You are on page 1of 36

Inuence Maximization in Social Network

Using A New Centrality Measure Diusion Degree


Suman Kundu
Center for Soft Computing Research
Indian Statistical Institute
Kolkata - 700108
June 21, 2011
Outline
Social Networks Overview
Centrality Measures
Problem Denition, Application and Challenges
Available Solutions
Proposed Solution
Diusion Model
Diusion Degree
Assumption & Experimental Set Up
Algorithm
Results
Conclusion
References
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 2 / 36
Social Networks
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 3 / 36
Overview
What is Social Networks?
A Social Structure made up of individual or organization called node
Nodes are tied by one or more interdependence (e.g. friendship,
common interest, nancial exchange etc.)
Many kind of ties between nodes
Operates on many levels
From family upto national level
Example
Online Social Networks - Facebook, Twitter, Orkut, LinkedIn etc.
Who-talks-to-Whom Networks - Telephonic Communication, Email
Communication etc.
Collaboration Networks - Co-Authorship Networks, Co-Appearance in a
movie etc.
Natural World Networks - Food webs are representations of the
predator-prey relationships between species, Biological Network of
Neural Connections.
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 4 / 36
Example
Figure: HEPph Citation Network
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 5 / 36
Centrality Measures
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 6 / 36
Overview
What is centrality?
Relative importance of a vertex within a graph
e.g. How important a person is within a social network
Measures of centrality
Degree Centrality
Number of edges incedent on a vertex
Betweenness Centrality
Ratio of number of shortest path passing through the vertex and total
number of shortest paths between all pairs in the network
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 7 / 36
Degree Centrality
Nieminen introduce a simple, natural and perfectly general measure of
centrality based upon degree
1
Count of the degree or number of adjacencies for a point
For a Graph G(V, E) with n vertex (Where V is the set of vertex and
E is the set of edges) degree centrality C
D
(v) for vertex v is:
C
D
(v) =
n

i =1
(u
i
, v)
Where (u
i
, v) = 1 if and only if u
i
and v are connected by a link
= 0 Otherwise
A point v, can at most be adjacent to n 1 other points in the
graph. So,
C

D
(v) =

n
i =1
(u
i
, v)
n 1
1
[Nieminen, 1974]
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 8 / 36
Betweenness Centrality
For a Graph G(V, E) with n vertex, betweenness of vertex v is:
C
B
(v) =

s=v=t

st
(v)

st
Where
st
is the number of shortest path between s and t and
st
(v)
is the number of shortest path from s to t passing through v
Freeman
2
proved that the max value taken by C
B
(v) is achieved only
by the central point in a star; and it is equal to
n
2
3n + 2
2
Therefore, the relative betweenness centrality of any point in a graph
may be expressed as a ratio of
C

B
(v) =
2C
B
(v)
n
2
3n + 2
2
[Freeman, 1977]
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 9 / 36
Problem Denition, Applications and
Challenges
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 10 / 36
Inuence Maximization in Social Network
Social Network - A Medium of Information Spread
Opinions, Ideas, Information, Innovations and more
Inuence in form of Word-of-Mouth
Signicant increase of prot
One of the major problem to achieve the above target is
How to select the inuential individuals quickly, to target in information
spreading? That is selecting the initial seed set for inuence spreading.
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 11 / 36
Application of Inuence Maximization
Marketing
Give limited free sample of products and/or applications
Wait for spreading of the information
Create potential buyer of the product and/or applications
Other Than Marketing
Spread of Innovation
Detect Stories in Blog
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 12 / 36
Challenges
Online Social Networks are obvious choice for marketing or
information spreading
3
Online Social Networks connects a huge number of people
Online Social Networks collects huge amount of information about the
Social Network Structure and Communication Dynamics
Challenges
The Social Networks are Large Scale
Complex Connection Structure
Dynamic Network
Solution needs to be very ecient and scalable
3
[Chen et al., 2009]
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 13 / 36
Complex Structure
Figure: HEPph Citation Network
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 14 / 36
Available Solutions
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 15 / 36
Classic Approach
High Degree Heuristics
Nodes are selected according to their degree rank
Random Selection
Seeding nodes are selected randomly
Most Central Heuristics
Nodes are selected according to their Betweenness or other centrality
rank
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 16 / 36
Hill Climbing Greedy Approach
Kempe, Kleinberg, and Tardos
4
are the rst to proposed greedy
solution for the problem
Present a greedy algorithm which guarantees that the inuence
maximization is within (1 1/e) of the optimal inuence spread
Also show through experiments that greedy algorithm outperforms the
classic degree and centrality based heuristics
Drawback of Greedy Algorithm
Eciency - Simulation based approach, needs to simulate sucient
amounts to get accurate estimation; Unlikely to get results for online
social networks contains millions of nodes.
Some researcher showed that even for a 15K vertics graph taking days
to compute the result
4
[Kempe et al., 2003]
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 17 / 36
Improvements over Greedy Algorithm
Several Attempts to improve the greedy approach is proposed-
Pabli A. et al proposed a Set Covering Greedy Algorithm
5
to improve
the performance and eciency of greedy algorithm
Leskovec et al.
6
present an optimized greedy algorithm referred by
Cost Eective Lazy Forward
Chen et al.
7
proposed two ecient algorithm to further improving on
greedy algorithm. This algorithm is known as NewGreedy and
MixedGreedy
Even after improvement, these greedy approach is not even closer to
the speed of centrality based heuristic model
5
[Estevez et al., 2007]
6
[Leskovec et al., 2007]
7
[Chen et al., 2009]
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 18 / 36
Degree Discount Heuristics
In their paper Chen et al.
8
also proposed a degree discount heuristic
General idea of degree discount algorithm of Chen et al. is that if one
node is considered as seed then the links connecting with the node will
not be counted as a degree of the other nodes i.e. when considering
the next node, the links connecting with the nodes already in the seed
set will be discounted.
The running time of the algorithm is comparable with high degree
heuristics. However, in our experiments we did not found
improvements over the classic high degree heuristics.
8
[Chen et al., 2009]
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 19 / 36
Proposed Solution
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 20 / 36
Diusion Model
Independent Cascade Model
Is a probabilistic information diusion model
Starts with a set of initial active nodes
In step t an active node gets single chance to activate an inactive
neighbor with diusion probability
Linear Threshold Model
In this model one node become active if the fraction of its active
neighbor is greater then the threshold value
Other Variants
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 21 / 36
Diusion Degree - Overview
Many of the available centrality measures considered only structural
property of a node
However, diusion Probability plays a vital role in inuence ow over
the network
Additionally in a social network, neighborhood has a signicant
impact on ones inuence
We proposed a new centrality measurement for vertex named
Diusion Degree considering the above points of social network and
diusion method
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 22 / 36
Diusion Degree - Mathematical Explanation I
The general Degree Centrality
9
measure of node v can be dened as
C
D
(v) =
n

i =1
(u
i
, v) (1)
where function (u
i
, v) dened as,
(u
i
, v) = 1 if and only if u
i
and v are connected
= 0 otherwise.
In a diusion process, a node v with propagation probability
v
, can
activate its neighbor u with probability
v
. So, considerable
contribution of node v in the diusion process is
C

DD
(v) =
v
C
D
(v). (2)
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 23 / 36
Diusion Degree - Mathematical Explanation II
When the diusion process propagates to the next level, active
neighbors of v will try to activate their inactive neighbors.
If the propagation probability of i
th
neighbor of v is
i
, considerable
contribution of i
th
neighbor in the diusion process is
C

DD
(i ) =
i
C
D
(i ). (3)
Thus the cumulative contribution in the diusion process by
neighbors of v will be maximized when all of its neighbors will be
activated in the previous step.
In this scenario, the total contribution of neighbors of v is
C

DD
(v) =

i neighbors(v)
C

DD
(i ). (4)
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 24 / 36
Diusion Degree - Mathematical Explanation III
The diusion degree of a node is dened as the cumulative
contribution score of the node itself and its neighbors.
So, from the equations 2 and 4 we can dene the diusion degree
C
DD
as
C
DD
= C

DD
+ C

DD
(5)
=
i
C
D
(i ) +

i neighbors(v)
C

DD
(i ) (6)
=
i
C
D
(i ) +

i neighbors(v)

i
C
D
(i ). (7)
9
[Nieminen, 1974]
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 25 / 36
Assumption & Experimental Data Set
Assumption
Independent Cascade Model for experiment and result comparison
Directed Social Networks
Only the in degree is contributing towards inuencing
Experimental Data Set
Large Scale Citation Network - DBLP(4.47Lac+ nodes & 23.27Lac+
Links) and HEPph(35K+ nodes & 4Lac+ links)
Large Scale Online Social Network - Twitter(4.15Lac+ Nodes & 8.2Lac
links)
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 26 / 36
Algorithm
Step 1: Compute diusion degree for all vertexes
Step 2: Order the vertexes based on the diusion degree
Step 3: Select top k nodes for top k-inuence maximization problem
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 27 / 36
Results
We compare our result with other centrality based heuristics model
like high degree and degree discount algorithm
To obtain approximate result of inuenced nodes we simulate 100
times
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 28 / 36
Results - DBLP with 0.05 Diusion Probability
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 29 / 36
Results - HEPph with 0.05 Diusion Probability
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 30 / 36
Results - Twitter with 0.05 Diusion Probability
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 31 / 36
Conclusion
Online Social Networks attracts marketing teams for exploring and
increasing market over the globe. Thus the demand of fast algorithm
with satisfying results are in demand
Based on our study so far and primary results, we believe that ne
tuned centrality based heuristics may provide truly scalable solutions
to the inuence maximization problem with satisfying inuence spread
and fast running time.
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 32 / 36
Your Question?
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 33 / 36
Thank You
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 34 / 36
Reference I
Chen, W., Wang, Y., and Yang, S. (2009).
Ecient inuence maximization in social networks.
In KDD 09: Proceedings of the 15th ACM SIGKDD international
conference on Knowledge discovery and data mining, pages 199208,
New York, NY, USA. ACM.
Estevez, P. a., Vera, P., and Saito, K. (2007).
Selecting the Most Inuential Nodes in Social Networks.
In 2007 International Joint Conference on Neural Networks, pages
23972402. Ieee.
Freeman, L. (1977).
A set of measures of centrality based on betweenness.
Sociometry, pages 3541.
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 35 / 36
Reference II
Kempe, D., Kleinberg, J., and Tardos, E. (2003).
Maximizing the spread of inuence through a social network.
In KDD 03: Proceedings of the ninth ACM SIGKDD international
conference on Knowledge discovery and data mining, pages 137146,
New York, NY, USA. ACM.
Leskovec, J., Krause, A., Guestrin, C., Faloutsos, C., VanBriesen, J.,
and Glance, N. (2007).
Cost-eective outbreak detection in networks.
In Proceedings of the 13th ACM SIGKDD international conference on
Knowledge discovery and data mining, pages 420429. ACM.
Nieminen, J. (1974).
On centrality in a graph.
Scandinavian Journal of Psychology, 15:322336.
S. Kundu (CSCR, ISI Kolkata) Inuence Maximization in Social Network June 21, 2011 36 / 36

You might also like