Antonia Godoy Lorite Long Term Evolution of Email ... · Antonia Godoy Lorite Department of...

2nd Workshop on Graph-based Technologies and Applications

Barcelona

Antonia Godoy Lorite

Department of Chemical Engineering Universitat Rovira i Virgili

Antonia.godoy@urv.cat

Advisors : Dr. Roger Guimerà and Dr. Marta Sales-Pardo

February 21st, 2014

Long Term Evolution of Email Communication Network

2

Techno-social system: the combination between technology and social interactions

British telephone recordsUniversity email network 2007

Air trasportation network

3

Techno-social system: the combination between technology and social interactions

University email network 2007

Air trasportation network

British telephone records

4

Current Methods focus on Short Term Dynamics

➔Information flow

➔Homofily leds to new connections

Gueorgi Kossinets and Duncan J. Watts. Science 311, (2006)

Kossinets, Gueorgi and Kleinberg, Jon and Watts, Duncan. KDD '08 (2008)

5

Current Methods focus on Short Term Dynamics

➔Information flow

➔Homofily leds to new connections

Gueorgi Kossinets and Duncan J. Watts. Science 311, (2006)

Kossinets, Gueorgi and Kleinberg, Jon and Watts, Duncan. KDD '08 (2008)What about long term

Evolution(on a yearly basis)?

6

Definition of the network which lead to the Evolution

2007 2008

20092010

7

The Growth of Email Communication has High Variability

ωij( t)

To study the Evolution of the network we calculate which is called the annual logarithmic growth rate (LGR) of each link.

g ijω=log (ωij(t+1)

ω ij(t) )

8

The Growth of Email Communication has High Variability

ωij( t)

To study the Evolution of the network we calculate which is called the annual logarithmic growth rate (LGR) of each link.

g ijω=log (ωij(t+1)

ω ij(t) )

The Fit is a convolution of a not centered Exponential distributions and a gaussian distribution.

σ exp≃0,4 σgauss≃0,1 x0≃0,08

9

The Growth of Email Communication has High Variability

σ exp≃0,4 σgauss≃0,1 x0≃0,08

ωij( t )

g ijω=log (ωij (t+1)

ω ij(t) )

10

The Growth of Email Communication has High Variability

ωij( t )

σ exp≃0,4 σgauss≃0,1 x0≃0,08

s i(t)=∑ j , j≠i(ωij(t ))

g ijs=log( si (t+1)s i( t) )

g ijω=log (ωij (t+1)

ω ij(t) )

11

The Growth of Email Communication has High Variability

σ exp≃0,4 σgauss≃0,1 x0≃0,08

s i(t)=∑ j , j≠i(ωij(t ))

σ exp≃0,5 σgauss≃0,009 x0≃0,2

g ijs=log( si (t+1)s i( t) )

g ijω=log (ωij (t+1)

ω ij(t) )ωij( t )

12

Individuals have Stable Patterns on Long Term

Gini Coefficient is measure of how equally is distributed the information (email) over each user's contacts.

G=0 Equally distributed G=1Unequally distributed

13

Individuals have Stable Patterns on Long Term

Gini Coefficient is measure of how equally is distributed the information (email) over each user's contacts.

G=0 Equally distributed G=1Unequally distributed

Gi (t) Gi (t+Δ t )

G j(t) , j≠i

dself i

dothers i

14

Individuals have Stable Patterns on Long Term

Gini Coefficient is measure of how equally is distributed the information (email) over each user's contacts.

G=0 Equally distributed G=1Unequally distributed

Gi (t) Gi (t+Δ t )

G j(t) , j≠i

dself i

dothers i

15

Individuals have Stable Patterns on Long Term

Gini Coefficient is measure of how equally is distributed the information (email) over each user's contacts.

G=0 Equally distributed G=1Unequally distributed

Gi (t) Gi (t+Δ t )

G j(t) , j≠i

dself i

dothers i

Persistence of Social Signatures at the Individual Level

16

Thank you!

➔Roger Guimera, Marta Sales-Pardo, Antoni Aguilar, Francesco Massucci, Arnau Galvalda, Antonia Godoy, Toni Vallès, Oriol Senan, Marc Tarrés, Manuel Miranda

➔Funding

➔More:➔http://seeslab.info➔http://twitter.com/sees_lab