TOWARDS SECURITY AWARE CROWDSOURCING

Abstract

Crowdsourcing has emerged as a novel problem-solving paradigm, which facilitates addressing problems by outsourcing them to the crowd. The openness of crowdsourcing renders it vulnerable to misbehaving workers that impair data trustworthiness. They may attempt to submit calibrated data/parameters to manipulate crowdsourcing outcomes for higher beneficial gain. Those misbehaviors would infringe crowdsourcing's process and, overall, its usefulness.

In this dissertation, I intend to secure the crowdsourcing platform from worker's untrustworthy data reporting. The main contributions are mainly threefold. First, we secure task allocation, an essential but vulnerable stage in crowdsourcing, from individual misreporting. To be specific, misbehaving workers may manipulate task allocation outcomes by uploading falsified parameters. Under the framework of incentive mechanism design, we propose a defense scheme that obtains accurate task allocation outcomes even with workers' manipulated parameters. Second, we further consider workers' collusive behaviors in the stage of task allocation. Strategic workers may form coalitions and rig their parameters together to game the system for extra benefit. To suppress collusion, we leverage incentive mechanism design to calibrate proper payment, leaving workers limited motivation to collude. Third, in addition to task allocation, we also investigate the misbehaviors from strategic workers in the stage of answer collection. A unified framework is developed to protect these two stages from workers' strategic manipulation simultaneously. Our approach still falls into the category of incentive design. Payment rule is carefully designed, such that workers gain more for truth-telling. It thus motivates workers to honestly report genuine data and parameters in both stages.