Social web 2.0 course notes:

• Trust and Reputation Systems
• Trust
  • A psychological state comprising the intention to accept vulnerability based upon positive expectation of the intentions or behavior of another
    • Process-based (past history of interaction)
    • Character-based (social similarity)
    • Institution-based
  • Entity (person, agent) vs content trust
  • Transitivity
    • Trust in performance (less so)
    • Trust in belief (more so)

• Stages of Trust in Site
  • Preliminary assessment (heuristic, affective)
    • Look and feel of site
    • Branding, familiar, trusted logos etc.
  • In-depth evaluation of information (analytic)
    • Quality of information
    • Personalization of advice, given by similar others
  • Long-term relationship with site

• Reputation Systems Online
  • Online interactions outside usual social constraints (disembodied)
    • Identified behavior
    • History of behavior over time
    • Social context: face-to-face increases normative behavior
  • People *will* break trust if not held accountable/ prosocial norms not activated by presence of others
  • Reputation
    • History of past interactions informs current expectation of reciprocity or retaliation in future
    • Accountability, trust
• Reputation Systems — Key Components
  • Long-lived entities that inspire expectation of future interaction
  • Capture and distribution of feedback about current interactions
  • Use of feedback to guide trust decisions
  • Issues:
    • Low incentive to provide feedback
    • People reluctant to provide negative feedback
    • Ensuring honest reports
• Types of Ratings
  • Implicit Ranking
    • Time in system, frequency of visits, frequency of posts, etc
  • Explicit Rating
    • Weighted average, explicit rating of object of interest
  • Collaborative filtering
    • People with similar rating patterns rate this highly, so you will probably like
    • Assumes high variability in preferences
  • Peer-based
    • Filter implicit/explicit ratings by relevance to self in network (e.g. friend of friend)

• Design Implications

• • “Look and feel” matters, at-a-glance judgments impact continuing analysis
  • Expose “related entities” around any content, with indicators of credibility
  • Filter both content and reputation metrics by relevance to self — emphasizing similarity
    • Often reduced overall average ratings the more information is exposed (voice, picture, profile information): indication of increased discrimination between good/bad, relevant content
  • Include both implicit and explicit ratings/rankings
  • Expect explicit ratings to be positively biased, so “absence of positive” matters
    • Ratings per hit rate for example meaningful
    • Count of ratings overall
    • Binary votes: e.g. “useful” or not
  • Metrics at both level of content and level of author important
  • Rate comments as well as content
• Opportunities for Innovation
  • Assessing a person’s/story’s reputation with “others like me” – localized reputation
  • Under the hood assessment of “trustability” of raters, use to influence their influence on aggregate scores, search results
    • Recency in system, deviance, phase of treatment, explicit ratings (ratings of raters)
  • Use interaction history with content to normalize ratings
    • % of positive ratings out of # of people read/hit vs. simple average
  • Search results, able to change sort by:
    • Overall ranking/ratings
    • Ranking/rating in my network
    • Similarity/relevance to me
    • Date updated/posted
    • Author

Recommended Readings

Jensen, C., Davis, J. P., & Farnham, S. D., Finding Others Online: Reputation Systems for Social Online Spaces, CHI 2002, ACM Press (2002), 447-454.

Resnick, P, Kuwabara, K., Zeckhauser, R., Friedman, E.. (2000). Reputation Systems. In Communications of the ACM. Vol. 43, 12, pp. 45-48. ACM Press.