Area coverage-based worker recruitment under geo-indistinguishability☆
Introduction
Mobile crowdsensing (MCS) is a useful paradigm of crowdsensing, which facilitates urban sensing to extract information in targeted areas by physically moving to the locations [1], [2], [3], [4]. Many important applications in MCS, such as those used for traffic monitoring [5], earthquake detection [6] and noise monitoring [7] will benefit greatly if the coverage area will be maximized [8], [9]. The core issue is how to recruit a set of workers with sensing radii to cover an interested region as much as possible. In particular, the sensing radii reflect how well these workers monitor their surrounding region [10], [11], [12]. However, participants are naturally required to disclose their sensitive locations to the MCS server. Without proper protection, it inevitably results in insufficient recruitment of workers to cover the interested region as participants may be reluctant to participate in the MCS system. Therefore, developing an approach to address such privacy concern is an urgent need.
Geo-indistinguishability (Geo-I) [13] has recently evolved into the de facto criterion for location protection. It is independent of the adversaries’ prior knowledge and is robust with respect to composition [14]. Moreover, it can be achieved in a simple and efficient way using planar laplacian mechanism (PL) [13]. Recently, PL has been adopted in many applications, including LP-Guardian [15], LP-Doctor [16] and the system for secure nearby-friends discovery [17].
Existing studies either are based on encryption technologies [12], [18], [19], [20] or are only designed for Target Coverage-based worker recruitment [21], [22]. The scenarios of Target Coverage and Area Coverage are shown in Fig. 1, Fig. 1 respectively. In particular, the goal of Target Coverage-based worker recruitment is to cover a set of tasks’ locations as more as possible. In practice, most of the time we need to maximize the coverage ratio of an interested region rather than covering several locations. For example, if a MCS system aims to provide traffic updates (congestion detection [23] or lane change monitoring [24]), obviously, it is more meaningful to maximize the area coverage. Additionally, participants can assist in finding criminals [25], taking photos for air quality analysis [22], [26] and collecting data for mobile advertisement dissemination [27]. Besides, there exists an excellent work [28] that most related to ours. However, since they do not consider the impact of participants’ sensing radii, and divide the target region into grids in advance, they potentially assume that if a participant’s obfuscated location is in a grid, the grid would be completely covered, and he could fully perceive the area formed by this certain grid in the future. This may not be the case in practice due to the impact of participants’ skills, located circumstances or their perceived equipments and so on [29].
Hence, in this paper, we, for the first time, investigate the problem of Area Coverage-based worker recruitment while considering participants’ sensing radii under Geo-I. Specifically, given the interested region and a crowd of active participants, there exist overlaps between each participant’s sensing region and . Each sensing region can be jointly defined by his location coordinate and self-determined sensing radius. The server aims to identify a suitable set of participants under a constraint on the number of the recruited workers . It is devoted to achieving the maximal coverage ratio for by designing an effective recruitment approach while preserving participants’ locations under Geo-I, where coverage ratio represents the ratio of the area of the sensing regions from all the recruited workers to the area of . The bigger, the better. A straightforward approach may be designed based on PL to obfuscate each participant’s location, and regard the top- participants as the recruited workers according to their degree of overlaps with in descending order. However, due to the randomness and boundlessness when obfuscating locations, and the arbitrariness of selecting participants, this straightforward approach will suffer from poor performance. Specifically, due to the randomness and boundlessness of PL [30], it may inject too much noise and inevitably leads to unreliable results for Area Coverage-based worker recruitment scenario. In particular, a participant located in may be obfuscated to a location outside of , which indicates that a participant whose sensing region has a large overlap with may be considered to have a small overlap with , and vice versa. In addition, arbitrarily selecting participants without considering their mutual spatial relationships will lead to large overlaps of the recruited participants’ sensing regions. Therefore, the area coverage ratio will be decreased under the number constraint .
To fill this gap, we present a geo-indiStinguishable arEa Coverage-based workeR rEcruitmenT approach, which is referred to as SECRET. Its main idea is to generate the obfuscated locations according to participants’ sensing regions and consider spatial relationships among participants to ensure large coverage for .
In SECRET, to generate an obfuscated location for each participant with high utility, we develop an optimized geographical exponential mechanism OptGEM with solid privacy and utility guarantees. It can lead to a high chance of the obfuscated location falling into the nearby locations with the real location by probability sampling a location within his sensing region. Overall, in OptGEM, the server first partitions the target region and broadcasts the partitioned grids to each participant. Then, each participant decides whether to further partition the received grids according to the number of grids he covers. Specifically, the server first determines the optimal two-level grid granularities and by bounding the information loss to minimize the total noise. Then, it partitions using and broadcasts the partitioned grids and to each participant. Finally, each participant generates his obfuscated location by probability sampling a location from the center of each covered grid. In addition, to avoid being unable to distinguish the participants, for those who only cover one grid, each of them further partitions his covered grid using and generates his obfuscated location with the new partitioned grids in the same manner.
To select the recruited workers with the obfuscated locations, we design a coverage-aware worker selection worker method CWS, which aims to meet and ensures large coverage for . In CWS, the server iteratively computes the degree of overlaps between the participants’ sensing regions and the uncovered part of , and regards the participant with the largest degree as a new recruited worker until meeting . To calculate the degree of overlap between a participant’s sensing region and the uncovered part of , we only need to know the degree of overlaps between this participant’s sensing region and the already recruited workers’ sensing regions, after giving the degree of overlap of this participant’s sensing region with . Recall that each participant’s sensing region is a circle centered on himself. To calculate the degree of overlap between two circles, it is only necessary to know the distance between two corresponding centers on the premise of giving the radii of these two circles. Since directly calculating the distance between the two centers according to the obfuscated locations will make multiple noise be concentrated in the result, to reduce the contained noise of the calculated distance, we estimate the true distance through the spatial relationships between the obfuscated locations and the real locations.
The key contributions of this paper are summarized as follows:
- •
We formulate the problem of area coverage-maximized worker recruitment while considering each participant’s sensing radius under Geo-I, and present a novel approach SECRET. To our best knowledge, this is the first effort in the literature to define and tackle this problem.
- •
To protect participants’ locations, we propose an optimized geographical exponential mechanism OptGEM. In OptGEM, we first design a two-level grid granularity determination method and then develop a probability sampling method to generate the obfuscated locations. We further theoretically analyze the privacy and utility guarantees of OptGEM.
- •
To select the recruited workers with the obfuscated locations, we design a coverage-aware worker selection method CWS by identifying a set of participants to ensure large coverage for through distance estimation.
- •
We show that SECRET guarantees the discrete version -Geo-I. Extensive experiments on two real-world datasets and a synthetic dataset confirm the effectiveness of SECRET.
The rest of this paper is organized as follows. We discuss related work in Section 2. We describe the preliminaries in Section 3. The details of SECRET are presented in Section 4. Experiments are discussed in Section 5. Finally, Section 6 concludes our work.
Section snippets
Related work
The related work of this paper falls into the following categories.
Geo-indistinguishability
Geo-indistinguishability (Geo-I) can provide provable guarantee of location privacy. Let be a set of workers’ possible locations, and be a set of possible obfuscated locations. To guarantee Geo-I for a probabilistic obfuscation algorithm , we have the following definition.
Definition 1 Given a privacy budget , a randomized algorithm achieves -Geo-I, if and only if where and are any two locations in , is the probability of obfuscating to , and is an observed-Geo-I
Overview of SECRET
To privately obtain the recruited workers with high coverage ratio, we present SECRET. It includes an optimized geographical exponential mechanism OptGEM and a coverage-aware worker selection worker method CWS. The overview of SECRET is shown in Fig. 3. Specifically, SECRET consists of the following four phases:
Phase 1. Each participant first sends his sensing radius to the server. Then, the server computes and , and partitions the target region using into grids. Finally, it
Datasets
We take Tokyo (TKY for convenience) and New York (NYC for convenience) [74] for evaluation. TKY contains the locations of 325 subway stations and 503 offices. There are about 573 708 check-ins. Meanwhile, NYC contains the locations of 142 subway stations and 492 offices. There are about 227 428 check-ins. We randomly choose a 15 km*15 km region as , and select 300 offices’ locations as participants’ locations for both datasets. These participants’ sensing radii range from 0.05 km to 1.5 km with
Conclusion
In this paper, we present a novel approach SECRET for recruiting workers while guaranteeing the discrete version geo- indistinguishability. In SECRET, an optimized geographical exponential mechanism OptGEM is developed to desensitize the participants’ locations. Moreover, a coverage-aware worker selection worker method CWS is designed to recruit workers. We show that SECRET satisfies -geo-indistinguishability. The experimental results on two real-world datasets and a synthetic dataset validate
CRediT authorship contribution statement
Pengfei Zhang: Conceptualization, Methodology, Data curation, Writing – original draft. Xiang Cheng: Project administration, Writing – review & editing, Funding acquisition. Sen Su: Writing – review & editing, Supervision, Funding acquisition. Ning Wang: Software, Validation.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work was supported by the National Natural Science Foundation of China [Grant No. 61872045 and No. 62072052] and the Foundation for Innovative Research Groups of the National Natural Science Foundation of China [Grant No. 61921003]. Corresponding author of this paper are Prof. Xiang Cheng and Prof. Sen Su.
Pengfei Zhang is a Ph.D. Candidate from Beijing University of Posts and Telecommunications in China. His major is Computer Science. His current research interest focuses on privacy protection in mobile crowdsensing systems.
References (77)
- et al.
Privacy-preserving QoI-aware participant coordination for mobile crowdsourcing
Comput. Netw.
(2016) - et al.
Incentive mechanism in platform-centric mobile crowdsensing: A one-to-many bargaining approach
Comput. Netw.
(2018) - et al.
A general location-authentication based secure participant recruitment scheme for vehicular crowdsensing
Comput. Netw.
(2020) - et al.
Poster: Towards a multitask worker recruitment framework for mobile crowdsensing
- et al.
Nearby friend discovery with geo-indistinguishability to stalkers
- et al.
Prediction based user selection in time-sensitive mobile crowdsensing
- et al.
Participant recruitment for coverage-aware mobile crowdsensing with location differential privacy
- et al.
Prediction based user selection in time-sensitive mobile crowdsensing
- et al.
Privacy preserving participant recruitment for coverage maximization in location aware mobile crowdsensing
IEEE Trans. Mob. Comput.
(2021) - et al.
Trust-based privacy-aware participant selection in social participatory sensing
J. Inf. Secur. Appl.
(2015)
On location-privacy in opportunistic mobile networks, a survey
J. Netw. Comput. Appl.
A survey on location privacy protection in wireless sensor networks
J. Netw. Comput. Appl.
Quantitative combinatorial geometry for concave functions
J. Comb. Theory Ser. A
Spatial crowdsourcing: a survey
VLDB J.
Heterogeneous stream processing and crowdsourcing for traffic monitoring: Highlights
Haiti earthquake photo tagging: Lessons on crowdsourcing in-depth image classifications
NoiseSense: Crowdsourced context aware sensing for real time noise pollution monitoring of the city
Spatial crowdsourcing: Challenges and opportunities
IEEE Data Eng. Bull.
Urban computing: Concepts, methodologies, and appl.
ACM Trans. Intell. Syst. Technol.
A survey on security, privacy, and trust in mobile crowdsourcing
IEEE Internet Things J.
Privacy-preserving user recruitment protocol for mobile crowdsensing
IEEE/ACM Trans. Netw.
Geo-indistinguishability: differential privacy for location-based systems
The algorithmic foundations of differential privacy
Found. Trends Theor. Comput. Sci.
Location privacy protection for smartphone users
Anatomization and protection of mobile apps’ location privacy threats
Scalable privacy-preserving participant selection in mobile crowd sensing
Secret-sharing-based secure user recruitment protocol for mobile crowdsensing
Scalable privacy-preserving participant selection for mobile crowdsensing systems: Participant grouping and secure group bidding
IEEE Trans. Netw. Sci. Eng.
Geographic differential privacy for mobile crowd coverage maximization
Differentially private unknown worker recruitment for mobile crowdsensing using multi-armed bandits
IEEE Trans. Mob. Comput.
High quality participant recruitment in vehicle-based crowdsourcing using predictable mobility
Push vs pull participant recruitment system for personalized vehicular crowdsensing
GoSense: Efficient vehicle selection for user defined vehicular crowdsensing
Duration-variable participant recruitment for urban crowdsourcing with indeterministic trajectories
IEEE Trans. Veh. Technol.
Recruitment framework for participatory sensing data collections
Collecting and analyzing multidimensional data with local differential privacy
A survey of mobile crowdsensing techniques: A critical component for the internet of things
ACM Trans. Cyber Phys. Syst.
A survey on task and participant matching in mobile crowd sensing
J. Comput. Sci. Technol.
Cited by (1)
Adapting Geo-Indistinguishability for Privacy-Preserving Collection of Medical Microdata
2023, Electronics (Switzerland)
Pengfei Zhang is a Ph.D. Candidate from Beijing University of Posts and Telecommunications in China. His major is Computer Science. His current research interest focuses on privacy protection in mobile crowdsensing systems.
Xiang Cheng received the Ph.D. Degree in Computer Science from Beijing University of Posts and Telecommunications, China, in 2013. He is currently a Professor at the Beijing University of Posts and Telecommunications. His research interests include privacy-enhanced computing, data mining and knowledge engineering.
Sen Su received the Ph.D. degree in 1998 from the University of Electronic Science and Technology, China. He is currently a Professor at the Beijing University of Posts and Telecommunications. His research interests include data privacy, cloud computing and internet services.
Ning Wang is working toward the master’s degree at the Beijing University of Posts and Telecommunications, China. His major is computer science. His research interests include data mining and data privacy.