BAYESIAN TRUST AND THE GRAMEEN MODEL: SOCIAL NETWORK METHODOLOGY APPLIED TO COMPUTER NETWORK TECHNOL

1
BAYESIAN TRUST AND THE GRAMEEN MODEL SOCIAL
NETWORK METHODOLOGY APPLIED TO COMPUTER NETWORK
TECHNOLOGY

• Project Presentation (ECE 568)
• Swapnil Dipankar
• Steve Adegbite
• Chaowu Lin
• Sameer Pai

2
Social vs. Computer Network

• Social Networks and Computer Networks are
  strongly related.
• Building Social Networks via Computer
  NetworksCreating and Sustaining Distributed
  Learning CommunitiesCaroline Haythornthwaite, The
  Grameen Model as Applied to Computer Networks
• A social network consists of any group of people
  connected through various social familiarities
• A computer network is a system for communication
  among two or more computers
• In a Social Network social bonds form connections
  between actors, while in a Computer Network, a
  network layer forms similar connections between
  nodes en.wikipedia.org

3
What is Trust?

• Trust A peers belief in another peers
  capabilities, honesty and reliability most often
  based on direct interactions with that peer
• Bayesian Trust A value directly proportional to
  the amount of trust in a peer and based on prior
  knowledge as derived from previous interactions
  with that peer.

Very Trusted
Trusted with my Life
Not Trusted
Trust 90
Trust 65
Trust (Current Trust Value Previous Trust Value)
4
Why Is Trust Important?

• Social Networks rely on trust because without
  trust, one would not know their friends from
  their enemies
• Computer Networks rely on trust to thwart
  mal-intentioned nodes from accessing network
  resources

5
Trust Models Found in Computer Networks

• Micropayment Model a naturally limited resource
  given in return for use of limited network
  resources
• Fungible Micropayments A small money amount is
  given in return for use of a network resource
• Non-Fungible Micropayments A small amount of
  time given in return for use of a network resource

22t mod n

• n is a product of two large primes
• t is proportional to the time required to solve

6
Trust Models Found in Computer Networks Continued

• Reputation Model Distributing or allowing
  access to network resources in direct proportion
  to the amount of Bayesian trust held in a node
• AOL Instant Messenger Warning Level
• Some P2P File Sharing Programs
• Grameen Bank Model Applied to Computer Networks

7
Our Model

• The concept of our network stems from two main
  models
• The Bayesian Model
• The Grammen Banking model

8
The Bayesian Model

• The Bayesian Model
• Offers a very flexible and logical way to combine
  different trust representations and interests
• Basically a relationship network which uses
  concepts from statistics to represent probability
  of interactions between various nodes thus the
  reason for its adaptation

9
The Grameen Banking Model

• Totally different concept from conventional
  banking methods
• Parallelism of the Grammen Banking with our
  Network model
• The requisite of mutual trust, accountability
  and participation
• Ease of initiation into a group
• The need for a continual renewal of trust

10
The Grameen Banking Model (Continued)
11
Interactions within a Network
Social Network Group 2
Social Network Group 1
Social Network Group 3
Social Network Group n
12
Underlying Assumptions in our model

• Trust and reputation are dynamic processes.
• Different issues of concern and areas of
  interest are foreseen in the advent of
  scalability of our Network simulator

13
Key Features of our model

• Efficiency considerations such as
• Evicting bad nodes in the minimum number of
  steps possible
• Maximizing successful interactions
• Minimizing unsuccessful interactions
• Adaptability and scalability
• Flexibility in the use of various configurations

14
Math Model

• Random Graph distribution model
• Trust model

15
Math Model

• Part ? Random Graph distribution model
• NodeA people in the social network
• NNumber of nodes
• Edges connections between the nodes
• Degree of a node Number of connections of it
• pthe independent probability of connecting edge

16
Math Models

• Part ? Random Graph distribution model(Cont)
• binomial function
• Where kdegree of a node
• Pkprobability distribution of
  a node
• It can be look as a Poisson distribution when p
  is small and N is large.
• Where z(N-1)p is the average degree of a
  node
• Design a large number of nodes who are random
  distribution in a large area
• Randomly only select a group inside of it
• This group is distributed by nodes who can
  connect with each other

17
Math Models

• Part ? Trust model
• The Theoretical foundation of trust model is
  Bayes rule
• where
• P(HE) - posterior probability of a hypothesis H
  after considering the evidence E,
• P(EH) is the likelihood and gives the
  probability of the evidence E assuming H
  (conditional probability),
• P(H) is the prior probability of H alone,
• P(E) is a normalising or scaling constant,
• P(E) is a normalising constant to ensure that the
  posterior probability adds up to 1 and is
  computed by summing the numerator over all
  possible values
• Consider the rule If H is true then E is true
  (with probability p) (i.e. If event H occurs then
  the probability that event E will occur is p)

18
Math Models

• Part ? Trust model
• When node A is not sure of node B, it will ask
  other nodes C for reference.
• 
  (C are other nodes in the same group,
  node A may not know all of them)
• If the reference untrustworthy, recommendations
  discarded
• Node A need to combine the Recommendation Rtotal
• Total Recommendation
• 
  Where
  k total number of trustworthy references
• 
  
  g total number of unknown references
• 
  
  the trust the node has in the node
  trustworthy
• 
  
  reference,
• 
  
  the trust the node trustworthy
  reference has in
• 
  
  node (node B),
• 
  
  the trust that node unknown
  reference has in node
• 
  
  (node B),
• 
  
  weights of the trustworthy reference,
• 
  
  weights of the unknown reference
• if Rtotal ?, (? is threshold value) ,the node A
  will interact with the node B, otherwise not.
• If the node A interacts with the Node B, it
• Updates its trust in the Node B,

19
Math Models

• Part ? Trust model
• Trust in reference
• Reference refer to the nodes who give
  recommendations
• Measures whether a node can provide
  reliable recommendations.
• 1. Trust evaluated by one interaction
• the trust value that the ith node has in
  the lth trustworthy reference
• new trust value that the ith node
  has in jth referencee after the update
• old trust value that the ith node has
  in jth reference.
• a learning rate real number in the
  range 0, 1.
• ea new evidence value,
• 1, successful interaction, -1,
  otherwise.

20
Math Models

• Part ? Trust model
• 2.Trust evaluated by frequent discussion
  (gossip)
• Nodes in the group know each other.
• Nodes exchange and compare opinion frequently,
• This can help them find other nodes who share
  similar preferences more accurately and faster.
• After each comparison, the nodes update their
  trusts in each other according to the following
  formula.
• new trust value that
  the ith node has in jth reference after the
  update.
• old trust value that
  the ith node has in jth reference.
• ß learning rate real
  number in the range 0, 1 (ßa).
• eß the new evidence value,
  
• number in the
  interval -1, 1.

21
(No Transcript)
22
Start
Termination condition met?
Choose a Random Client
Evict the bad node from the group.
Choose a Random Server
Is trust less than threshold?
Start Client-Server Interaction
Update sever trust value
Is the server file corrupt?
Update successful interactions
Update unsuccessful interactions
Normal Mode Outline
End
23
Start
Termination condition met?
Choose a Random Client
Evict the bad node from the group.
Choose a Random Server
Is trust less than threshold?
Start Client-Server Interaction
Update sever trust value
Is the server file corrupt?
Update successful interactions
A
Update unsuccessful interactions
A
Update Server Trust Database
End
Bayesian Model Outline
24
Start
Gossiping condition met?
Termination condition met?
Gossip
Evict the bad node from the group.
Is trust less than threshold?
Choose a Random Client
Choose a Random Server
Start Client-Server Interaction
Update sever trust value
Is the server file corrupt?
Update successful interactions
A
Update unsuccessful interactions
A
Update Server Trust Database
End
Bayesian Model with Gossiping Outline
25
Client-Server Interaction
Go through the clients Distrusted nodes list.
Is server ID in that list?
Make the transaction
Is the file corrupt?
Successful transaction
Add server to the Trusted nodes list.
Unsuccessful transaction
Add server to the Distrusted nodes list.
Transaction avoided.
Update Server trust value
The Bayesian Model
Proceed
26
Check for condition of gossip
Condition met?
Iterate through each node in the group
Compare its distrusted nodes list with the
distrusted nodes list of its trusted nodes
Update the Client distrusted nodes list
Proceed
Gossiping Model
27
Results
28
Results continued
29
Results continued
30
Results and Conclusions

• Our model is more efficient in terms of
  eliminating bad element increasing the successful
  transactions and decreasing bandwidth misuse.
• Our model enhances the currently existing trust
  and reputation model.
• Our model does not need any member to refer, the
  bad behavior is automatically taken care of.

31
References

• Complexity Theory and Models for Social Networks
• Emerging Social Networks in Peer-to-Peer Systems
• Random Graph Models of Social Networks.
• Dynamics of Social Networks A Deterministic
  Approach.
• Identity and Search in Social Networks
• Referral Web - Combining Social Networks and
  Collaborative Filtering
• Searching Social Networks
• Settings In Social Networks - A Measurement Model
• Social Networks - From the Web to the Enterprise
• Social networks in the Virtual Science Laboratory
• Social Networks
• The Link Prediction Problem for Social Networks
• The Structure of Growing Social Networks
• The Web and Social Networks
• Wallop - Designing Social Software for Co-located
  Social Networks
• Wearable Communities - Augmenting Social Networks
  with Wearable Computers
• Electronic Mail and Social Networks
• Trust and Reputation Model in Peer-to-Peer
  Networks