Research On Privacy-Preserving Truth Discovery Methods Based On Distribution Estimation

With the rapid popularity of intelligent devices,crowdsourcing has become an effective method of data capture and has been widely used in research fields such as information retrieval and machine learning.However,the collection and analysis of data by crowdsourcing platform bring great risk of privacy leakage to users.Furthermore,out of concerns about the leakage of personal sensitive information,users in crowdsourcing may provide low-quality data.The difficulty of the tasks and expertise of users may also lead users to submit inconsistent data.Therefore,effective protection of personal sensitive data and accurate inference of real data are of great significance to crowdsourced data management.In recent years,the research of privacy-preserving truth discovery scheme has provided for many scenarios with above problems.Nevertheless,the research on the truth discovery with local differential privacy for crowdsourcing-oriented is not yet complete.In response to above questions,the thesis mainly conducts related research from two aspects of non-personalized and personalized truth discovery scheme with local differential privacy protection in crowdsourcing,and proposes following schemes for multiple discrete data type:(1)A truth discovery scheme with local differential privacy for multi-variate crowdsourced data is proposed.In the scheme,a local differential privacy mechanism based on random response is adopted to protect personal sensitive data,and two distribution estimation methods are used to reduce the adverse effects of injected noise on quality of user and data.In addition,this thesis theoretically analyzes the impact of privacy budget,data volume and other parameters on the accuracy of truth discovery,as well as the impact on user credibility under different privacy intensities.The experimental results on the real and synthetic datasets show that this scheme is more suitable for multi-variate crowdsourcing dataset,and the accuracy of truth discovery is better than comparison algorithms.(2)A truth discovery scheme with personalized multi-layer privacy protection for crowdsourcing-oriented is proposed.Considering the different privacy requirements of each user in the actual scenario,the scheme sets privacy layers with different privacy protection intensity.Then,users choose the corresponding privacy layer according to personal privacy requirement to achieve user-based personalized privacy protection.Moreover,the scheme designs methods of data recovery and hierarchical aggregation to solve the problem of data dispersion which caused by personalized privacy protection.Distribution estimation method is also used to further improve the utility of truth discovery.The final experimental results show that this scheme can effectively improve the accuracy of truth discovery under different privacy requirements of users.