An Incentive Compatible Multi-Armed-Bandit Crowdsourcing Mechanism with Quality Assurance

Consider a requester who wishes to crowdsource a series of identical binary labeling tasks from a pool of workers so as to achieve an assured accuracy for each task, in a cost optimal way. The workers are heterogeneous with unknown but fixed qualities and moreover their costs are private. The problem is to select an optimal subset of the workers to work on each task so that the outcome obtained from aggregating labels from them guarantees a target accuracy. This problem is challenging because the requester not only has to learn the qualities of the workers but also elicit their true costs. We develop a novel multi-armed bandit (MAB) mechanism for solving this problem. We propose a framework, {\em Assured Accuracy Bandit (AAB)}, which leads to an adaptive, exploration separated MAB algorithm, {\em Strategic Constrained Confidence Bound (CCB-S)}. We derive an upper bound on the number of exploration steps which depends on the target accuracy and true qualities. We show that our CCB-S algorithm produces an ex-post monotone allocation rule which can be transformed into an ex-post incentive compatible and ex-post individually rational mechanism that learns qualities of the workers and guarantees the target accuracy in a cost optimal way.