Lightweight Distributed Trust Propagation
Download as PPT, PDF3 likes1,100 views
1) The document proposes a new method called Lightweight Distributed Trust Propagation for setting initial trust values between devices based on their relationships to other devices. 2) It involves building a graph of trust relationships and defining a "good" rating function f that minimizes loss and rates neighboring nodes similarly. 3) Testing on real data with over 55,000 user ratings showed the method was useful, robust, fast and lightweight, requiring only 30KB of data transmission and 2.8ms of run time in worst cases.
Lightweight Distributed Trust Propagation
- 1. Lightweight Distributed Trust Propagation (daniele quercia & stephen hailes & licia capra) U C L
- 2. What I do...
- 6. Reputation Systems for Mobiles
- 7. Daniele Quercia Situation : People exchange digital content
- 8. Daniele Quercia Problem : Content overload
- 9. Daniele Quercia Proposal: Accept content only from reputable people
- 10. Daniele Quercia Take two devices… A B
- 11. Daniele Quercia A B Shall I accept content from B ?
- 12. Daniele Quercia A B A should set its initial trust for B
- 13. Daniele Quercia A B A should set its initial trust for B Yes, but how?
- 14. Daniele Quercia Traditional way: Trust propagation ? A B C
- 15. Daniele Quercia That way works on the Web and has been tested on “binary” ratings
- 16. Daniele Quercia If that way is tested on 3-level ratings
- 17. Conclusion: A needs a new way for propagating its trust in B
- 18. Conclusion: A needs a new way for propagating its trust in B But which way?
- 19. Our proposal
- 20. Daniele Quercia 1 ? A B C 2 A B A C C B
- 21. Daniele Quercia 1 ? A B C 2 ? Put trust relationships in a new graph & label graph nodes 1 2 A B A C C B
- 22. Daniele Quercia 1 ? A B C 2 ? Put trust relationships in a new graph & label graph nodes Define a “good” function that rates A B f 1 2 A B A C C B
- 23. Daniele Quercia 1 ? A B C 2 ? build_graph ( trust_relationships, A B ) f find_f 1 2 A B A C C B
- 24. <build_graph>
- 25. A B D C 3 2 1 1 2 ? A B ? ? 2 3 1 0.5 1 M M
- 26. A B D C 3 2 1 1 2 ? A B ? ? 2 3 1 0.5 1 M M What are those ?
- 27. Remember, we’ll define f We will see that f rates neighbouring nodes alike It’s a neighbour of A B; therefore, it rates as A B does ?
- 28. Remember, we’ll define f We will see that f rates neighbouring nodes alike It’s a neighbour of A B; therefore, it rates as A B does ? Find relationships that rate as A B does
- 29. Remember, we’ll define f We will see that f rates neighbouring nodes alike It’s a neighbour of A B; therefore, it rates as A B does ?
- 30. 2 Types of those relationships
- 31. 1. The relationships with same rater ( A ) A B D C 3 2 1 1 2 As long as B and D behave alike
- 32. A B D C 3 2 1 1 2 2. The relationships with same rated node ( B ) As long as A and C rate alike
- 33. OK, 2 Types
- 34. Taking those relationships, one comes up with …
- 35. … a graph! A C A B A D C B 2 3 1 0.5 1 M M
- 36. </build_graph>
- 37. <find_f>
- 38. A C A B A D C B 2 3 1 0.5 1 M M
- 39. Take f that returns the ratings already there ( ) & similar ratings for neighbouring nodes A C A B A D C B 2 3 1 0.5 1 M M 2 3
- 40. More formally, …
- 41. We find f that minimizes that loss!
- 42. </find_f>
- 43. Daniele Quercia 1 ? A B C 2 ? build_graph f find_f 1 2 A B A C C B
- 44. Does it work? Useful, Robust, Fast, “Light”?
- 45. Daniele Quercia Useful? Tested on real data (Advogato: > 55K user ratings)
- 46. Daniele Quercia Useful? Tested on real data (Advogato: > 55K user ratings)
- 49. Daniele Quercia Fast and “ Light ”?
- 50. Daniele Quercia Fast and “ Light ”? For propagating A B (worst case) Transmit 30KB & run for 2.8ms
- 51. Daniele Quercia All this on …
- 52. Daniele Quercia All this on …
- 53. Daniele Quercia