Sorting from Crowdsourced Comparisons using Expert Verifications

Abstract: We introduce a novel noisy sorting model motivated by the Just Noticeable Difference (JND) model from experimental psychology. The goal of our model is to capture the low quality of the data that are collected from crowdsourcing environments. Compared to other celebrated models of noisy sorting, our model does not rely on precise data-generation assumptions and captures crowdsourced tasks' varying levels of difficulty that can lead to different amounts of noise in the data. To handle this challenging task, we assume we can verify some of the collected data using expert advice. This verification procedure is costly; hence, we aim to minimize the number of verifications we use. We propose a new efficient algorithm called CandidateSort, which we prove uses the optimal number of verifications in the noisy sorting models we consider. We characterize this optimal number of verifications by showing that it is linear in a parameter $k$, which intuitively measures the maximum number of comparisons that are wrong but not inconsistent in the crowdsourcing data.