mobile phones & social networks Jari Saramki Dept. of - - PowerPoint PPT Presentation

mobile phones & social networks

Jari Saramäki

• Dept. of Biomedical Engineering & Computational Science

Aalto University Finland

EIB Seminar, Feb 11, 2014, Luxembourg

• The basic need to form & maintain social relationships

has been the largest technological driver for a few decades!

• Still new technologies take everyone by surprise

(including their providers!)

Why Study Social Networks?

1988 1995 2004

1

Big Data & Computational Social Science

• Massive electronic records:
• Mobile operators: calls, text messages, WiFi, ...
• Online social networks: Facebook, Twitter, ...
• Online purchasing and browsing histories
• Allow studies of human behaviour
• n an unprecedented scale

From Calls to Network Analysis

apply mathematical & computational tools to understand network structure, its evolution, and its effects on dynamical processes

• n networks

• ut-star

call detail records

• perator’s billing system

anonymized network data

network: people linked if they have called

• ne another

call network analysis modelling pattern detection

• inhomogeneity: different social habits
• homophily: similar people like to connect
• assortativity: highly connected people like

to connect

• group structure: circles of friendship
• limited personal network size:

time and cognitive constraints

Features of human social networks

This talk

Why Study Social Networks?

1

Weak Ties, Strong Ties

2

Persistent Social Habits

3

Patterns of Conversation

4

Weak Ties, Strong Ties

2

The weak ties hypothesis

5 min 8 min 7 min 20 min

1) 2) 3)

• The weak ties hypothesis (Granovetter 1973):

The relative overlap of two individual’s friendship networks varies directly with the strength of their tie to one another.

• We used anonymized call records to investigate

this

• Call data for 7 million people over 18 weeks
• network:
• nodes = people,
• two people linked if mutual calls,
• tie strength = total duration of calls

• Define the overlap Oij of

a link as the fraction of common friends

• Calculate average overlap

as a function of tie strength

• Increasing tendency
• bserved
• hypothesis verified

Verification of the Weak Tie Hypothesis

Onnela, Saramäki, et al.,

• Proc. Natl. Acad. Sci. (USA) 104, 7332 (2007),

New Journal of Physics 9, 179 (2007)

tie strength = duration of calls tie strength = number of calls

Weak ties are crucial for connectivity!

small network sample 80% of strongest links removed 80% of weakest links removed

diluted fragmented

Onnela, Saramäki, et al.,

• Proc. Natl. Acad. Sci. (USA) 104, 7332 (2007),

New Journal of Physics 9, 179 (2007)

Weak ties act as bottlenecks for information diffusion

Onnela, Saramäki, et al.,

• Proc. Natl. Acad. Sci. (USA) 104, 7332 (2007),

New Journal of Physics 9, 179 (2007)

Persistent Social Habits

3

Limits to numbers of relationships

time: we only have so much of it! brain power: the same applies!

Networks in flux

• Data on 24 volunteer students
• Students finished high school &

went to university

• All outgoing calls for 18 months
• Three social surveys

105 57 56 35 13 8 6 4 3 2

ego A B C D E F G

H

I J

a)

rank

2 4 6 8 10 10−3 10−2 10−1 100

A B C D E F G H I J

fraction of calls b)

1) count calls to everyone in a 6-month interval 2) rank everyone, see what % of calls goes to #1, what % to #2, etc

Social signatures

very large numbers of calls concentrated at top ranks top 3: males 40% of calls females 48% of calls signatures do not change

• ver time

Average signatures

2 10 15 5 g u A

• r

a M : I 10-3 10-2 10-1 100 fraction ) a

1

2 10 15 5 b e F

• p

e S : I 10-3 10-2 10-1 100 ) b

2

2 10 15 5 g u A

• r

a M : I alter I alter I alter I kin 10-3 10-2 10-1 100 ) c

3 1 2 3

• Communication mainly with a small

number of others

• Very persistent pattern, even when

friends are replaced by newcomers

• Individual-level persistence:

“If you like to have two best friends, this never changes, irrespectively of who those friends are”

One in, one out

average for all 24 students

Patterns of Conversation

4

Call & text message sequences

call sms

green lines = social ties

Burstiness: universal feature of human dynamics

all calls by one person calls to each friend

time

Karsai et al, Phys. Rev. E 83, 025102(R) (2011)

correlated call sequences

temporal motifs

• We want to detect temporal patterns, where links

activate within short time periods

• Patterns should be grouped into equivalence classes (motifs)

based on the order of events (calls in this case)

• Study pattern frequency vs properties of involved nodes

B C

1.,2. 1. 2. 2. 1. 1. 2. 1. 2. 2. 1.

repeated contact returned contact causal chain non-causal chain

• ut-star

in-star t=1 t=3 t=4 t=8

d a b c

t=1 t=3 t=4

a b c

1. 2. 3. t=1, t=2, t=7

b a

t=1, t=2

b a

1., 2.

event sequence temporal subgraph (t=3) temporal motif

Kovanen, Kaski, Kertész, Saramäki, PNAS 2013

temporal motifs in call sequences

• there is temporal homophily:

patterns where participants are similar (age, gender) are

• verexpressed

Kovanen, Kaski, Kertész, Saramäki, PNAS 2013

temporal motifs in call sequences

• there is temporal homophily:

patterns where participants are similar (age, gender) are

• verexpressed
• there are gender differences:

females: chains & stars males: “ping-pong”

2 . 1 . 1 . 2 . 1 . 2 . 2 . 1 .

causal chain non-causal chain

• ut-star

in-star

• 1.,2.

1. 2.

repeated contact returned contact

• Kovanen, Kaski, Kertész, Saramäki, PNAS 2013

temporal motifs in call sequences

• there is temporal homophily:

patterns where participants are similar (age, gender) are

• verexpressed
• there are gender differences:

females: chains & stars males: “ping-pong”

• there are group talk patterns:

chains & stars within social groups

2 . 1 . 1 . 2 . 1 . 2 . 2 . 1 .

causal chain non-causal chain

• ut-star

in-star

• 1.,2.

1. 2.

repeated contact returned contact

• Kovanen, Kaski, Kertész, Saramäki, PNAS 2013

• We have a small number of

close relationships and many “weak links”

• Those weak links are important!
• Our closest friends (who

resemble us!) are typically also friends

• Much of our communication is

with closest friends & family

• nly
• Our social patterns

change only slowly, if at all

Summary

5

• Call records provide information that cannot be
• btained with traditional methods (e.g. surveys)
• This allows statistical detection of behavioral

patterns

• Also a lot of commercial interest:

“data scientist” is one of the hottest professions currently

Summary

5