looking for communities in networks
TRANSCRIPT
![Page 1: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/1.jpg)
Looking for communities innetworks
Krzysztof Kułakowski
in cooperation with
Małgorzata Krawczyk
Przemysław Gawroński
Piotr Gronek
Faculty of Physics and Applied Computer Science, AGH University of Sci and Tech, Kraków, Poland
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 2: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/2.jpg)
Overview
• The task
• A case study – the Southern Women
• The modularity after Newman
• A social problem – the Heider balance
• A continuous approach
Review: S. Boccaletti et al., Complex networks: Structure and dynamicsPhysics Reports 424 (2006) 175, section 7.1
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 3: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/3.jpg)
The task
• Graph representation of a network
N nodes
A(i,k)– connectivity matrix
A(i,k)= 0,1 or
A(i,k) ∈ R (weighted networks)
Jasmine’s blog Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 4: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/4.jpg)
The task
• The communities:
- subsets ofnodes, internallyconnectedmore than withother parts ofthe network
Our aim is to find separated communities.
Cornell > D.Constantine, NYT
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 5: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/5.jpg)
• The task
• A case study – the Southern Women
• The modularity after Newman
• A social problem – the Heider balance
• A continuous approach
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 6: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/6.jpg)
The Southern Women, Natchez 1935
L. C. Freeman, Finding social groups: a meta-analysis of the southern women data, in Ronald
Breiger, Kathleen Carley and Philippa Pattison (eds.) Dynamic Social Network Modeling andAnalysis. Washington, D.C.:The National Academies Press, 2003.
![Page 7: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/7.jpg)
Participation of the Southern Women in Events
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 8: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/8.jpg)
Results of 21 methods
![Page 9: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/9.jpg)
Comparison of 21 methods
![Page 10: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/10.jpg)
• The task
• A case study – the Southern Women
• The modularity after Newman
• A social problem – the Heider balance
• A continuous approach
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 11: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/11.jpg)
Basic tool - the modularity
),(),(1
ji
ij
jicc
w
kkjiA
wQ δ∑
−=
∑=ij
jiAw ),(where
∑=
j
i jiAk ),(
[M. E. J. Newman, Analysis of weighted networks, Phys. Rev. E 70, 056131 (2004)]
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 12: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/12.jpg)
Special cases
• One fully connected cluster of N nodesplus one separated node: Q = 1/N. Whenthe node is added to the cluster, Q=0.
• R separated fully connected clusters, each
of N nodes: Q=1-(N-1)/(RN) ≈ 1 if R>>1. When one fully connected cluster is
formed, Q=1/(RN) ≈ 0.
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 13: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/13.jpg)
Algorithm
• For a given set of data (N, {A(i,k)}) we have to constructa new connectivity matrix {cik=0,1}, which reveals thecommunities.
• start from cik=0 for all i,k.
• find the pair (n,m) such that when we put cnm=1,
Q increases most.
• repeat the search of such
pairs, until Q starts to
decrease. We accept the
division where Q has a
maximum.
![Page 14: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/14.jpg)
Results – a dendrogram
![Page 15: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/15.jpg)
• The task
• A case study – the Southern Women
• The modularity after Newman
• A social problem – the Heider balance
• A continuous approach
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 16: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/16.jpg)
The Heider balance• Let us transform the connectivity matrix as:
1→ 1 (friendly relation)
0 → -1 (hostile relation)
• Nodes = group members.
• In the balanced state, some rules are obeyed:
Friend of my friend is my friend A(i,k)>0 , A(k,n)>0 ⇒ A(i,n)>0
Enemy of my friend is my enemy A(i,k)<0 , A(k,n)>0 ⇒ A(i,n)<0
Friend of my enemy is my enemy A(i,k)>0 , A(k,n)<0 ⇒ A(i,n)<0
Enemy of my enemy is my friend A(i,k)<0 , A(k,n)<0 ⇒ A(i,n)>0
• If the state is not balanced, some members suffer the cognitivedissonance.
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 17: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/17.jpg)
•They try to get the balanced state, changing some of their relations.
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 18: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/18.jpg)
The Harary theorem
• At the balanced state, the group is divided into two
mutually hostile subgroups. The relations within
both subgroups are friendly.
[Z. Wang, W. Thorngate, Sentiment and social mitosis: Implications of Heider's
Balance Theory, J. of Artificial Society and Social Simulation vol. 6, no. 3 (2003)]
![Page 19: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/19.jpg)
Problems with the dynamics
• Sociological:
Relations are more subtle than just friendlyor hostile; more possibilities should be included.
• Computational:
The final balanced state can depend on the order of modified relations.
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 20: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/20.jpg)
A continuous dynamics• The relations are described by real numbers
A(i,k)∈(-R,R)
• Their time evolution is determined by differentialequations
∑
−=
n
knAniAR
kiA
dt
kidA),(),(
),(1
),(2
2
[More details in KK, Some recent attempts to simulate the Heider balance problem,
Computing in Science and Engineering, July/August 2007, p.80.]
The results agree with the best six methods,
reported by Freeman.
![Page 21: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/21.jpg)
• The task
• A case study – the Southern Women
• The modularity after Newman
• A social problem – the Heider balance
• The continuous approach
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 22: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/22.jpg)
Looking for communities in general–- the continuous approach
• In the Heider balance problem, thedifferential equations were motivated by the specific nature of the cognitivedissonance.
• In a general approach, for each pair we ask: do other connections confirm that thispair belongs to the same set?
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 23: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/23.jpg)
The continuous approach
• The differential equations are
]),(),([)),(1()),((),(
∑ −−ΘΘ=
n
bknAniAkiAkiAdt
kidA
The product of the step functions ensures that once
A(i,k)=0 or 1, it does not change any more.
Once the product A(i,n)A(n,k)>b , the connection of
sites i,k via site n proves that i,n belong to the same
cluster and so do n,k. Then A(i.k) should increase.
(more details in M. Krawczyk and K. K., arXiv:0709.0923 )
![Page 24: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/24.jpg)
The continuous approach – a test
• For separated and fullyconnected clusters, theconnectivity matrix has a block form
1
1
1
We add a noise: random numbers with theuniform distribution (0,e) are added to zeros and subtracted from units. The taskis to reproduce the original structure.
0
0
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 25: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/25.jpg)
The continuous approach – a test
R=2 R=3 R=4
R=6 R=7 R=8P
eP – probability that the initial structure is reproduced
e – upper limit of noise
modularity random walk continuousNR=60÷72, #=100, b=1/4,
![Page 26: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/26.jpg)
Final remarks
• The method is deterministic. Itproduces some results even for a purely random graph.
• Direct comparisons with socialexperiments (Southern women, Zachary karate club) can be dubious.
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 27: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/27.jpg)
Thank you
Summer School on Socio-Econo-Physics 2007 in Windberg
![Page 28: Looking for communities in networks](https://reader031.vdocument.in/reader031/viewer/2022011913/61d76b76f4773d6f070bbfcf/html5/thumbnails/28.jpg)
Control questions
• To calculate the modularity, one takes into account:
A. Only existing connections
B. Only a proposed division into clusters
C. Both A and B
• Differential equations raise difficulties because
A. high numerical precision is needed
B. the order of modified relations does matter
Summer School on Socio-Econo-Physics 2007 in Windberg