1 / 28

A New Approach for Trust Calculation in Social Networks

A New Approach for Trust Calculation in Social Networks. Mehrdad Nojoumian (student) Timothy C. Lethbridge (supervisor) University of Ottawa, Canada tcl@site.uottawa.ca. Objectives of this talk. Explore the behavior of various trust calculation approaches

brasen
Download Presentation

A New Approach for Trust Calculation in Social Networks

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. A New Approach for Trust Calculation in Social Networks Mehrdad Nojoumian (student) Timothy C. Lethbridge (supervisor) University of Ottawa, Canada tcl@site.uottawa.ca Trust Calculation - Nojoumian and Lethbridge

  2. Objectives of this talk • Explore the behavior of various trust calculation approaches • Describe an approach that has an improved combination of characteristics. Trust Calculation - Nojoumian and Lethbridge

  3. Some definitions • Social network • Nodes are actors (buyers, sellers, partners, brokers) • Arcs are relationships (buying, selling, advising, consulting, sharing, etc.) • Reputation: Perception an agent has of another’s intentions • Derived from one’s own observations and those in one’s social network • Reputation is a social quantity, but everyone has their own perception of it Trust Calculation - Nojoumian and Lethbridge

  4. Trust • Personal expectation about another’s behavior in a particular encounter (Mui 2002) • Derived from reputation • Parties in a transaction must establish trust to do business effectively • If party A has low trust of party B, party A will be willing to pay party B less, and will need to consider insurance • So party B has an incentive to be trustworthy Trust Calculation - Nojoumian and Lethbridge

  5. Reputation systems • Gather experiences from participants as transactions take place • Trustworthy agents increase in reputation • Untrustworthy agents drop in reputation • Reputation systems can be ‘centralized’ • E.g. in EBay, sellers receive ratings (-1, 0,1) for reliability. • Reputation can be the sum or some other function of those ratings Trust Calculation - Nojoumian and Lethbridge

  6. Decentralized reputation systems • A1 can query others who have transacted with A2 • Overall reputation can be a combination of A1’s: • Direct experience with A2 • Feedback from others who have interacted with A2 • Reputation of others (A3, A4 and A5) as witnesses Trust Calculation - Nojoumian and Lethbridge

  7. Trust is built up over time • Through a series of transactions • Co-operations (C) = good experiences with the agent in question • Delivery occurred in a timely manner • Merchandise was as advertised • Payment was received in full and on time • Acted as a truthful or reliable witness • Defections (D) = bad experiences • Delivery excessively late • Merchandise wrong or inferior to expectation • Payment excessively late or not received • Acted as an untruthful or unreliable witness Trust Calculation - Nojoumian and Lethbridge

  8. A sample trust function from the literature • Y&S: Yu and Singh (2000) • Compute Tt+1 = f(Tt, CorD) Trust Calculation - Nojoumian and Lethbridge

  9. Effect of Y&S = 0.1 and  = -0.2 Increment in trust after cooperation Decrement in trust after defection Trust value before transaction Yellow region: The better you are, the less co-operation benefits Trust value after cooperation Trust value after defection Yellow region: The worse you are, the less defection costs

  10. Y&S ‘Increment’ view = 0.1 and  = -0.2 Increment on defection Increment on cooperation Trust value, Tt Trust Calculation - Nojoumian and Lethbridge

  11. Y&S ‘Next value’ (Tt+1) view = 0.1 and  = -0.2 Next value on defection Next value on cooperation Trust value, Tt Trust Calculation - Nojoumian and Lethbridge

  12. Y&S ‘Sequence’ view = 0.1 and  = -0.2 • Sequences of • 30 cooperates • 10 cooperates + 10 defects + 10 cooperates • 30 defects Inflection point Penalty for D after C Trust Calculation - Nojoumian and Lethbridge

  13. A new proposed formula family: N&L (Nojoumian and Lethbridge) • Key changes: • As trust increases above threshold , keep increasing the reward for co-operation • Up to the maximum • As trust decreases below threshold keep increasing the cost of defection • Down to the minimum • Between  and keep cost and reward fixed Trust Calculation - Nojoumian and Lethbridge

  14. The N&L trust function:In case of Cooperation • Tt in [-1, ) • Bad agent (for now): Encourage • Reward increases linearly from • Xencourage (default 0.01) to • Xgive (default 0.05) • Tt in [, ] • Agent about which you are indifferent: Give Xgive • Tt in (, +1] • Good agent: Reward • Reward increases linearly from • Xgive to • Xreward (default 0.09) Trust Calculation - Nojoumian and Lethbridge

  15. The N&L trust function:In case of Defection • Tt in [-1, ) • Bad agent: Penalize • Increment increases linearly from • Xpenalize (default -0.09) to • Xtake (default -0.05) • Tt in [, ] • Agent about which you are indifferent: Take Xtake • Tt in (, +1] • Good agent (for now): Discourage • Increment increases linearly from • Xtake to • Xdiscourage (default -0.01) Trust Calculation - Nojoumian and Lethbridge

  16. Comparison of ‘increment’ views = 0.1 and  = -0.2 Next value on defection Next value on cooperation Y&S N&L -1 Trust value, Tt +1 -1 Trust value, Tt +1 Trust Calculation - Nojoumian and Lethbridge

  17. Comparison of ‘next value’(Tt+1) views = 0.1 and  = -0.2 Increment on defection Increment on cooperation Y&S N&L -1 Trust value, Tt +1 -1 Trust value, Tt +1 Trust Calculation - Nojoumian and Lethbridge

  18. Comparison of ‘sequence’ views = 0.1 and  = -0.2 Yellow: 10C 10D 10C Y&S N&L ‘Maxed out’ Inflected asymptotic Less severe penalty for D after C, but can be adjusted

  19. Effect of adjusting N&L parameters: 0.1 to 0.3 and  -0.2 to -0.4 Original Result Longer indifferent period Slight delay only

  20. Effect of adjusting N&L parameters:Xencourage 0.01 to 0.015 and Xpenalize -0.09 to -0.15 Original Result Slight effect of increased encouragement Effect of increased penalty Larger D after C penalty

  21. Effect of different N&L sequencesXencourage remains 0.015 and Xpenalize remains -0.15 20C10D end <0 10C10D10C end >010D20Cend <0 10C20D10D10C10D Trust Calculation - Nojoumian and Lethbridge

  22. Same sequences from Y&S function 20C10D end <0 10C10D10Cend <010D20C end >0 10C20D10D10C10D Trust Calculation - Nojoumian and Lethbridge

  23. Microsoft Excel formula for calculating N&L trust values =prevTrustValue+(IF(CorD="C", MIN(1-PrevTrustValue, IF(GoodOrBad="B", X_encourage+(PrevTrustValue+1)/(beta2--1)*(X_give-X_encourage), IF(GoodORBad="I", X_give, X_give+(PrevTrustValue-alpha2)/(1-alpha2)*(X_reward-X_give) ))), MAX(-1-Y50, IF(GoodORBad="B", X_penalize+(PrevTrustValue+1)/(beta2--1)*(X_take-X_penalize), IF(GoodORBad="I", X_take, X_take+(PrevTrustValue-alpha2)/(1-alpha2)*(X_discourage-X_give) ))) )) Trust Calculation - Nojoumian and Lethbridge

  24. You can simplify calculations by using an approximation • Results of quadratic regression for the N&L for default parameters Trust Calculation - Nojoumian and Lethbridge

  25. Varying the function for varying transaction value • E.g. You could apply the formula N=floor(Log10(V)) times where V is the transaction value • I.e. • $10-$99 - Apply once • $100-$999 - Apply twice • $1000-$9999 - Apply 3 times • Etc. • The base of the logarithm can be changed for different effects Trust Calculation - Nojoumian and Lethbridge

  26. Main drawback of N&L • ‘Maxing out’ or ‘hitting rock bottom’ • No further increase in trust after you reach 1 • No further decrease in trust after you reach -1 • Asymptotic approach corresponds to ‘diminishing returns’ • Could be rectified by making the function open-ended Trust Calculation - Nojoumian and Lethbridge

  27. Conclusions - • It seems reasonable to consider that trust functions should • Reward more (or same) the better an agent is • Penalize more (or same) the worse an agent is • Y&C trust function does not have these properties • But has asymptotic approach / diminishing returns

  28. Conclusions - 2 • We propose a family of trust functions • Reward always increases the better an agent is, and vice-versa • Eight parameters can be adjusted to fine tune behavior • Future work: • Empirically evaluate the ability of the variously parameterized Y&C or N&L functions to predict actual trustworthiness Trust Calculation - Nojoumian and Lethbridge

More Related