ITEC 810 Workshop Paper: A Survey of Web-based Social Network Trust

1
ITEC 810 Workshop PaperA Survey of Web-based
Social Network Trust

• Student Eric Wang
• 41176774
• Supervisor Yan Wang
• Subject Coordinator Robert Dale
• 5th Jun 2009

2
Agenda

• Trust and WBSN
• Trust inference in a WBSN
• Survey of trust inference mechanisms
• Findings and discussion
• Possible area for future studies
• Conclusion
• Question time

Slide 2
3
Problem

• How trustworthy is the opinion of a friend of my
  friend who I do not have direct relationship
  with?
• How to infer trust on people I do not have direct
  relationship via the people I do have direct
  relationship?
• How to accurately estimate this inferred trust
  without negative impact from malicious users?

Slide 3
4
Trust and WBSN

• Trust
• Have confidence or faith in a person or a piece
  of information
• Web-based Social Network
• MySpace, Facebook, LinkedIn
• Decisions are made based on trust
• Which movie should I watch next weekend?
• Should I buy Windows Vista or wait for Windows 7?
• How can we estimate trust from someone not
  directly connected to me in a WBSN?

Slide 4
5
Properties of trust in a WBSN

• Asymmetry
• Personalisation
• Transitive

Slide 5
6
Trust inference in a WBSN

• Trust inference in a trust diagram

Slide 6
7
Survey of Trust Inference Mechanisms

• 9 trust inference mechanisms
• TidalTrust
• Binary Trust Algorithms
• Advogato Trust Metric
• Appleseed
• SocialTrust
• FuzzyTrust Algorithm
• SUNNY
• Trusted Gossip
• RN-Trust
• We select 5 to present here

Slide 7
8
TidalTrust

• Jennifer Golbeck, 2005
• Simple algorithm - scalable
• Generic applicable to any WBSN
• Applies two conditions to increase accuracy
• Only consider the shortest path
• Trust threshold limit on highly trusted
  neighbours
• Most frequently cited and compared by others

Slide 8
9
SocialTrust

• James Caverlee, et al, 2008
• Distinguishing relationship quality from trust
• Incorporating personalised user feedback
• Three components
• Current quality component
• History component
• Adaptation to change component
• More robust against malicious attacks than
  PageRank from Google

Slide 9
10
FuzzyTrust Algorithm

• James Caverlee, et al, 2008
• Fuzzy linguistic terms instead of numeric trust
  values
• Computes trust from stronger and shorter paths
• More meaningful information when there is
  contradictory information

Slide 10
11
Trusted Gossip

• Arindam Mitra and Mucumaru Maheswaran, 2007
• Encourages spread of good information while
  restrict spam messages
• Perform more than one step of trust estimation
• Bayesian trust estimation
• Recommender system
• Three approaches on where node-level and message
  level filtering applied
• Receiver Initiated
• Sender Initiated
• Using both as Hybrid
• Robust against reputation distribution

Slide 11
12
RN-Trust

• Mohsen Taherian et al, 2008
• Electrical resistance network model
• Simple and generic
• Scalable
• Improve on weakness of TidalTrust
• All paths considered, not only shortest paths

Slide 12
13
Findings and discussion

• Approach of trust inference estimation
• Only trust values in the shortest or strongest
  paths
• Perform more than one step of trust estimation
• Include user feedback
• Fuzzy linguistic terms instead of numeric trust
  values
• Electrical resistance model
• Merits and Weaknesses
• Increased accuracy
• Simplicity
• Scalability
• Dynamic update by feedback
• Robust against reputation distribution
• Robust against malicious attacks

Slide 13
14
Possible topics for future studies

• Further improve accuracy and precision
• Develop event based trust inference algorithm
• Develop context aware extensions
• Other applications of inference trust
• Military Intelligence
• Data mining
• Computing requirement
• CPU process
• Storage
• Network

Slide 14
15
Conclusion

• Define trust and its properties in WBSN
• Survey of existing trust inference mechanisms
• Merits and weaknesses in existing trust inference
  mechanisms
• 5 possible topics for future studies

Slide 15
16
Question Time

• Question from the audience

Slide 16