Ars Localcom

I NTRODUCTION Community detection
Local Community Computation
Rushed Kanawati
LIPN, CNRS UMR 7030; USPC
http://lipn.fr/∼kanawati
rushed.kanawati@lipn.univ-paris13.fr
1 / 32
P LAN
1 I NTRODUCTION
2 Community detection
Local community detection
2 / 32
C OMPLEX NETWORK
Definition
Graphs modeling (direct/indirect) interactions among actors.
Basic topological features

I Low Density
I Small Diameter
I Heterogeneous degree
distribution.
I High Clustering
coefficient
I Community structure
3 / 32
C OMMUNITY ?
Definitions
I A dense subgraph loosely coupled to other modules in the
network
I A community is a set of nodes seen as one by nodes outside the
community
I A subgraph where almost all nodes are linked to other nodes in
the community.
I ...
4 / 32
C OMMUNITY DETECTION PROBLEM
I Local community identification (ego-centred).
I Network partition computing
I Overlapping community detection
5 / 32
L OCAL COMMUNITY
6 / 32
L OCAL COMMUNITY
C: Core (noyau)
B: Border (frontière)
S: Shell
U: extérieur de la commu
7 / 32
L OCAL COMMUNITY
1 C ← {φ}, B ← {n0 } S ← Γ(n0 )

2 Q ← 0 /* a community quality function */
3 While Q can be enhanced Do
1 n ← argmaxn∈S Q
2 S ← S − {n}
3 D ← D + {n}
4 update B, S, C
4 Return D
8 / 32
Q UALITY FUNCTIONS : E XEMPLES I
Local modularity R [Cla05]

Bin
R= Bin +Bout
Local modularity M [LWP08]

Din
M= Dout
Local modularity L [CZG09]

P P
kΓ(i)∩Dk kΓ(i)∩Sk
Lin i∈D i∈B
L= Lex where : Lin = kDk , Lex = kBk
And many many others . . . [YL12]
9 / 32
L OCAL MODULARITY LIMITATIONS : A N EXEMPLE
Blank nodes enhance Bin and Din without affecting Bout and Dout
Bank nodes will be added if M or R modularities are used
Low precision computed communities
Proposed solution: Ensemble approaches
10 / 32
M ULTI - OBJECTIVE LOCAL COMMUNITY

IDENTIFICATION
Three main approaches

Combine then Rank
Ensemble ranking
Ensemble clustering
11 / 32
C OMBINE THEN R ANK
Principle
Let Qi (s) be the local modularity value induced by node s ∈ S

Qi (s)−min Qi (u)
u∈S
if max Qi (u) 6= min Qi (u)


] max Qi (u)−min Qi (u)
Qi (s) = u∈S u∈S u∈S u∈S
1

otherwise
k
1 P ]
Qcom (s) = k Qi (s)
i=1
12 / 32
E NSEMBLE RANKING APPROACHES
Principle
I Rank S in function of each local modularity Qi
I Select the winner after applying ensemble ranking approach
I What stopping criteria to apply ?
Stopping criteria
I Strict policy : All modularities should be enhanced
I Majority policy : Majority of local modularities are enhanced
I Least gain policy : At least one local modularity is enhanced.
13 / 32
E NSEMBLE RANKING
Problem
I Let S be a set of elements to rank by n rankers
I Let σi be the rank provided by ranker i
I Goal: Compute a consensus rank of S.
14 / 32
E NSEMBLE RANKING
Problem
I Let S be a set of elements to rank by n rankers
I Let σi be the rank provided by ranker i
I Goal: Compute a consensus rank of S.
Déjà Vu: Social choice algorithms, but . . .

I Small number of voters and big number of candidates
I Algorithmic efficiency is required
I Output could be a complete rank
15 / 32
E NSEMBLE RANKING : APPROACHES
Borda
I Borda’s score of i ∈ σk :
Bσk (i) = {count(j)|σk (j) < σk (i) ; j ∈ σk }.
I Rank elements in function of
B(i) = kt=1 wt × Bσt (i).
P
Jean-Charles de Borda [1733-1799]
16 / 32
Condorcet
I The winner is the candidate who defeats
every other candidate in pairwise
majority-rule election
I The winner may not exists
Marquis de Condorcet [1743-1794]
17 / 32
Condorcet
I The winner is the candidate who defeats
every other candidate in pairwise
majority-rule election
I The winner may not exists
Marquis de Condorcet [1743-1794]
Condorcet 6= Borda
I Votes : 6 × A B C, 4 × B C A
I Borda winner : B
I Condorcet winner : A
18 / 32
Extended Condorcet criterion

If for every a ∈ A and b ∈ B, majority prefers a to b the all elements in A
should be ranked before any element in B.
Arrow’s Theorem
No vote rule can have the following desired
proprieties :
I Every result must be achievable somehow.
I Monotonicity.
I Independence of irrelevant attributes.
Kenneth Arrow, 1921-
I Non-dictatorship.
19 / 32
Optimal Kemeny rank aggregation

I Let d() be distance over rankings σi (ex.
Kendall τ , Spearman’s footrule)
P
I Find π that minimise i d(π, σi )
I NP-Hard problem
I Approximation : Local Kemeny : two

adjacent candidats are in the good order.
John Kemeny 1926-1992
I Local Kemeny : Apply Bubble sort using the
majority preference partial order relationship
I Approximate Kemeny : Apply QuickSort
20 / 32
E NSEMBLE CLUSTERING APPROACHES
Principle
I Let CQ i
vq be the the local community of vq applying Qi .
I We have a natural partition : PQi = {CQ i Qi

vq , Cvq }
I Apply an ensemble clustering approach.
21 / 32
E NSEMBLE CLUSTERING : PRINCIPLE
from A. Topchy et. al. Clustering Ensembles: Models of Consensus and Weak Partitions. PAMI, 2005
22 / 32
E NSEMBLE CLUSTERING : APPROACHES
CSPA: Cluster-based Similarity Partitioning Algorithm

I Let K be the number of basic models, Ci (x) be the cluster in model
i to which x belongs.
K
P
δ(Ci (v),Ci (u))
i=1
I Define a similarity graph on objects : sim(v, u) = K
I Cluster the obtained graph :

Isolate connected components after pruning edges
Apply community detection approach
I Complexity : O(n2 kr) : n # objects, k # of clusters, r# of clustering

solutions
23 / 32
CSPA : E XEMPLE
from Seifi, M. Cœurs stables de communautés dans les graphes de terrain. Thèse de l’université Paris 6, 2012
24 / 32
HGPA: HyperGraph-Partitioning Algorithm

I Construct a hypergraph where nodes are objects and hyperedges
are clusters.
I Partition the hypergraph by minimizing the number of cut
hyperedges
I Each component forms a meta cluster
I Complexity : O(nkr)
25 / 32
MCLA: Meta-Clustering Algorithm

I Each cluster from a base model is an item
I Similarity is defined as the percentage of shared common objects
I Conduct meta-clustering on these clusters
I Assign an object to its most associated meta-cluster
I Complexity : O(nk2 r2 )
26 / 32
E XPERIMENTS
Protocol ([Bag08])
1 Apply the different algorithms on nodes in networks for which a
ground-truth community partition is known.
2 For each node compute the distance between the real-partition

and the computed one (Ex. NMI [Mei03])
3 Compute average and standard deviation for the network.
27 / 32
D ATASETS
Zachary’s Karate Club Football network
US Political books network Dolphins social network
28 / 32
R ESULTS : EVALUATING STOPPING CRITERIA (NMI)
29 / 32
R ESULTS : C OMPARATIVE RESULTS (NMI)
30 / 32
B IBLIOGRAPHY I
J. P. Bagrow, Evaluating local community methods in networks, J. Stat. Mech. 2008 (2008), no. 5, P05001.
Aaron Clauset, Finding local community structure in networks, Physical Review E (2005).
Jiyang Chen, Osmar R. Zaı̈ane, and Randy Goebel, Local community identification in social networks, ASONAM, 2009,
pp. 237–242.
Feng Luo, James Zijun Wang, and Eric Promislow, Exploring local community structures in large networks, Web Intelligence
and Agent Systems 6 (2008), no. 4, 387–400.
Marina Meila, Comparing clusterings by the variation of information, COLT (Bernhard Schölkopf and Manfred K. Warmuth,
eds.), Lecture Notes in Computer Science, vol. 2777, Springer, 2003, pp. 173–187.
31 / 32
B IBLIOGRAPHY II
Jaewon Yang and Jure Leskovec, Defining and evaluating network communities based on ground-truth, ICDM
(Mohammed Javeed Zaki, Arno Siebes, Jeffrey Xu Yu, Bart Goethals, Geoffrey I. Webb, and Xindong Wu, eds.), IEEE
Computer Society, 2012, pp. 745–754.
32 / 32

Ars Localcom

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ars Localcom

Uploaded by

Copyright:

Available Formats

I NTRODUCTION Community detection

Local Community Computation

Basic topological features

C OMMUNITY DETECTION PROBLEM

I Local community identification (ego-centred).

I Network partition computing

I Overlapping community detection

1 C ← {φ}, B ← {n0 } S ← Γ(n0 )

Q UALITY FUNCTIONS : E XEMPLES I

Local modularity R [Cla05]

Local modularity M [LWP08]

Local modularity L [CZG09]

And many many others . . . [YL12]

L OCAL MODULARITY LIMITATIONS : A N EXEMPLE

M ULTI - OBJECTIVE LOCAL COMMUNITY

Three main approaches

C OMBINE THEN R ANK

E NSEMBLE RANKING APPROACHES

Déjà Vu: Social choice algorithms, but . . .

E NSEMBLE RANKING : APPROACHES

E NSEMBLE RANKING : APPROACHES

E NSEMBLE RANKING : APPROACHES

E NSEMBLE RANKING : APPROACHES

Extended Condorcet criterion

E NSEMBLE RANKING : APPROACHES

Optimal Kemeny rank aggregation

I Approximation : Local Kemeny : two

E NSEMBLE CLUSTERING APPROACHES

I We have a natural partition : PQi = {CQ i Qi

E NSEMBLE CLUSTERING : PRINCIPLE

E NSEMBLE CLUSTERING : APPROACHES

CSPA: Cluster-based Similarity Partitioning Algorithm

I Cluster the obtained graph :

I Complexity : O(n2 kr) : n # objects, k # of clusters, r# of clustering

E NSEMBLE CLUSTERING : APPROACHES

HGPA: HyperGraph-Partitioning Algorithm

E NSEMBLE CLUSTERING : APPROACHES

MCLA: Meta-Clustering Algorithm

2 For each node compute the distance between the real-partition

3 Compute average and standard deviation for the network.

Zachary’s Karate Club Football network

US Political books network Dolphins social network

R ESULTS : EVALUATING STOPPING CRITERIA (NMI)

Zachary’s Karate Club Football network

US Political books network Dolphins social network

R ESULTS : C OMPARATIVE RESULTS (NMI)

Zachary’s Karate Club Football network

US Political books network Dolphins social network

You might also like