Transcript A Social Network-Based Trust Model for the Semantic Web

A Social Network-Based
Trust Model for the
Semantic Web
Yu Zhang, Huajun Chen, and Zhaohui Wu
Grid Computing Lab, College of Computer
Science, Zhejiang University
Speaker: Yi-Ching Huang
Outline
Introduction
Related Work
Trust Model
Basic Definitions
Basic Mechanisms
Conclusion
Introduction
Trust is essential to secure and high quality
interactions on the Semantic Web
Semantic Web can be view as a collection of
intelligent agent
RDF (Resource Description Framework)
machine-understandable
Example: aspirin can cure headache
effectively
subject
aspirin
cure
predicate
headache
object
Introduction
contribution
increase efficiency
evaluate trust from two dimensions
exploit formulas in probability and statistics
provide an algorithm to compute trust
values simultaneously
Related Work
Small World - Milgram’s experiment(1960s)
FilmTrust
EigenTrust algorithm
a reputation management algorithm for P2P
networks
focus on security problems
Trust Model
Source: FOAF data
RDF/XML Semantic Web vocabulary
Easy to process and merge by machine
Allows users to specify who they know and build a
web of acquaintances
Use a graph to describe a social structure
G = (V, E)
V: resources, E: predicates
Basic Definitions
Def. 1 : Trust Rating
which degree a consumer’s evaluation about a
provider’s ability
Def. 2 : Reliable Factor
which degree that a consumer agent believes the
trust information
Def. 3 : Neighbor
Def. 4 : Friend
Example: Neighbor and
Friends
Basic Mechanisms
Local Database Storage
Trust Report Mechanism
Routine Report
Update Report
On-demand Report
Pull Mode and Push Mode
Honor Roll and Blacklist
Local Database Storage
•
Linked List
Trust Report Mechanism
Problem: Semantic Web is “openness”
it is hard to know whether our past
experience is valuable or meaningless
3 types
Routine Report
Update Report
On-demand Report
Pull Mode and Push
Mode
Pull mode
when the consumer needs some trust
information, it takes the initiative to “pull”
trust news from its acquaintances
Push mode
the publisher pushes the trust information
directly to the consumer
Honor Roll and Blacklist
•
Problem: it consumes much time to calculate trust
values and transfer information
•
Want to speed up the process
•
Solutions
•
Honor Roll: behave well
•
Blacklist: behave badly
•
Both store in linked list in the local database
•
Need to update
The Algorithm of the
Trust Model
Easy-to-compute
Avoid client to wait the results
Parallel arithmetic
Use BFS to expands out from source to sink
through the trust network
All the path compute trust values simultaneously
Two dimensions
Trust rating
Reliable factor
The Algorithm of the
Trust Model
N: # of paths from P to Q
Di: # of steps between P and Q
Wi: weight of the i-th path
Mi: Q’s immediate friend or neighbor on the
i-th path
Similarity of Preference
Probability and statistics theory
Assumption: the closer of each pair of trust
ratings, the more similar of the two agents
Define
| E(x) | < 0.3 and S(x) < 0.1
Similarity of Preference
Similarity of Preference
Similarity of Preference
•
A and B are similar
•
A and C are not similar
Conclusion
the algorithm of the model is simple, efficient
and flexible
the trust model does not provide a mechanism
to deal with lying or betrayal of agents
plan to incorporate reasoning and learning
abilities