More about the eBay feedback model and trust metrics attacks

My last post reminded myself of some paragraphs I wrote in a paper some time ago. I know my writing ability is not comparable to Shakespeare’s one but maybe you find some interesting information in this passage from A Survey of Trust Use and Modeling in Current Real Systems reported below:

EBay’s feedback ecology is a large and realistic example of a technology mediated market. The advantage of this is that a large amount of data about users’ interactions and behaviors can be recorded in a digital format and can be studied. In fact, there have been many studies on eBay and in particular on how the feedback system influences the market, see for example (Resnick and Zeckhauser, 2002). A very interesting observation is related to the distribution of feedback values: “Of feedback provided by buyers, 0.6% of comments were negative, 0.3% were neutral, and 99.1% were positive” (Resnick and Zeckhauser, 2002). This disproportion of positive feedbacks suggests two considerations: the first is actually a challenge and consists of verifying if these opinions are to be considered realistic or distorted by the interaction with the media and the interface. We will discuss this point later in Section 3. The second is about possible weaknesses of the eBay model. The main weakness of this approach is that it considers the feedback of every user with the same weight, and this could be exploited by the malicious user. Since on eBay there are so few negative feedbacks, a user with just few negative feedbacks is seen as highly suspicious and it is very likely nobody will risk into engaging in a commercial transaction with her. Moreover, having an established and reputable identity helps a lot the business activity. A controlled experiment on eBay (Resnick et al., 2003) found that an high reputation identity is able to get a selling price 7.6% higher than a newcomer identity with little reputation. For this reason, there are users who threaten to leave a negative feedback (and therefore destroy the other user’s reputation) unless they get a discount on their purchase.
This activity is called “feedback extortion” on eBay’s help pages (“EBay help: Feedback extortion”, n.d.) and in a November 2004 survey (Steiner, 2004) 38% of the total respondents stated that they had “received retaliatory feedback within the prior 6 months, had been victimized by feedback extortion, or both“.
These users are “attacking” the system: as eBay’s help page puts it “Feedback is the foundation of trust on eBay. Using eBay feedback to attempt to extort goods or services from another member undermines the integrity of the feedback system” (“EBay help: Feedback extortion”, n.d.). The system could defend itself by weighting in different ways the feedback of different users. For example, if Alice has been directly threatened by CoolJohn12 and thinks the feedback provided by him is not reliable, his feedback about other users should not be taken into account when
computing the trust Alice could place in the other users. In fact, a possible way to overcome this problem is to use Local Trust Metric (Massa and Avesani, 2005, Ziegler and Lausen, 2004), that considers only (or mainly) trust statements given by users trusted by the active user and not all the trust statements with the same, undifferentiated weight. In this way, receiving negative feedback from CoolJohn12 does not influence reputations as seen by the active user if the active user does not trust explicitly CoolJohn12. For a short discussion of Global and Local Trust Metrics, see Section 3. However, eBay at the moment uses the Global Trust Metric we described before, which is very simple. This simplicity is surely an advantage because it is easy for users to understand it and the big success of eBay is also due to the fact users easily understand how the system works and hence trust it (note that the meaning of “to trust” here means “to consider reliable and predictable an artifact” and not, as elsewhere on this chapter, “to put some degree of trust in another user”). Nevertheless, in November 2004, a survey on eBay’s feedback system (Steiner, 2004) found that only 3% of the respondents found it excellent, 19% felt the system was very good, 39% thought it was adequate and 39% thought eBay’s feedback system was fair or poor. These results are even more interesting when compared with numbers from a January 2003 identical survey. The portion of “excellent” went from 7% to 3%, the “very good” from 29% to 19%, the “adequate” from 35% to 39%, the “fair or poor” from 29% to 39%. Moreover, the portion of total respondents who stated that they had received retaliatory feedback within the prior 6 months passed from 27% of 2003 survey to 38% of 2004 survey. These shifts seem to suggest that the time might have come for more sophisticated (and, as a consequence, more complicated to understand) Trust Metrics.

Bibliography for this portion:
– Resnick, P., & Zeckhauser, R. (2002). Trust Among Strangers in Internet Transactions: Empirical Analysis of eBay’s Reputation System. The Economics of the Internet and Ecommerce. Advances in Applied Microeconomics, 11.
– Resnick, P., & Zeckhauser, R., & Swanson, J., & Lockwood, K. (2003). The value of reputation on eBay: A controlled experiment.
– eBay help: Feedback extortion. (n.d.) Retrieved December 28, 2005, from
– Steiner, D. (2004). Auctionbytes survey results: Your feedback on eBay’s feedback system. Retrieved December 28, 2005, from
– Massa, P., & Avesani, P. (2005). Controversial users demand local Trust Metrics: an experimental study on community. In Proceedings of 25th AAAI Conference.
– Ziegler, C., & Lausen, G. (2004). Spreading activation models for trust propagation. In IEEE International Conference on e-Technology, e-Commerce, and e-Service (EEE’04).

Leave a Reply

Your email address will not be published. Required fields are marked *