Collusion-Proof Result Inference in Crowdsourcing

摘要: In traditional crowdsourcing, workers are expected to provide independent answers tasks so as ensure the diversity of answers. However, recent studies show that crowd is not a collection workers, but instead communicate and collaborate with each other. To pursue more rewards little effort, some may collude repeated answers, which will damage quality aggregated results. Nonetheless, there few efforts considering negative impact collusion on result inference in crowdsourcing. this paper, we specially concerned Collusion-Proof problem for general crowdsourcing public platforms. end, design metric, worker performance change rate, identify colluded by computing difference mean before after removing Then incorporate detection into existing methods guarantee results even occurrence behaviors. With real-world synthetic datasets, conducted an extensive set evaluations our approach. The experimental demonstrate superiority approach comparison state-of-the-art methods.