Providers : A.Shekari E.Ehsani E.Golzardi

1
Providers A.Shekari E.Ehsani
E.Golzardi
In the name of god
Science Research Breanch Islamic Azad University
Limiting the Spread of Misinformation in Social
Networks

• Dear Professor Mr.Sheykh Esmaili

2
Table of Contents

• Introduction
• Related Work
• Diffusion OF Misinformation
• Eventual Influence Limitation
• Evaluation
• EIL With Incomplete Data
• Conclusion
• Strong Week Points
• Refrence

3
Introduction

• Online social networks have many benefits
• social networks can be very beneficial
• It can also have disruptive effects
• social networks are the main source of news for
  many people today

4
Table of Contents

• Introduction
• Related Work
• Diffusion OF Misinformation
• Eventual Influence Limitation
• Evaluation
• EIL With Incomplete Data
• Conclusion
• Strong Week Points
• Refrence

5
Related Work

• The experimental results show that
• the greedy approach performs better than the
  heuristics
• the best strategy for the first player is to
  choose high degree nodes
• the first player, the first to decide, is not
  always advantageous

6
Table of Contents

• Introduction
• Related Work
• Diffusion OF Misinformation
• Eventual Influence Limitation
• Evaluation
• EIL With Incomplete Data
• Conclusion
• Strong Week Points
• Refrence

7
Diffusion OF Misinformation

• A social network can bemodeled as a directed
  graph G (N,E)
• A node w is a neighbor of a node v if and only if
  there is ev,w ? E, an edge from v to w in G.
• pv,w is assigned to each edge ev,w

8
Diffusion Modle

• Independent cascade model (ICM)
• Multi-Campaign Independent Cascade Model (MCICM)
• Campaign-Oblivious Independent Cascade Model
  (COICM)

9
Table of Contents

• Introduction
• Related Work
• Diffusion OF Misinformation
• Eventual Influence Limitation
• Evaluation
• EIL With Incomplete Data
• Conclusion
• Strong Week Points
• Refrence

10
Eventual Influence Limitation

• Given a network and the MCIC Model, campaign C
  spreading bad information is detected with delay
  r
• budget k, select AL as seeds for initial
  activation with the limiting campaign L

11
Table of Contents

• Introduction
• Related Work
• Diffusion OF Misinformation
• Eventual Influence Limitation
• Evaluation
• EIL With Incomplete Data
• Conclusion
• Strong Week Points
• Refrence

12
Problem De?nition

• Problem De?nition
• One strategy to deal with a misinformation
  campaign is to limit the number of users
• we will assume that the spread of in?uence for
  campaign C starts from one node n and at that
  point campaign L is initiated
• we will focus on minimizing the number of nodes
• We refer to this problem as the eventual in?uence
  limitation problem (EIL)

13
General Purpose

• limiting the in?uence of a misinformation
  campaign
• Submodularity Of EIL
• f(S U V) f(S) gt f(T U V)-f(T)

14
General Purpose

• for all elements v and all pairs of sets S ? T
• as high as the marginal gain from adding the
  same element to
• a superset of S

15
General Purpose

• The greedy hill-climbing
• ? starting with the empty set, and
  repeatedly adding an
• element that gives the maximum marginal
  gain

16
Evaluation

• in our speci?c problem each simulation involves
  the expensive computation of shortest paths which
  is crucial to EIL and this makes EIL even more
  computationally
• intense then the in?uence maximization problems

17
Evaluation

• In this Figure we present our evaluation of the 4
  methods on MCICM
• delay20 delay50

18
Evaluation

• Figure 4 Evaluation for COICM
• delay20 delay50

19
Table of Contents

• Introduction
• Related Work
• Diffusion OF Misinformation
• Eventual Influence Limitation
• Evaluation
• EIL With Incomplete Data
• Conclusion
• Strong Week Points
• Refrence

20
EIL With Incomplete Data

• sets of active, inactive and newly activated
  nodes for campaign C be denoted Agiven , Igiven
  Ngiven respectively
• Apred , Ipred Npred

21
Prediction Algorithms

• 1. Identifying A and I
• Three reasons 1) for identifying newly
  activated nodes
• 
  2,3) storage , Incorrectly identified
• A good heuristic should 1) nodes in Ap should
  form a connected component
• 2) have as many arborescences as possible

22
Prediction Algorithms(cont.)
1 Given (Agiven, Igiven,G,Ca) where G (N,E)
is the network graph, Agiven, Igiven are
the incomplete sets of active and inactive nodes
and Ca is an approximate value of A 2
Apred Agiven 3 Create a refined graph G '
that consists of nodes in N - I given 4 Select
a node ni at random from A 5 Tstein min
Steiner tree rooted at ni in G ' covering
Agiven 6 Nstein nodes in Tstein 7 Apred
Apred Nstein 8 while Apred lt Ca do 9
Nchoose ni ? N - Igiven - Apred 10 Apred
Apred argmaxn ? Nchoose deg(n) Apred 11
Output Apred
Ipred N - Apred
23
Prediction Algorithms(cont.)
2. Identifying N 1) In set Apred ? Shortest
average path ? bfs ? the leaves 2) Random
spanning tree on the Gpred ? the leaves
24
Predictive Hill Climbing Approach (PHCA)

• Agiven , Igiven Ngiven
• Identify ALP , the set of k nodes to influence by
  campaign L in graph G

25
Evaluation of PHCA

• Accuracy, precision and recall statistics
• Accuracy refers to the ratio of the nodes whose
  true states are correctly identified
• Precision refers to the ratio of nodes that are
  active
• recall refers to the ratio of nodes identified as
  active to the total number of active nodes
• with decreasing Pknown Greater amount of
  missing information

26
Evaluation of PHCA(cont.)

• Select the nodes that are unknown to be infected
• (a) Delay 30 (b) delay
  70

27
Table of Contents

• Introduction
• Related Work
• Diffusion OF Misinformation
• Eventual Influence Limitation
• Evaluation
• EIL With Incomplete Data
• Conclusion
• Strong Week Points
• Refrence

28
Conclusion

• Introduced PHCA algoritm
• Predicts the all the nodes of the network
• Then uses the PHCA to choose the set of
  influentials using the predicted data
• PHCA provides good performance, within 96-90
  that would be achieved with no missing
  information
• For large delays the performance degrades to 75

29
Table of Contents

• Introduction
• Related Work
• Diffusion OF Misinformation
• Eventual Influence Limitation
• Evaluation
• EIL With Incomplete Data
• Conclusion
• Strong Week Points
• Refrence

30
Strong Points

• Choose influentials ? largely achieved with no
  missing information
• Inactive nodes ? correct result
• Provider Two Model MCICM COCIM
• Paper presented with proof

31
Weak Points

• We have identified more with the possibilitie
• ? Results are not correct
• Work on a synthetic graph

32
Table of Contents

• Introduction
• Related Work
• Diffusion OF Misinformation
• Eventual Influence Limitation
• Evaluation
• EIL With Incomplete Data
• Conclusion
• Strong Week Points
• Refrence

33
Refrence
1 C. Budak, D. Agrawal, A. El Abbadi, Limiting
the Spread of Misinformation in Social
Networks,Department of Computer Science
University of California, Santa Barbara,
Santa Barbara CA 93106-5110, USA