ε-collusion-proof mechanism for the aggregation of unverifiable cluster labels

摘要

In this paper we study the collusion-proof property of a mechanism that is proposed for solving classification/clustering aggregation among selfish agents who tend to lie about their correct clustering/classes during the aggregation process. The ground-truth of the class labels is unknown, yet the agents are interested in the majority choice. The mechanism is first studied in [1], which shows the feasibility of it to foster agents' truth-telling behavior, but failed to address its collusion-proof property. In this paper we show that the proposed mechanism is indeed satisfying the collusion-proof property under some conditions. A further analysis based on the property of the participating agents is also provided here.