4 cliques clusters

26
Cliques, Clans and Clusters Finding Cohesive Subgroups in Network Data

Upload: maksim-tsvetovat

Post on 18-Dec-2014

1.330 views

Category:

Education


2 download

DESCRIPTION

 

TRANSCRIPT

Page 1: 4 Cliques Clusters

Cliques, Clans and Clusters

Finding Cohesive Subgroups in

Network Data

Page 2: 4 Cliques Clusters

Social Subgroups

Frank & Yasumoto argue that actors seek social capital, defined as the access to resources through social ties

a) Reciprocity Transactions Actors seek to build obligations with others, and thereby gain in the ability to extract resources.

b) Enforceable Trust “Social capital is generated by individual members’ disciplined compliance with group expectations.”

c) Group Cohesion

Page 3: 4 Cliques Clusters

Goals

• Find a meaningful way to separate larger networks into groups

• Meaningful = • Reduce overlap• Locate cohesive groups

Page 4: 4 Cliques Clusters

Reciprocity

Page 5: 4 Cliques Clusters

Reciprocity

• Ratio of reciprocated pairs of nodes to number of pairs that have at least 1 tie• In example, reciprocity = 0.5• Called “dyad method”

Page 6: 4 Cliques Clusters

Transitivity

• Types of triadic relations (in undirected networks):• Isolation• Couples only• Structural holes• Clusters (also cliques)

Page 7: 4 Cliques Clusters

In directed networks

• There are 16 types of triads

• Triad language:• A-xyz-B form…• A= 1..16 (number of the triad in the catalogue)• X = number of pairs of vertices connected by

bidirectional arcs• Y = number of pairs of vertices connected by a

single arc; • z = number of unconnected pairs of vertices.

Page 8: 4 Cliques Clusters

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

Page 9: 4 Cliques Clusters

Triad Catalogue

• 9, 12, 13, 16 are transitive

• 6, 7, 8, 10, 11, 14, 15 are intransitive

• 1, 2, 3, 4, 5 do not contain arcs to meet the conditions of transitivity (they are vacuously transitive)

Page 10: 4 Cliques Clusters

Triad #16…

• …is known as a clique

• Cliques are a particular type of cohesive subgroups

• We can count the number of cliques in the network to estimate overall cohesion or evaluate local properties of nodes

Page 11: 4 Cliques Clusters

Cliques

• Definition • Maximal, complete subgraph

• Properties • Maximum density (1.0)

Minimum distances (all 1) • overlapping • Strict

Page 12: 4 Cliques Clusters

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

Page 13: 4 Cliques Clusters

Relaxation of Strict Cliques

• Distance (length of paths) • N-clique, n-clan, n-club

• Density (number of ties) • K-plex, ls-set, lambda set, k-core, component

Page 14: 4 Cliques Clusters

N-Cliques• Definition

• Maximal subset such that:

• Distance among members less than specified maximum

• When n = 1, we have a clique

• Properties • Relaxes notion of clique• Avg. distance can• be greater than 1

Page 15: 4 Cliques Clusters

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

Page 16: 4 Cliques Clusters

Issues with n-cliques• Overlapping

• {a,b,c,f,e} and {b,c,d,f,e} are both 2-cliques

• Membership criterion satisfiable through non- members

• Even 2-cliques can be fairly non-cohesive • Red nodes belong to same

2-clique but none are adjacent

Page 17: 4 Cliques Clusters

N-Clan• Definition

• An n-clique in which geodesic distance between nodes in the subgraph is no greater then n

• Members of set within n links of each other without using outsiders

• Properties • More cohesive

than n-cliques

Page 18: 4 Cliques Clusters
Page 19: 4 Cliques Clusters

N-Club

• Definition • A maximal subset S whose

diameter is <= n • No n-clique requirement

• Properties • Painful to compute• More plentiful than n-clans• Overlapping

Page 20: 4 Cliques Clusters

K-core:• A maximal subgraph such that:

• In English:• Every node in a subset is connected to at

least k other nodes in the same subset

Page 21: 4 Cliques Clusters

Example

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

Page 22: 4 Cliques Clusters

Notes

• Finds areas within which cohesive subgroups may be found

• Identifies fault lines across which cohesive subgroups do not span

• In large datasets, you can successively examine the 1-cores, the 2-cores, etc. • Progressively narrowing to core of network

Page 23: 4 Cliques Clusters

K-plex:

• Maximal subset such that:

• In English:• A k-plex is a group of nodes such that every

node in the group is connected to every other node except k

• Really a relaxation of a clique

Page 24: 4 Cliques Clusters

Example

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

Page 25: 4 Cliques Clusters

Notes

• Choosing k is difficult so meaningful results can be found

• One should look at resulting group sizes - they should be larger then k by some margin

Page 26: 4 Cliques Clusters

Next time…

• Making sense of triads - structural holes, brokerage and their social effects