Adaptive task assignment in spatial crowdsourcing
Hassan, Umair ul
MetadataShow full item record
This item's downloads: 279 (view details)
Spatial crowdsourcing has emerged as a new paradigm for solving difficult problems in the physical world. It engages a large number of human workers in scenarios such as crisis management and smart cities. The utility of spatial crowdsourcing is generally dependent on who performs the task. On one hand, the workers may not perform assigned tasks within time. On the other hand, the performed tasks may not meet the prede ned quality criteria. In the case of spatial crowdsourcing, the success of an assignment depends on factors such as task location, travel distance, or expected reward. This necessitates an appropriate task assignment process to optimize the utility of spatial crowdsourcing. The design of the task assignment process faces three primary research challenges: dynamism, heterogeneity, and uncertainty. The goal of this thesis is to address these research challenges; therefore, it proposes a conceptual framework to analyze the dimensions of dynamic task assignment in spatial crowdsourcing. To formalize the research problem, this thesis de nes dynamic task assignment as a repeated decision-making problem under uncertainty and heterogeneity. Uncertainty de nes the limited knowledge about the reliability of workers for assigned tasks, and heterogeneity characterizes the di erences in reliability and expertise of workers. The proposed formalization is referred to as the adaptive assignment problem in spatial crowdsourcing. The adaptive assignment problem combines online optimization with heuristic-based learning for balancing the exploration-exploitation trade-o during repeated assignments. The problem is instantiated in four di erent scenarios of spatial crowdsourcing to highlight speci c requirements of assignment algorithms. An agent-based simulation methodology is employed to evaluate the algorithms proposed for each scenario. Empirical evaluation provides evidence of the performance of algorithms using synthetic and real-world data.
The following license files are associated with this item: