Content-Driven Reputation for Collaborative Systems

Abstract

We consider collaborative editing systems in which users contribute to a set of documents, so that each document evolves as a sequence of versions. We describe a general technique for endowing such collaborative systems with a notion of content-driven reputation, in which users gain or lose reputation according to the quality of their contributions, rather than according to explicit feedback they give on one another. We show that content-driven reputation systems can be obtained by embedding the document versions in a metric space with a pseudometric that is both effort preserving (simple changes lead to close versions) and outcome preserving (versions that users perceive as similar are close). The quality of each user contribution can be measured on the basis of the pseudometric distances between appropriately chosen versions. This leads to content-driven reputation systems where users who provide contributions of positive quality gain reputation, while those who provide contributions of negative quality lose reputation. In the presence of notification schemes that prevent the formation of “dark corners” where closed groups of users can collaborate without outside interference, these content-driven reputation systems can be made resistent to a wide range of attacks, including attacks based on fake identities or specially-crafted edit schemes.