Data Quality Analysis And Quality-Aware User Management For Mobile Crowdsensing

With the growing proliferation of mobile devices and the rapid development of wireless technology,mobile crowdsensing has become a new sensing paradigm.It utilizes the sensors of people's mobile devices to sense,collect,analyze,and share the local information of their surrounding environment,so as to achieve the purpose of collecting specific data.Compared with traditional sensor networks,mobile crowdsensing,based on the devices and intelligence of humans,has better feasibility and autonomy and can cover regions that are hard or impossible for sensor networks to cover.Besides,people's smart devices usually have more powerful computing,storing,and communication functions that are able to benefit the fulfillment of sensing tasks.The concept of mobile crowdsensing has already been widely applied in various real scenarios,including but not limited to indoor localization,indoor floorplan construction,traffic condition monitoring,environment condition monitoring,bus arrival prediction,last mile navigation,and so on.It is one of the keychains in achieving smart cities.Data are the foundation of a mobile crowdsensing system,and data quality is the key problem in recent mobile crowdsensing researches.On one hand,due to the factors of devices and environment,the data collected by mobile crowdsensing platforms usually contain different levels of noise,thus the data quality varies.On the other hand,in some circumstances,we may need mobile device users to follow certain specific behaviors in collecting sensing data.However,due to the uncertainty of the users' behaviors,some users may consciously or unconsciously deviate from the standard sensing behaviors,causing large errors in the collected data.The contribution of low-quality data reduces the accuracy and effectiveness of the platform's follow-up data analysis procedure,wasting both the users' and the platform's resources and diminishing the crowdsensing platform's quality of service.Therefore,effectively addressing the data quality problem is the key to improve the effectiveness of the mobile crowdsensing platform.In recent years,the data quality problem of mobile crowdsensing has gained increasing attention in research communities.In this paper,we have conducted extensive research works focusing on data quality analysis and quality-aware user management problem.Our works include the following three parts.First,data quality analysis.Analyzing and quantifying users' data quality are the basis to address the data quality problem in mobile crowdsensing.Given users' contributed data,how to,without the knowledge of ground truth,accurately and effectively quantifying the quality of each data item,modeling the users' behaviors,and estimating the values of the monitored objects are the core problems of data quality analysis.To this end,we design a novel framework to estimate the users' data quality and to determine the users' payments.In the framework,we propose an unsupervised method to quantify the users' short-term data quality,capture abnormal data contribution based on outlier detection techniques,and design a reputation system to characterize the users' longterm data contribution behaviors.Further,we integrate the users' data quality into the determination of their payments,so as to motivate users' high-quality data contributions.Second,quality-aware user selection and user coordination.The problems of user selection and user coordination are two important problems in mobile crowdsensing.Specifically,the crowdsensing platform needs to select part of the users from a large number of registered users to perform sensing tasks,and calculates specific coordinations towards the users' sensing locations and sensing paths,in order to maximize the utility of the crowdsensing platform.Earlier researches usually adopt techniques from game theory and combinatorial optimization theory to address these problems,while ignoring the important factor of quality.In this paper,we incorporate the quality factor into the problems of user selection and user coordination.Our works include the follows:(1)We study the problem of designing sensing paths for users based on the quality of data fusion.We consider different mobility scenarios and design paths for each scenario respectively so as to maximize the sensing coverage while satisfying the constraint of sensing quality.(2)We investigate the user selection problem under the constraints of both time and budget,without the prior knowledge of users' utilities.We propose a combinatorial multi-armed bandit algorithm to learn the users' utilities and select users in an online manner.Third,quality-aware task matching and task recommendation.In mobile crowdsensing,a platform-centric method is usually adopted to address the task matching problem,i.e.,the platform determines which users to conduct which tasks.A major drawback of the platform-centric methods is that users cannot choose their interested tasks,preventing users from expressing their preferences.To address this issue,we propose a novel user-centric task matching method.Based on a fine-grained user profiling,we learn the users' preference and reliability towards different tasks,and then recommend personalized tasks to each user based on the joint consideration of her preference and reliability scores.Specifically,we propose a hybrid preference profiling method based on users' implicit feedback information.In profiling users' reliability,we propose a semisupervised method to character users' reliability levels of their involved tasks and a matrix factorization algorithm to estimate their reliability of uninvolved tasks.The experiment results show that our personalized task recommendation system can effectively model the users and improve the estimation quality and task acceptance ratio.