Research On Benefit Allocation Method For Spatial Crowdsourcing Platforms Cooperation

With the popularity of smartphones and the development of the mobile Internet,spatial crowdsourcing has become a new direction for the development of traditional crowdsourcing in the era of sharing economy.The core operation of spatial crowdsourcing is task assignment,i.e.,assigning tasks with spatial and temporal characteristics to free workers.The existence of multiple spatial crowdsourcing platforms makes cooperation possible.Each platform cooperates by sharing tasks and workers to help other platforms complete tasks that would otherwise be difficult to complete,thus achieving better task assignments and an expansion of total revenue.In addition,spatial crowdsourcing platforms can also use big data to drive smarter task assignments,and share their data resources based on federal learning technology to collaboratively train better performing models,which can aid and support platform decision making and bring more benefits.In both models,platform cooperation can bring more benefits,and it is crucial to incentivize each platform to actively participate in the cooperation and distribute these benefits.Fair benefit allocation,as one of the effective incentive mechanisms,can promote the willingness of each platform to cooperate.However,existing incentive mechanisms in spatial crowdsourcing are mainly studied for workers,and their design goals are usually to maximize platform benefits or minimize social costs,which do not apply to the benefit allocation problem in multi-platform cooperation scenarios;existing incentive mechanism methods in federal learning cannot incrementally measure and update each platform’s contribution in scenarios where platforms dynamically join the cooperation.Therefore,this thesis focuses on the efficient benefit allocation methods in these two cooperation models.The specific contributions are mainly in the following two areas:(1)Research on benefit allocation methods in cross-platform task distribution.In this thesis,we propose a fair benefit allocation method based on Shapley value to motivate the platforms to participate in cooperation.In addition,considering that the exponential time complexity of the method limits its practicality,an efficient approximation is proposed based on the characteristics of task assignment,using the spatio-temporal information of tasks and workers and related auxiliary information.Finally,the efficiency and effectiveness of the proposed methods are verified by adequate experiments on a real data set.(2)Research on benefit allocation methods in cross-platform federated learning.Shapley value is an important contribution measure and benefit allocation method.In this thesis,we propose a Shapley value contribution measure and benefit allocation method applicable to dynamic scenarios and design an efficient approximation method using intermediate information such as gradients and parameters saved during the training of the federal learning model.The efficiency and effectiveness of the proposed methods are verified on two real datasets with different experimental settings.