network analysis

30
Network Analysis 안안안 2004.05.02

Upload: prince

Post on 07-Jan-2016

31 views

Category:

Documents


2 download

DESCRIPTION

Network Analysis. 안용열 2004.05.02. Austin Powers: The spy who shagged me. Let’s make it legal. Robert Wagner. Wild Things. What Price Glory. Barry Norton. A Few Good Man. Monsieur Verdoux. Degree = 3. Links. A shortest path with path length=3 (Equivalent with 3 clicks in WWW). - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Network Analysis

Network Analysis

안용열2004.05.02

Page 2: Network Analysis

A Few Good Man

Robert Wagner

Austin Powers: The spy who shagged me

Wild Things

Let’s make it legal

Barry Norton

What Price Glory

Monsieur Verdoux

Page 3: Network Analysis

Basic Concepts

Nodes

Links

Degree = 3

A shortest path with path length=3 (Equivalent with 3 clicks in WWW)

Page 4: Network Analysis

Basic Concepts : Degree Distribution

Degree

Occurrence

k

o

Number of k-degree-Nodes is o

Page 5: Network Analysis

Basic Concepts : Power Law

Exponential Network

Scale-free Network

Page 6: Network Analysis

Small World

• Milgram’s mail experiment

• Small world project : http://smallworld.columbia.edu/

• WWW’s diameter ~ 20 clicks

Page 7: Network Analysis

Strength of weak ties

• Granovetter shows the strength of weak ties

• Weak ties = Shortcut between heterogeneous communities

Page 8: Network Analysis

High Clustering

내 친구 A 가 내 친구 B 를 알고 있을 확률

A 라는 임의의 사람이 B 라는 임의의 사람을 알고 있을 확률

Network has many triangles!!

Page 9: Network Analysis

Small-world network model

• Networks are small world, and highly clustered.

• Duncan Watts & Steven Strogatz made a model

Page 10: Network Analysis

Small world network model

• Shows Small-World behavior (Of course..)

• Shows High clustering

• But, does not shows Power-law degree distribution

Page 11: Network Analysis

Scale-Free Network

• Barabasi & Albert & Jeong shows that Internet has Power-law degree distribution

• Power-Law degree distribution = Scale-free

• Power-Law degree distribution means, “very large hubs exist”

Page 12: Network Analysis

Network Centralities

• Degree• Closeness• Betweenness(=load)• Range

Page 13: Network Analysis

Degree

• 얼마나 많은 링크를 가지고 있는가 ?

• High degree centrality ‘Hub’

• 보통 가장 중요한 centrality

Page 14: Network Analysis

Closeness

• 얼마나 다른 이들과 가까운가 ?

• 1/ ( 모든 노드들의 쌍들의 거리 합 )

Page 15: Network Analysis

Betweenness

• 얼마나 주요 길목에 위치하는가 ?

• 모든 노드쌍들에 대해서 그 둘을 잇는 가장 짧은 길을 찾고 , 그 길 위에 있는 노드들의 Betweenness 값을 올려준다 .

Page 16: Network Analysis

• Betweenness Centrality (BC) [Freeman, 1977]

Example: the BC at k contributed by the communication from i to j is

Accumulate over all ordered pairs:

( ) ( ) 2 / 3i jb k

i j

k11

23

13

13

23

( )

,

i jk k

i j

b b

bij(k) (fraction in the number of the shortest

paths between i and j that pass through k.) “How much is the k-th node influential to the communication between i and j”

Page 17: Network Analysis

• Load

Example: load at k due to a packet from i to j is

Accumulate over all ordered pairs:

43)( jikl

i j

k11

2

1

2

1

4

1

4

3

ji

jikk ll

,

)(

lij(k) (fraction of a unit packet sent from node i to node j along the shortest paths, that pass through k, assuming even division at branching points and accumulation at merging points.)

Page 18: Network Analysis

Range

• 어떤 링크가 있을 때 , 그 링크가 얼마나 ‘숏’컷인가 ?

• 링크를 자른 뒤 그 링크가 연결하고 있던 두 노드사이의 거리를 잰다 .

Page 19: Network Analysis

In Computer Program.. : Network

• Two column format

0 1

1 5

3 2

2 1

Means

01

2

3

5

Page 20: Network Analysis

In Computer Program.. : Network

• Neighbor Array

0 1 2 3 4 5 6 7

Degree is the number of neighbors

Page 21: Network Analysis

In Computer Program..: Closeness, Betweenness

• 이런 centrality 들을 계산하기 위해서는 모든 node pair 에 대한 계산이 필요 네트워크의 노드개수가 n 개라면 , node pair 의 수는 n(n-1)/2 ~ n^2

• 네트워크가 커질수록 계산이 대단히 힘들어짐 .

Breadth-first algorithm 을 이용

Page 22: Network Analysis

In Computer Program..: Closeness, Betweenness

• 한 노드로부터 출발하여 다른 모든 노드로 가는 shortest path 를 한 번에 구한 뒤에 각 path 를 거꾸로 밟아오면서 Betweenness 를 구한다 .

• 자세한 알고리즘 : http://143.248.182.187/brandes.pdf

Page 23: Network Analysis

Centrality 의 이용• Epidemics

• Community identification

• …..

Page 24: Network Analysis

Epidemic spreading, idea spreading

• Hub 때문에 Scale-free network 위에서는 전염병이 사라지지 않음

• Hub 만 감염시키면 삽시간에 전 네트워크로 어떤 idea 나 정보들을 퍼뜨릴 수 있다 .

Page 25: Network Analysis

Immunization strategy

• 임의의 한 명을 골라서 그 사람을 접종시키지 말고 그 사람의 친구에게 예방접종을 시키는 방법

• 사실상 링크를 임의로 선택하는 것이기 때문에 링크를 많이 가지고 있는 허브에게 예방접종이 될 확률이 높아진다 .

Page 26: Network Analysis

Community Identification

• Every social networks have community structure.

• Email network

Page 27: Network Analysis

Community Identification

• 대표적인 알고리즘 : Girvan-Newman algorithm– Based on betweenness centrality

Page 28: Network Analysis

Community Identification

• Edge clustering coefficient 를 이용한 알고리즘

• Voltage 를 이용한 알고리즘• Flow 를 이용한 알고리즘• …

Page 29: Network Analysis

Tools

• Pajek

• Netminer

Page 30: Network Analysis

참고사이트

• http://stat.kaist.ac.kr : 실험실 홈페이지• http://www.nd.edu/~networks/ : 노틀담

대학 network 홈페이지• http://vlado.fmf.uni-lj.si/pub/networks/paje

k/ : pajek 홈페이지

• http://www.netminer.com/ : netminer 홈페이지

• http://cafe.naver.com/sna.cafe