14 dynamic networks

52
1 Time and Social Networks

Upload: maksim-tsvetovat

Post on 30-Nov-2014

790 views

Category:

Technology


1 download

DESCRIPTION

 

TRANSCRIPT

Page 1: 14 Dynamic Networks

1

Time and Social Networks

Page 2: 14 Dynamic Networks

2

Background:•Most social network research has been static, though there is a growing interest in modeling network dynamics. This is occurring in two related directions:

•Two ways of thinking about time:•Movement of things through a network: Diffusion processes

•Change in the network itself: Structural implications of relation change.

Time and Social Networks

Page 3: 14 Dynamic Networks

3

Time and Social Networks

Historically, time has been incorporated into the network looking at

• changes in the distribution of an item over the population, over time

• the adoption of an innovation• the spread of an idea, etc.)

• different cross-sectional slices of the network

Page 4: 14 Dynamic Networks

4

Network Dynamics

• Limitations• Don’t allow us to explicitly model the changes within the

network, • Explain changes in the distribution of goods as a function of

timing.

• This static bias is built into some views of the network

• What we want is to be able to account for the dynamics of the network in “real time”• to account for changes in relations as a function of changes in

relations occurring around ego.\

Page 5: 14 Dynamic Networks

5

Network Forces

• Reciprocity• If A links to B, B links to A• “…essential for running of complex

societies”

• Dyads:• Reciprocated : A->B, B->A• Unreciprocated: A->B, not B->A• Null: no ties

Page 6: 14 Dynamic Networks

6

Measuring Reciprocity• Assumption: Link strength coded for

“liking” or “friendship”• Reciprocity = difference between sent

and returned liking

• Scale undefined, only in comparison:

Page 7: 14 Dynamic Networks

7

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

Dyadic Forces

Page 8: 14 Dynamic Networks

8

Transitivity

• Transitive triads:• A->B, B->C, A->C• A->B, not B->C or A->C

• Intransitive triads:• A->B, B->C, not A->C

• Null triads are transitive• No edges

• Complicated to compute on valued graphs

• Count the number of “forbidden triads”

Page 9: 14 Dynamic Networks

9

Enemy of my Friend is an Enemy

B

A

C

9

Page 10: 14 Dynamic Networks

10

Friend of my Enemy is an Enemy

B

A

C

10

Page 11: 14 Dynamic Networks

11

Enemy of my Friend is an Enemy

B

A

C

11

Page 12: 14 Dynamic Networks

12

Enemy of my Enemy is a Friend

B

A

C

12

Page 13: 14 Dynamic Networks

13

Group Balance

• Combination of friends and enemies:• “A friend of a friend is a friend”• “A friend of an enemy is an enemy”• “An enemy of a friend is an enemy”• “An enemy of an enemy is a friend”

• Assume binary (friend-enemy) or ternary (friend-enemy-neutral) relations

Page 14: 14 Dynamic Networks

14

Closure Force

Page 15: 14 Dynamic Networks

15

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

Closure Force

Page 16: 14 Dynamic Networks

16

Structural Holes

Page 17: 14 Dynamic Networks

17

Structural Hole Force

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

Page 18: 14 Dynamic Networks

18

Hierarchy

Page 19: 14 Dynamic Networks

19

Page 20: 14 Dynamic Networks

20

Group Balance

• Only allows for 2 opposing groups• Relax rule 4 (“enemy of an enemy”)

• Balanced graph will be partitioned into subsets• Only positive ties within subsets• Only negative ties outside subsets

Page 21: 14 Dynamic Networks

21

Measuring Balance

• Partition into groups• Count departures from balance:

• Positive lines between plus-sets• Negative lines within plus-sets

• Move nodes from one set to another to minimize number of departures

• Classic optimization problem

Page 22: 14 Dynamic Networks

22

Time 1

Page 23: 14 Dynamic Networks

23

Time 5

Page 24: 14 Dynamic Networks

24

Time 10

Page 25: 14 Dynamic Networks

25

Reciprocity

Page 26: 14 Dynamic Networks

26

Transitivity

Page 27: 14 Dynamic Networks

27

40

30

20

10

01 2 3 4 5 6 7 8 9 10 11 12 13 14

Week

% C

hang

e in

ties

Relational Stability

Relational Stability

• Percentage of ties changing at each step

Page 28: 14 Dynamic Networks

28

1 2 3 4 5 6 7 8 9 10 11 12 13 14 159

11

13

15

17Extent of Structural Imbalance

Structural Imbalance

Generalized Imbalance

Week

Structural (im)balance

Page 29: 14 Dynamic Networks

29

Result…

• Stable state of small oscillations• Similar to dynamic mechanical systems

Page 30: 14 Dynamic Networks

30

Doreian, Kapuscinski, Krackhardt & Szczypula:A brief history of balance through time.

“…the micro-level processes can be viewed as generating social forces that move the structure toward group balance.”

…negative ties within groups are likely less tolerated than positive ties between groups, as negatives within group may threaten the group in ways that positive ties between groups do not.

Page 31: 14 Dynamic Networks

31

An extension: A balance model of friendship change among adolescents

The basic hypothesis of social balance is that people will make choices that bring the entire group into balance.

But, consider how a given relationship looks from different perspectives: A transition that generates transitivity for one person can generate intransitivity for another. As such, there is no guarantee that friendship change will result in a globally balanced outcome.

Page 32: 14 Dynamic Networks

32

State-Transition diagram of triads

Page 33: 14 Dynamic Networks

33

Testing in simulationWhat are the implications for relationship change if people follow transitivity, from their own point of view?

J. Moody used a simulation, based on data from Add Health, to answer this question.

Rules: For every agent,

Attempt to achieve closureAttempt to achieve transitivity

Resolved through adding and dropping edges

Page 34: 14 Dynamic Networks

34

Strong transitive and intransitive rules:

Results in cliques forming quickly and maintaining over time.

Page 35: 14 Dynamic Networks

35

Moderate values for seeking transitivity and avoiding intransitivity: Results in a fluid network structure.

Page 36: 14 Dynamic Networks

36

Rules for network evolution• Seeking reciprocity (peer approval)

• Seeking transitivity (legitimation)

• Seeking balance (cognitive dissonance)

• Institutional Constraint• Fit to the institutional network

• Institutionalization• Escalation of tie strength - first mover advantage

• Planar optimization

• Feature selection

• Limited number of outgoing ties

• What about seeking power?

• Memory

• Any other forces?

Page 37: 14 Dynamic Networks

3737

Page 38: 14 Dynamic Networks

38

Friend of my Friend

Becomes my friend (with some probability)

B

A

C

38

Page 39: 14 Dynamic Networks

39

Balanced Triad

• The most stable configuration of network (Newcomb)

• 3 Simmelian ties (Krackhardt)• Reinforces conformity and similarity• Generates stable social norms of

behavior

39

Page 40: 14 Dynamic Networks

40

First Phase Transition

• After a small number of links is added, Triadic Closure rule strikes

• After critical mass is reached, transition from linear growth to quadratic

• Critical mass ~= 0.04

• Fully Connected network 40

Page 41: 14 Dynamic Networks

41

Let Conflict Strike

• Conflict inverts an edge from 1 (Friend) to -1 (Enemy)

• Conflict will decay at a linear rate and enemy link will disappear (0) in a short time

• What happens to triads when conflict strikes?

41

Page 42: 14 Dynamic Networks

42

Friend of my Friend

Can be an enemy???

B

A

C

42

Page 43: 14 Dynamic Networks

43

...dissonance causes action

43

Page 44: 14 Dynamic Networks

44

Can Conflict Spread?

44

Page 45: 14 Dynamic Networks

45

Can Conflict Spread?

45

Page 46: 14 Dynamic Networks

46

The greater the network density, the greater is the

probability of a catastrophic cascade

46

Page 47: 14 Dynamic Networks

47

Like the forest fire?

47

Page 48: 14 Dynamic Networks

48

Phase Transition #2

48

Page 49: 14 Dynamic Networks

49

alpha = 1.2886

49

Page 50: 14 Dynamic Networks

50

Degree Distribution

50

Page 51: 14 Dynamic Networks

51

Some results...

• Mean network density ~= 0.06• Mean density on Russian-language

LiveJournal ~=0.057• Network density is self-regulating;

Rapid increase in density quickly results in a collapse

51

Page 52: 14 Dynamic Networks

52

Instability - or complexity?

• The model produces viable networks at a set of parameter “sweet spots”

• Collapses otherwise!• e.g. low prior probability of conflict

results in super-critical density of links; when conflict finally strikes, all hell breaks loose!

• Rugged parameterlandscape

52