Accession Number:



Robust Reputations for Peer-to-peer Markets

Descriptive Note:

Doctoral thesis

Corporate Author:


Personal Author(s):

Report Date:


Pagination or Media Count:



This thesis investigates on-line reputation systems for peer to-peer markets and presents a number of systems which increase robustness against attacks on individual reputations while distinguishing between honest and dishonest user behavior. The fluidity of identity on-line and unavailability of practical legal recourse make evaluating trust and risk in on-line markets both vital and difficult. Reputation systems have been proposed as one possible means of building trust among strangers by aggregating the experience of many users. However, the attributes that make reputation systems helpful also make them a target for fraud. A good reputation is valuable, so some users may try to circumvent the system. We look at two specific ways in which users attack reputation systems in peer-to-peer markets and discuss ways in which the damage can be mitigated. We first address retaliatory negative feedback, where a user leaves a negative feedback for someone who complained about their behavior. We show that allowing retaliation can result in a reputation system that is incapable of identifying low-quality users and allows cheating to go unpunished. We then present EM-Trust, a system that is better able to estimate true user quality even with high levels of retaliation. We next look at the issue of sybil attacks, where a single user creates a large collection of identities to increase his own reputation. We show that EigenTrust, a widely discussed algorithm that purports to resist similar collusion attacks, does not work against sybils. We then present Relative Rank, a transformation of EigenTrust that is both sybil resistant and better suited to peer-to- peer marketplaces. Finally, we discuss RAW, a variation of PageRank that offers additional guarantees of sybil-resistance. We demonstrate that it is possible to design reputation systems that are as effective as existing non-robust ones at discriminating between honest and dishonest user behavior.

Subject Categories:

  • Operations Research
  • Computer Programming and Software
  • Computer Systems
  • Computer Systems Management and Standards

Distribution Statement: