Research On Privacy Preservation Of Spatial Crowdsourcing Based On Local Differential Privacy

In spatial crowdsourcing task allocation,workers are required to send their real information to the crowdsourcing platform in order to achieve goals such as maximising the total number of tasks allocated or maximising worker compensation.However,there is a risk of privacy information being leaked or misused by untrustworthy platforms,so researchers have proposed a number of privacy protection mechanisms,and local differential privacy is one of the widely used approaches.However,in some spatial crowdsourcing tasks with special requirements,direct application of existing local differential privacy mechanisms may result in reduced utility of worker data and significant degradation of task completion quality.To address the privacy protection concerns in these new scenario,this thesis carries out relevant optimisation work based on local differential privacy to significantly improve the quality of task completion.The main work in this thesis is as follows.(1)We propose an innovative differential privacy mechanism for variety crowdsourcing tasks.The method can effectively select a subset of workers with diverse kinds of attributes while providing privacy protection for worker attributes.In addition,we propose a 2-step local differential privacy method to optimize the results under the unbalanced distribution of worker attributes.Experiments on synthetic and real datasets demonstrate the effectiveness of the method.(2)We design privacy-preserving methods for maximum worker coverage tasks in spatial crowdsourcing.The method uses a geo-indistinguishability perturbation mechanism to protect worker location privacy,and designs a true distribution estimation method based on this perturbation method.The optimal coverage location is then determined and the nearest worker is recruited by a budgeted maximum coverage greedy algorithm.Experiments on synthetic and real datasets demonstrate the effectiveness of the method.(3)We design a perturbation method for multi-platform cooperative worker location release.The method selects feature locations based on the real distribution of workers in different platforms and aggregates the feature locations and constructs a quadtree,designs a random walk location perturbation method based on the quadtree and proves that the method satisfies geo-indistinguishability.Experiments on synthetic and real datasets prove the effectiveness of the method.