Research On Data Credibility Enhancement Method In Crowd Sensing Environments

With the development of modern science and technology and the popularity of smart phones,mobile crowd sensing,as an emerging data collection method,is gaining popularity in more and more application scenarios.Its main data collection method is to collect the surrounding information by using sensors in various smart devices.In mobile crowd sensing system,sensing nodes are generally characterized by high flexibility and densely distributed among nodes.Therefore,compared with traditional fixed sensor network,mobile crowd sensing system has wider coverage,lower task cost,and more timeliness and comprehensiveness of collected data.However,the high flexibility of the mobile crowd sensing system also means that it cannot determine the credibility of each sensing node,thereby affecting the quality of the upload data,resulting in the need to eliminate low-quality data when the sensing platform analyzes the data to ensure the accuracy of the results,and increasing the additional cost of the system operation.In order to ensure the quality of the collected sensing data,it is an important content of mobile crowd sensing system to select suitable sensing nodes to complete the data collection task according to the temporal and spatial requirements and task characteristics of the sensing task.Based on this,this paper conducts relevant research on data processing and sensing user selection in mobile crowd sensing system,and the main research results are as follows:1.A multi-label local embedding dimension reduction algorithm based on knearest neighbor selection is proposed.Firstly,according to the relationship between different labels in the perceptual data,metric learning method is used to determine the k-nearest neighbor selection range of the sensing data.Secondly,after the evaluation of each data label,the corresponding weights are set to make each data label have different utility during dimensionality reduction.Finally,the local dimensionality reduction algorithm is used to reduce the dimensionality of perception data,reducing the negative impact of redundant and disorderly data labels on data analysis and reducing the amount of calculation of data analysis by mobile crowd sensing platform without affecting the analysis results.Experimental results based on My Signals data sets show that the proposed method can effectively improve the dimension reduction effect of high-dimensional data and reduce the time required in the dimension reduction process compared with other algorithms.2.A trusted user selection method based on generative adversarial network and user reputation is proposed.First of all,the neural network model is built.After a large amount of data training,the analysis module and the simulation module form an adversarial game relationship.At this time,the training of the adversarial network model is completed,and the judgment accuracy of the input data is very high.Secondly,the completion degree of the user's historical task is counted,and the realtime credit value of the user is calculated by combining with the time decay factor,which is used as the basis for generating adversarial network to analyze the user's credibility.Finally,the data is input into the analysis module,which can accurately analyze whether the user is trusted or not in a short time.Experimental results based on the simulation data set show that this method can effectively improve the accuracy of trusted user selection,and can obtain higher quality perception data compared with other user selection methods.3.An optimal selection method for participants oriented to user preference is proposed.Firstly,the sensing data provided by users and the subjective evaluation of the sensing environment are analyzed,and the user reputation value is calculated according to the credibility matrix,so as to ensure the credibility of the data uploaded by users.Secondly,according to the content of the sensing task and the user's interest preference,the interest correlation is calculated.According to the user ' s reputation value and interest correlation,the comprehensive evaluation index is calculated.The users with higher evaluation index are added to the user list of the sensing task.Through this method,users can collect data in their areas of interest,and play an incentive role for users.Finally,the sensing platform sets the start-up cost of this task according to the remuneration required by the sensing users,so as to attract more users to participate.Based on the analysis of experimental results,this method improves the accuracy of user credibility sensing and ensures that the sensing platform can obtain high-quality sensing data.