Research On LDP-based Quality-assured Data Aggregation Mechanism In Mobile Crowdsensing

Mobile crowdsensing is a data collection and information sharing model from group intelligence.In this model,the sensing platform publishes sensing tasks with content characteristics,time-space characteristics,and social characteristics.Task participants receive tasks and then collect sensing data based on their portable electronic devices.The platform aggregates and analyzes these data to provide a variety of services to society.On the one hand,due to the data's differences,the sensing platform needs to accurately aggregate sensing data from different sources to identify the correct and available information.The value of data usually involves sensitive information,and different participants have different privacy requirements for the context of similar perception tasks(including contextual information such as time,space,and identity).If the sensing platform cannot meet participants' privacy needs,they will provide untrusted data,which would reduce the quality of data aggregation.On the other hand,due to differences in participants' ability to collect data,and devices,such as sensors,may also be damaged to a certain extent,which affects the accuracy of the data.Therefore,it is necessary to select a group of high-quality participants to perform the task.In real life,the platform cannot obtain the context of participants.To solve the above problems,this thesis researches two aspects: high-quality data aggregation and the selection of high-quality participants under the condition of privacy protection,to achieve quality-guaranteed data aggregation.In response to high-quality data aggregation,this thesis proposes a multi-level privacy protection truth discovery aggregation mechanism(MLTD).In this mechanism,the sensing platform first divides the degree of privacy protection into multiple levels and provides corresponding feedback for participants to compensate for their privacy costs.Secondly,the sensing platform informs participants of the privacy budget and feedback at each level.Participants locally obfuscate the data based on the d-bit histogram estimation mechanism and send the obfuscated data and privacy protection level to the sensing platform.Finally,the sensing platform aggregates the obfuscated data according to requirements based on the truth discovery algorithm.This thesis proves through rigorous theoretical analysis that the MLTD mechanism satisfies the local differential privacy(LDP)and(?,?)-accuracy.The MLTD mechanism is compared with the typical personalized privacy protection data aggregation mechanism in different scenarios.The experimental results verify the accuracy of the MLTD mechanism to aggregate data.In response to the selection of high-quality participants,this thesis proposes a participant selection mechanism for privacy protection.Firstly,the problem of selecting unknown participants is modeled as a multiple-play game of multi-armed bandit problems,and the ability attribute of the participants is modeled as an ability measurement parameter.Secondly,the ability measurement parameter is trained to realize the participant group's iterative selection based on the Thompson Sampling algorithm.In the iterative process,the participant ability parameter is updated.During the update process,the participant uses a random response mechanism to perturb its feedback value so that the parameter update process meets the LDP constraints.This thesis conducts a detailed theoretical analysis of the mechanism's performance,and the analysis results show that the optimal participant group can be selected within a limited number of iterations.The experimental results show that compared with existing typical mechanisms with real data sets,the proposed mechanism is highly efficient.It is an effective participant selection mechanism adapted to mobile crowdsensing without context.Combining the truth discovery aggregation mechanism with the participant selection mechanism can achieve quality assurance data aggregation,protect the sensing participants' sensitive attributes,ensure the availability of aggregated data,and finally provide a data aggregation mechanism based on local differential privacy for mobile crowdsensing system.