Transcript Reality Mining

Inference in Complex Social Systems
Insights and Applications from the Behavior of the Aggregate
Nathan Eagle
Research Scientist / Postdoctoral Fellow / Fulbright Lecturer
Massachusetts Institute of Technology / The Santa Fe Institute / U of Nairobi et al.
Fastest Technology Adoption in Human History

1.15B Phones Purchased in 2007

1.9T Text Messages sent in 2007

Processing Power

A New Era of Wearable Computing…
Data to Science to Engineering
Data
Science
Engineering
Behavioral Data in the 21st Century


12 vs. 250 million people…
Stop-Watch vs. Computational Social Science
1930s
Roethlisberger, F.J., and Dickson, W.J. (1939), Management and the
Worker, Cambridge, MA: Harvard University Press.
2008
Communication Networks of 250M people, 12B edges,…
Behavioral Data in the 21st Century

400,000 continuous hours…
Eagle, N. “Machine Perception and Learning of Complex Social
Systems", PhD Thesis, Massachusetts Institute of Technology. 2005.
Reality Mining Data

100 Nokia 6600s with Context logging software




Location: Celltower ID / User-Defined Names
date, area, cell, network, name
Bluetooth: Proximate Bluetooth Devices every 5 minutes
date, MAC, device name, device type
Communication: Phone Call/Text Log
date, text/call, incoming/outgoing, duration, number
Total Data

Over 400,000 hours continuous human behavior data collected
over the 2004-2005 academic year.
Mika Raento, Antti Oulasvirta, and Nathan Eagle. “Smartphones: an emerging tool for social
scientists”, Sociological Methods and Research (in submission).
BlueDar : Bluetooth Radar

Currently Deployed around MIT


Infinite Corridor, Media Lab, Muddy Charles
Pub, Sloan Business School, Student Center, …
Coming Soon...
Cafeterias
 Elevators
 Gym
 ...

Nathan Eagle, “Can Serendipity Be Planned?”, MIT Sloan Management Review, Vol. 46, No.
1, pp 10-14, 2004.
Outline: Science
Data
Science
Engineering
Bias in Self Report Data
Bias in Self Report Data: Friendships
Bias in Self Report Data: Recency
High-Level Situation Classification

Probabilistic Graphical Models for Data Filtering

Conditioned HMM
Q Home,Work , Elsewhere, No Signal
Ii  Q1 , Q2 , Q3 ,...Qn 
The Entropy of Life

Shannon Information Entropy Applied to Everyday
Life

Estimate of the amount of structure / randomness in a
subject’s routine.
n
H ( I )   p( j ) log 2 p( j )
j 1
Low Entropy Subject, I1
High Entropy Subject, I2
H ( I1 )  30.9
H ( I 2 )  48.5
Behavioral Entropy

Which demographic is the most entropic?

Which demographic is the most “infectious”?
Eigenbehaviors: Transformation

1
N

N
i 1
i
 i  i  
A  1 ; 2 ; 3 ;...N 
2
1 N
k   uk  i 
N i 1
Turk, M., and Pentland, A., "Eigenfaces for Recognition", J. of
Cognitive Neuroscience. Vol 3, Number 1., (1991) 71-86
Individual Behavior Space
u2
2
4
5
3
6
1
1 7
u1
Eigenbehaviors: Behavior Space

u1
Daily human behavior can be clustered in a low
dimensional space.
u2
Nathan Eagle and Alex Pentland. “Eigenbehaviors: Identifying Structure in Routine”,
Proc. Roy. Soc. (in submission).
Eigenbehaviors: Affiliation Inference

u j2
Can we determine a subject’s affiliation based on
short samples of behavior?
j  I j
   i 1 i j uij
j
b
M'
   
2
j
j
j 2
b
I3
I4
 j3
u j2
j
I2
4
I1
 j1
 j2
I2
u j1
Group j Behavior Space
Friendship vs. Proximity Networks
Self-Report Friendship
1-Day Proximity
Relationship Inference

Is it possible to infer friendship based on proximity
behavior?
Nathan Eagle and A. Pentland, “Reality Mining: Sensing Complex Social Systems”, Journal
of Personal and Ubiquitous Computing, Vol. 10 (4), February 2006.
Dyadic Variables
Factor Analysis: “In-Role” v. “Extra-Role”
Table S5: Loadings from a Factor Analysis for Friendship
Variable Name
Specific
Factor1: ‘InVariance
Role’
Communication
Proximity at Work
0.2064
0.9194
Number of Unique Locations
0.1171
0.9927
Proximity with no Reception
0.4749
0.6697
Proximity Outside Work
0.6288
0.3535
Proximity at Home
0.7694
0.0716
Proximity on Saturday Nights
0.6689
-0.1584
Phone Communication
0.6476
-0.1418
Factor 2: ‘ExtraRole’
Communication
-0.0595
-0.1162
0.0990
0.3491
0.4401
0.6387
0.6523
Inferred v. Reported Network
Nathan Eagle, Alex Pentland, and David Lazer. “Inferring Social Network Structure using
Mobile Phone Data”, PNAS (in submission).
Team Proximity Networks
Bi(tj)
1


0

if vertices i and j are ever connected
between time t and t  
otherwise
  1 hour
Network Evolution

Can mobile phone usage reflect an emphasis on
‘networking’ and social network evolution?
Organizational Rhythms

How the deadlines of an institution can be seen in
the collective behavior of its individual members.
168 hrs (7 days)
24 hrs
Outline: Engineering
Data
Science
Engineering
One Day in the Life...

Automatic Diary Generation:

A life log from cell tower IDs
Life Inferences

Class: Sleeping?





Class: Lunch?




Location: {Home}
Phone status: {idle/charging}
Time: {late night / early morning}
Alarm Clock: {interval}
Location: {!= office}
People: {lunch crowd={Mike, Push, Martin}}
Time: {lunchtime}
Class: Partying?



Location: {hang outs ={b-side, sevens, BHP}}
People: {party friends={Mike, Jon, Aisling}}
Time: {evening / late night}
Life Query

AutoDiary
How much sleep did I get last week?
 When was the last time I had lunch with Josh?
 How much time did I spend driving when I was last
in Mountain View?
 Where did I go after leaving Marvin’s house last
week?


Prediction
What are the chances of seeing Mike in the next
hour?
 How likely is it that Caroline will call me tonight?
 Will I be in lab this weekend?

Automatic Diary
MetroSpark
Nathan Eagle and A. Pentland, “Mobile Matchmaking: Proximity Sensing and Cuing”, IEEE
Pervasive Computing, 4 (2): 28-34, 2005.
Nathan Eagle and Alex Pentland, "Combined short range radio network and cellular telephone
network for interpersonal communications." U.S. Patent Application Serial No. 60/568,482.
Filed May 6, 2004. MIT ID: 10705T. Assignee: Massachusetts Institute of Technology.
Talk Takeaways



http://reality.media.mit.edu
Behavior Prediction
Relationship Inference
Computational Social
Science
Questions?
Nathan Eagle
[email protected]