Finding Effectors in Social
Networks
T. Lappas (UCR), E. Terzi (BU),
D. Gunopoulos (UoA),
H. Mannila (Aalto U.)
Presented by: Eric Gavaletz
04/26/2011
The Motivation
If we have as input
the resulting
activation state
of a network...
Keep in mind that things
have already happened...
The Motivation
Can you identify
the most likely
effectors?
Or, can you find the nodes that could
have started a propagation leading to
the observed state?
:~$ diff effectors sota
top-k influential nodes
-- metric determining a node's
influence potential
++ how groups of effectors
collectively affect the network
influence maximization
-- nodes that will cause the
greatest propagation effect
++ nodes that best explain an
observed pattern
Effectors vs. top-k influential...
-- no influential potential
++ info-propagation model
++ influence weights as input
Suggested Input Sources:
Simply ask people how much they are influenced by
their peers, and machine learning algorithms...
Effectors vs. Influence Maximization...
k-effectors(0) = Influence Maximization
In many cases these problems are
similar...but they are different in some
very interesting cases...
Targeting niche markets
Technologically savvy groups
Tightly knit subcultures
Unwanted propagation is damaging
Targeted social networking
Hippster and Independent groups
Subversive and sensitive propaganda
How about an ideal activation state?
Triggering a regime change
You know the supporters
and the authoritarians…
Can you target effectors so that the
maximum supporters and minimum
authoritarians get the message?
Formal Problem Specification
Independent
Cascade
Propagation
Model
An active node
gets one shot to
activate a neighbors.
If it fails it does not try again.
Cost
Given a set of
effectors, X,
Cost(X) = input vector expected activations
Results
5508 DBLP Authors in
Databases, Data Mining,
Artificial Intelligence,
and Theory
Results
Results
Qualitative Results
Sets of effectors included
"very prolific authors"
Five of the authors had
156 - 250 papers each.
Follow-up
Influence Maximization in Social Networks When
Negative Opinions May Emerge and Propagate
Wei Chen et al.*
Lappas et al. [14] study k-effectors problem, which contains
influence maximization (without negative opinions) as a special
case. They also use a tree structure to make the computation
tractable, and then approximate the original graph with
a tree structure. The difference, besides not considering the
negative opinion, is that they use one tree structure...
* 10 authors for a 10 page paper?
Follow-up
Finding Influential Mediators in Social Networks
Cheng-Te Li et al.*
Lappas et al. [5] propose to find a set of effectors who can
cause an activation pattern as similar as possible to the given
active nodes in a social network.
REFERENCES
[1] W. Chen, A. Collins, R. Cummings, T. Ke, Z. Liu, D. Rincon, X. Sun,
Y. Wang, W. Wei, and Y. Yuan. Influence maximization in social networks when
negative opinions may emerge and propagate. Technical report, Technical Report
MSR-TR-2010-137, Microsoft Research, 2010.
[2] B. Chor and T. Tuller. Finding a maximum likelihood tree is hard. Journal of the
ACM (JACM), 53(5):722–744, 2006.
[3] T. Lappas, E. Terzi, D. Gunopulos, and H. Mannila. Finding effectors in social
networks. In Proceedings of the 16th ACM SIGKDD international conference on
Knowledge discovery and data mining, pages 1059–1068. ACM, 2010.
[4] C.T. Li, S.D. Lin, and M.K. Shan. Finding influential mediators in
social networks. In Proceedings of the 20th international conference companion
on World wide web, pages 75–76. ACM, 2011.
Questions?