Data Quality Control Mechanism Design For Mobile Crowdsourcing

As a cost-effective working model,Mobile crowdsourcing can solve time and location-sensitive crowdsourced tasks by using human intelligence or human sensor.However,it is difficult to guarantee the data quality of mobile crowdsourcing due to the heterogeneous capabilities and strategic behavior of crowdworkers.Therefore,data quality control is necessary for mobile crowdsourcing.Especially,through affecting the behavior of crowdworkers,the task allocation mechanism and incentive mechanism will have significant impact on the data quality of crowdsourcing systems,and are huge challengesThe main contributions of this thesis are following threefold(1)First,this thesis thoroughly investigates the domestic and foreign research on quality control mechanisms for mobile crowdsourcing,and analyzes the factors affecting data quality in mobile crowdsourcing from the perspective of crowdworkers,crowdsourcers and platforms.Then,the existing schemes are summarized,and divided into four strategies improving task design,worker selection,incentive mechanism and data processing,and explaining and analyzing representative quality control techniques.Finally,the shortcomings of existing works and future research directions are pointed out(2)Spatial crowdsourcing based on time and space constraints has become a new mobile crowdsourcing model.How to efficiently match the workers and tasks that appear dynamically on the platform in real time and maximize the matching utility is a challenge In order to solve this problem,this thesis proposes the TGOA-ExtendedAT algorithm.The main idea is to first divide crowdworkers and tasks into two parts of equal length according to their arrival orders.For the first half,a greedy strategy is adopted,in which each new arrival crowdworker(or task)is assigned to a task(or crowdworker)with the largest utility and satisfying the limitations.For the second half,TGOA-ExtendedAT adopts an adaptive threshold strategy which the utility of matching object no less than a set threshold which constantly change with the current total utility besides meeting the constraints.The minimum total utility that the algorithm can obtain is proved theoretically and the time complexity of this algorithm for processing each new arrival object is 0(max(|T|,|W|))Finally,simulations based on synthetic and real datasets prove the efficiency and effectiveness of the algorithm(3)Incentive mechanism plays an important role on ensuring the quantity of pariticipants and the quality of data.Although the existing incentive mechanism takes into account the strategic behavior and self-interest of crowdworkers,it ignores the impact of the distance between crowdworker and task on the utility of platform or crowdworker.This thesis proposes a Reputation Based Location-aware Incentive Mechanism.This mechanism distributes rewards according to the actual contributions of workers and the distance between workers and tasks,not only motivating workers to invest in their efforts,but also motivating long-distance workers to participate.Equilibrium analysis shows that input effort is the best strategy for workers;And the mechanism uses the truth discovery algorithm to estimate the truth of the task,which not only provides a benchmark for contribution evaluation,but also can effectively counteract the deception behavior's disturbance to the discovered truth.Finally,the efficiency and effectiveness of the mechanism are proved by experimental simulations on real datasets.