend considerably more time on health sites
compared to men, although means can be misleading

Individual-level prediction
How well can one predict an individual’s demographics from their
browsing activity?
• Represent each user by the set of sites visited
• Fit linear models4 to predict majority/minority for each
attribute on 80% of users
• Tune model parameters using a 10% validation set
• Evaluate final performance on held-out 10% test set
4http://bit.ly/svmperf

Reasonable (⇠70-85%) accuracy and AUC across all attributes
Over/Under 25
Years Old
Female/Male
White/Non−White
Under/Over $50,000
Household Income
College/No College
Accuracy
●
●
●
●
●
.5 .6 .7 .8 .9 1
AUC
●
●
●
●
●
.5 .6 .7 .8 .9 1

Highly-weighted sites under the fitted models
Large positive weight Large negative weight
Female
winster.com
lancome-usa.com
sports.yahoo.com
espn.go.com
White
marlboro.com
cmt.com
mediatakeout.com
bet.com
College Educated
news.yahoo.com
linkedin.com
youtube.com
myspace.com
Over 25 Years Old
evite.com
classmates.com
addictinggames.com
youtube.com
Household Income
Under $50,000
eharmony.com
tracfone.com
rownine.com
matrixdirect.com
Table 2: A selection of the most predictive (i.e., most highly weighted) sites for each classification task.
Over/Under 25
Years Old
Female/Male
White/Non−White
Under/Over $50,000
Household Income
College/No College
AUC
!
!
!
!
!
.5 .6 .7 .8 .9 1
Accuracy
!
!
!
!
!
.5 .6 .7 .8 .9 1
Figure 7, a measure that eectively re-normalizes the ma-jority
and minority classes to have equal size. Intuitively,
AUC is the probability that a model scores a randomly se-lected
positive example higher than a randomly selected neg-ative
one (e.g., the probability that the model correctly dis-tinguishes
between a randomly selected female and male).
Though an uninformative rule would correctly discriminate
between such pairs 50% of the time, predictions based on
browsing histories are relatively reliable, ranging from 74%

Substantially better performance when restricted to “stereotypical”
users (⇠80-90%)
● Age
Sex
Race
Education
Income
Fraction of Users
AUC
0.95
0.90
0.85
0.80
0.75
0.70
● ● ●
●
●
●
●
0.0 0.2 0.4 0.6 0.8 1.0
● Age
Sex
Race
Education
Income
Fraction of Users
Accuracy
0.95
0.90
0.85
0.80
0.75
0.70
● ● ●●
●
●
●
0.0 0.2 0.4 0.6 0.8 1.0

Similar performance even when restricted to top 1k sites
Number of Domains
AUC
0.9
0.8
0.7
0.6
0.5
●
●
● ●
● Age
Sex
Race
Education
Income
102 102.5 103 103.5 104 104.5 105
Number of Domains
Accuracy
0.9
0.8
0.7
0.6
0.5
●
●
● ●
● Age
Sex
Race
Education
Income
102 102.5 103 103.5 104 104.5 105

Site-level skew
Proportion Female Visitors
Density
0.0 0.2 0.4 0.6 0.8 1.0
Proportion White Visitors
Density
0.0 0.2 0.4 0.6 0.8 1.0
Proportion College Educated Visitors
Density
0.0 0.2 0.4 0.6 0.8 1.0
Proportion Adult Visitors
Density
0.0 0.2 0.4 0.6 0.8 1.0 Proportion of Visitors With
Household Incomes Under $50,000
Density
0.0 0.2 0.4 0.6 0.8 1.0
Many sites have skew close the overall mean, but there also
popular, highly-skewed sites

Proof of concept browser demo
http://bit.ly/surfpreds

Summary
• Highly active users spend disproportionately more of their
time on social media and less on e-mail relative to the overall
population
• Access to research, news, and healthcare is strongly related to
education, not as closely to ethnicity
• User demographics can be inferred from browsing activity with
reasonable accuracy
• “Who Does What on the Web”, Goel, Hofman Sirer,
ICWSM 2012

Outline
Search predictions Right Round
Week
Rank
10
20
30
40
c
Billboard
Search
Web diversity
70
60
50
40
30
20
10
0
Under $25k
●
White
●
Some College
●
Under 65
●
●
Over $25k
Black

Hispanic
No College
Over 65
Female
Male

The structual virality of online di↵usion
with Ashton Anderson, Sharad Goel, Duncan Watts (201?)

“Going Viral”?

“Going Viral”?
“Therefore we ... wish to proceed with great care as is
proper, and to cut o↵ the advance of this plague and
cancerous disease so it will not spread any further ...”5
-Pope Leo X
Exsurge Domine (1520)
5http://www.economist.com/node/21541719

“Going Viral”?
Rogers (1962), Bass (1969)

“Going viral”?

Data
• Examined one year of tweets from July 2011 to July 2012

Data
• Restricted to 1.4 billion tweets containing links to top news,
videos, images, and petitions sites

Data
• Aggregated tweets by URL, resulting in 1 billion distinct
“events”

Data
“events”
• Crawled friend list of each adopter

Data
“events”
• Inferred “who got what from whom” to construct di↵usion
trees

Data
“events”
• Inferred “who got what from whom” to construct di↵usion
trees
• Characterized size and structure of trees

The Structural Virality of Online Di↵usion
A
B
D
C
E
Time
Group posts by URL

A
B
D
C
E
Time
Label each friend who previously adopted as a potential parent

A
B
D
C
E
Time
Select each node’s most recent adopting friend as its parent

A
B
D
C
E
Generations
Characterize size and structure of trees

Cascade size distribution
10%
1%
0.1%
0.01%
0.001%
0.0001%
0.00001%
1 10 100 1,000 10,000
Cascade Size
CCDF
Focus on the rare hits that get at least 100 adoptions

Quantifying structure
Measure the average distance between all pairs of nodes6
⌫(T) =
1
n(n − 1)
Xn
i=1
Xn
j=1
dij
6Weiner (1947); correlated with other possible metrics

Quantifying structure
Measure the average distance between all pairs of nodes6
⌫(T) =
2n
n − 1

1
n
X
S2S
|S|−
1
n2
X
S2S
|S|2
#
6Weiner (1947); correlated with other possible metrics

Size and virality by category
Remarkable structural diversity across across categories
100%
10%
1%
0.1%
0.01%
0.001%
100 1,000 10,000
Cascade Size
CCDF
Videos
Pictures
News
Petitions
100%
10%
1%
0.1%
0.01%
0.001%
3 10 30
Structural Virality
CCDF
Videos
Pictures
News
Petitions

Structural diversity

Structural diversity
Size is relatively poor predictive of structure
Petitions News Pictures Videos
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ● ●● ●●● ●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
● ●
●●
● ●
● ●
●
●
●●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
● ●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
● ●
● ●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
● ●
●
●
● ●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
● ●
● ●
● ● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
● ● ●
●
●
●
●
●
● ●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
● ●
●
●
●
● ●
●
●
●
●
●
●
●
● ●
●
●
●●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●●●
●●
●
●
●
● ●●
● ●
●
●
●
●
●●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●●●
●
● ●
●
●
●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
30
10
3
100
300
1,000
100
300
1,000
3,000
100
300
1,000
3,000
10,000
100
300
1,000
3,000
10,000
Cascade size
Structural virality

Simulations
Simulate cascades with a simple SIR model7,
varying infectivity and degree skew 662 CHAPTER 21. EPIDEMICS
t
y
x z
u
r v
w
s
(a)
t
y
x z
u
r v
w
s
(b)
t
y
x z
u
r v
w
s
(c)
t
y
x z
u
r v
w
s
(d)
8
Figure 21.2: The course of an SIR epidemic in which each node remains infectious for a
number of steps equal to tI = 1. Starting with nodes y and z initially infected, the epidemic
spreads to some but not all of the remaining nodes. In each step, shaded nodes with dark
borders are in the Infectious (I) state and shaded nodes with thin borders are in the Removed
(R) state.
7Kermack McKendrick (1927)
8Easley Kleinberg (2010)

Simulations
This reproduces the observed marginal distributions of size and
structure
100
30
10
3
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
● ●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
● ●
●
●
●
●
●
● ●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
● ●
●
●
●
● ●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●● ●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
100
300
1,000
3,000
100,000
10,000
30,000
Cascade size
Structural virality
... but fails to account for the variance in structure given size

Summary
• Most cascades fail, resulting in fewer than two adoptions, on
average
• Of the hits that do succeed, we observe a wide range of
diverse di↵usion structures
• It’s difficult to say how something spread given only its
popularity
• “The structural virality of online di↵usion”, Anderson, Goel,
Hofman Watts (Under review.)

Outline
Search predictions Right Round
Week
Rank
10
20
30
40
c
Billboard
Search
Web diversity
70
60
50
40
30
20
10
0
Under $25k
●
White
●
Some College
●
Under 65
●
●
Over $25k
Black

Hispanic
No College
Over 65
Female
Male

Conclusion

Lessons learned
Data jeopardy
Regardless of scale, it’s difficult to find the right questions to ask
of the data

Lessons learned
Hacking
Cleaning and normalizing data is a substantial amount of the work

Lessons learned
Modeling
Understanding human activity is often useful for detecting
malicious activity

Lessons learned
Modeling
Simple methods (e.g., linear models) work surprisingly well,
especially with lots of (diverse) data

Thanks. Questions?
jake@jakehofman.com
Also, we’re hiring:
bit.ly/msrnyc_appsci
bit.ly/msrnyc_eng

NYC Data Science Meetup: Computational Social Science

More Related Content

What's hot

Viewers also liked

Similar to NYC Data Science Meetup: Computational Social Science

More from jakehofman

Recently uploaded

NYC Data Science Meetup: Computational Social Science