marin stamov cs 765 nov 14 2011. goals my model believe trust tolerance the simulation expected...
TRANSCRIPT
Information spread in social networks (part 2)
Marin StamovCS 765Nov 14 2011
OutlineGoals
My modelBelieveTrustTolerance
The simulation
Expected results
Conclusion
GoalsModel that allows the agents to change their believe in the information truthfulness
Useful for testing the spreading of information such as rumors, expectations, prognosis and other uncertain kinds of information
How the network topology affects the spread and the average believe of the agents
How will believe and disbelieve interfere with each otherDo we need large starting seed if for successful spread
My model
Spreading of a probability, based on connection trust
Uses weighted undirected network
Allows agents to be prejudiced
Different then threshold or binary like models
BelieveRepresented by a single number from 0 to 100
Estimate the chance that this information is truthful
Disbelieve is also an information which can spread the network
Disbelieve Believe0 50 100
Believe example
Believe
I am 90 % sure the president of our company will
be reelected
Disbelieve Believe
Based on my previous
believes and how much I trust him I
would say 75%
TrustThe weight of each edges (0 to 100)
How much the neighbor will affect our beleive
Difficult to obtain in the real world
Relationship between trust and information spread
Tolerance
Willingness of the agent to change his current believes
Opinion confirmation should be accumulated
Useful for representing forceful agents
75
77
73
78
72
20
Road mapGoals
My modelBelieveTrustTolerance
The simulation
Expected results
Conclusion
The simulationSmall scale
Easier to visualize Monitor each step
Large scaleMay show different results
Test different network topologies
My simulation programWritten in c++ (QT)Works with .net filesCan represent graphically the network
The simulationB2`=50(T/100+1)(B1-B2)/100+B2
90
6347
57 71
35
79 82
72 70
80
41
64
Expected results
Improve the model based on the data from the simulations
What topologies are best and worst for good spread
How the average believe changes over timeCreate graphicsEach activated agent use one of his edges at each step
ConclusionWe spread information, but we measure the probability that the information is true based on each agent estimation
Believe is the most important parameter in this model
Trust of the connection is important for the calculating of the estimated believe, but other parameters can also be used
Threshold can be used on the values of believe and trust
References[1] Daron Acemoglu,Asuman Ozdaglar, Spread of (Mis)Information in Social
Networks Games and Economic Behavior 7 (2010)[2] D. Acemoglu, Munther Dahleh, Ilan Lobel, Bayesian learning in social
networks, Preprint, (2008)[3] A. Banerjee and D. Fudenberg, Word-of-mouth learning, Games and
Economic Behavior 46 (2004)[4] V. Bala and S. Goyal, Learning from neighbours, Review of Economic Studies
65(1998)[5] A. Banerjee, A simple model of herd behavior, Quarterly Journal of Economics
107(1992)[6] S. Sreenivasan, J. Xie, W. Zhang, Influencing with committed minorities, NetSci
(2011)[7] Cindy Hui, Modeling the Spread of Actionable Information in Social
Networks, (2011)[8] Lada Adamic, Co-evolution of network structure and content, NetSci (2011)[9] Andrea Apolloni, Karthik Channakeshava, Lisa Durbeck, A Study of Information
Diffusion over a Realistic Social Network Model, Computational Science and Engineering, (2009). CSE '09
Questions ?Thank you!