Conclusions and Managerial Implications

I began this chapter by arguing that managers of online marketplaces should pay special attention to the design of effective trust management mechanisms that will help guarantee the stability, longevity, and growth of their respective communities. This chapter has contributed in this direction by presenting a number of novel techniques for "immunizing" online reputation systems against unfair ratings. The proposed techniques are summarized in Figure 6. The combination of frequency filtering and median filtering is capable of guaranteeing reputation biases of less than 5% (e.g., less than 0.5 points when ratings range from one to 10) when the ratio of unfair raters is up to 15% to 20% of the total buyer population for a given seller.

Technique	Description	Effect	Prerequisites
Controlled anonymity	Market-maker conceals the true identities of buyers and sellers from one another and only reveals their respective reputation estimates	Prevents bad-mouthing and/or negative discrimination	Ability to practically implement with reasonable cost
Median filtering	Calculation of reputation estimate using the median of the ratings set	Results in robust estimations in the presence of up to 30% to 40% of unfair ratings	Ratio of unfair ratings less than 50%
Frequency filtering	Ignores raters whose ratings submission frequency for a given seller is significantly above average	Eliminates raters who attempt to flood the system with unfair ratings; maintains the final ratio of unfair raters at low levels	Ability to authenticate the true identity of online raters

Figure 6: Summary of Proposed Immunization Techniques

The conclusions of this chapter are directly applicable to the design of current and future electronic marketplaces. More specifically, the analysis of the proposed techniques has resulted in a number of important guidelines that managers of online marketplaces should take into account in order to embed effective reputation systems into their respective communities:

• It is important to be able to authenticate the identity of rating providers. Unauthenticated communities are vulnerable to unfair rating "flooding" attacks.

• Concealing the (authenticated) identity of buyers and sellers from one another can prevent negative unfair ratings and discriminatory behavior. Managers of electronic marketplaces and B2B hubs can consider adding this function into the set of services they provide to their members.

• Numerical reputation estimates should be based on the median (and not the mean) of the relevant rating set. Also, frequency filtering should be applied in order to eliminate raters who might be attempting to flood ("spam") the system with potentially unfair ratings.

This chapter suggests several topics for further research. The calculation of robust estimates of reputation variance, the development of "immunization" techniques that avoid unfair ratings "flooding" in non-authenticated online communities, and the analysis of unfair ratings in environments where bi-directional ratings are possible (that is, both parties can rate one another) are just some of the issues left open by this work. It is our hope that the analysis and techniques proposed by this work will provide a useful basis that will stimulate further research in the important and promising field of online reputation systems.