Aggregating Reputation Feedback

Abstract

A fundamental task in reputation systems is to aggregate multiple feedback ratings into a single value that can be used to compare the reputation of different entities. Feedback is most commonly aggregated using the arithmetic mean. However, the mean is quite susceptible to outliers and biases, and thus may not be the most informative aggregate of the reports. We consider three criteria to assess the quality of an aggregator: the informativness, the robustness and the strategyproofness, and analyze how different aggregators, in particular the mean, median and mode, perform with respect to these criteria. The results show that the arithmetic mean may not always be the best choice.