Reputation Systems Guest Lecture Paul Resnick Associate Professor Univ. of Michigan School of Information
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Learning Objectives n Understand –What a reputation system is –Theory about when and why it should work –Open research questions n Participate in design –Recognize situations when it might be helpful –Raise some of the difficult design challenges
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Outline n What is a reputation system? n Theory: when/why they should work n Empirical results n Design space n Case study: NPAssist recommender
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Definition n A Reputation System… –Collects –Distributes –Aggregates n …information about behavior
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Examples n BBB n Bizrate n eBay n Expertise sites –Epinions “top reviewers” –Slashdot karma system
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Why Reputation Systems n Interacting with strangers n Sellers (Exchange Partners) Vary –Skill –Effort –Ethics
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Other Trust-Inducing Mechanisms in E-commerce n Insurance n Escrow n Fraud Prosecution
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu How Reputation Systems Should Work n Information n Incentive n Self-selection
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Some Issues n Anonymity n Name changes n Name trades n Lending reputations n Eliciting evaluation n Honesty of evaluations
Anonymity Analysis
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu 1L Pseudonyms n Third-party issues pseudonyms –No cost –Not replaceable –Reveal name to third party –Don’t reveal mapping of name to pseudonym
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Empirical Results: eBay n Feedback is provided n It’s almost all positive n Reputations are informative n Reputation benefits –Effect on probability of sale –Effect on price
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Provision of Feedback n Negatives: paid but did not receive; seller cancelled; not as advertised; communication n Neutrals: slow shipping, not as advertised, communication
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Feedback Profiles of Buyers and Sellers
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Predicting Problematic Transactions n Logistic Regression f(0,0) = 1.91% f(100,0) =.18% f(100,3) =.53% N = Beginning Block Number 0. Initial Log Likelihood Function -2 Log Likelihood Log Likelihood Dependent Variable.. NEGNEUT Variables in the Equation Variable B S.E. Wald df Sig R Exp(B) LNNPOS LNPOS Constant
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Predictive Value
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Some Recently Completed Work n Experiment: does reputation affect profit? –Many positives: Yes, but only a little (8.1%) –One or two negatives: No n Incentives for quality feedback provision –Can pay based on agreement among raters
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Studies Currently Underway n Feedback provision (empirical) –Reciprocation, altruism, and free riding n Dynamics: learning and selection (empirical) n Geography: trust and trustworthiness by state
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Design Space n Rating scales n Aggregation of ratings n Who rates? n Incentives for raters n Identification/Anonymity –Exchange partners –Evaluation providers
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Case Study n Goal: help Michigan non-profits select consultants and other service providers n Is this a good candidate for a reputation system?
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Case Study n Goal: help Michigan non-profits select consultants and other service providers n Is this a good candidate for a reputation system? Interacting with strangers Sellers (Exchange Partners) Vary Skill Effort Ethics
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Case Study Design Choices n Rating scales n Aggregation of ratings n Who rates? n Incentives for raters n Identification/Anonymity –Exchange partners –Evaluation providers
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu
SCHOOL OF INFORMATION UNIVERSITY OF MICHIGAN si.umich.edu Summary n RS inform, incent, select n Opportunity for RS: interactions with strangers n Design space –Scales, aggregation, raters, incentives, anonymity