Mobile crowd sensing (MCS) network collects scenario, environmental, and individual data within a specific range via the intelligent sensing equipment carried by the mobile users, thus providing social decision-making services. MCS is emerging as a most important sensing paradigm. However, the person-centered sensing itself carries the risk of divulging users’ privacy. To address this problem, we proposed a variable weight privacy-preserving algorithm of secure multiparty computation. This algorithm is based on privacy-preserving utility and its effectiveness and feasibility are demonstrated through experiment.

1. Basic Theories

1.1. Architecture of Mobile Crowd Sensing Network

Mobile crowd sensing (MCS) network [1] takes the ordinary mobile terminals as the basis sensing units. The sensing task distribution and sensing data collection are achieved through collaboration via the mobile Internet. This represents a large-scale complex social sensing task. “Crowd” refers to the aspect of mobilizing the power and intelligence of the general public, and “sensing” is the process of acquiring the users’ behavioral data under different scenarios using the sensors.

Figure 1 shows a typical MCS framework, which consists of the mobile users and the sensing platform. Mobile users are millions of mobile intelligent terminals, into which sensors are embedded (e.g., GPS, gravity sensor, temperature sensor, camera, microphone, and acceleration sensor). These sensors collect various sensing data, which are updated to the sensing platform via the mobile network or short-range wireless communication network. Upon receiving the data, the sensing platform will commence data analysis and processing. The processed data will be directly applied to a diversity of universal social sensing services. After the data analysis and processing are finished, each parcel of data will be evaluated. The mobile users participating in the sensing tasks will be awarded based on the specific incentive mechanism, so as to attract more users into the large-scale sensing task. Liu proposed schemes based on both the Monopoly and Oligopoly models enhancing the location privacy of MCS applications by reducing the bidding and assignment steps in the MCS cycle [2]. Jin proposed a differentially private incentive mechanism that preserves the privacy of each worker’s bid against the other honest-but-curious workers [3]. Furthermore, many researchers focused on the detailed information extraction processing in MCS including Hybrid Deep Learning Architecture [4] and Fog Computing and Data Aggregation Scheme [5, 6].

1.2. Application of the MCS Network

The MCS network comprising the mobile intelligent terminals and the mobile sensors is capable of large-scale, complex, fine-grained, and thorough data sensing and collection. For example, the use of MCS network for the collection, analysis, and fusion of the urban traffic flow information can provide highly efficient and convenient path planning and driver assistance system for the mobile users. The MCS network can also provide the decision-making support for urban transport planning and for the formulation of a safe and highly efficient urban transportation network. The MCS network-based sensing and monitoring of urban domestic infrastructures offer convenient life services for the local residents. The wide prevalence of the mobile intelligent terminals is a solid guarantee for the high-efficiency and low-cost and large-scale monitoring of natural environment in the cities.

2. Privacy Protection Mechanism for MCS Network Users

2.1. Privacy Preserving in MCS Network

The sensing data collected by the MCS network are largely user privacy. Location data usually contain the sensitive information such as users’ address, scope of activity, and transportation route. The mining of users’ state of motion can obtain the sensitive information of users’ living habits and health conditions. The biological data collected contains the information of users’ voice, fingerprints, and basic physiological characteristics. The routine usage data of the mobile intelligent terminals are associated with the user privacy of a deeper level, including the users’ hobbies and behavioral traits. Once the user privacy is divulged, there may be violation of privacy, harassment, fraud, or even direct economic loss. Therefore, designing the data security architecture for dynamic privacy protection under the MCS network is an urgent issue.

2.2. Related Technology of Privacy Preserving

The major privacy-preserving techniques used for MCS network are divided into the following types.

(1) Generalized Privacy-Preserving Algorithm. Anonymization is performed while sharing the sensing data, so that the sensitive information about the user’s identity is removed without harming the meaningful deduction based on the anonymized sensing data. However, the currently used anonymization methods are usually greedy algorithms which have low execution efficiency.

(2) Perturbation-Based Privacy-Preserving Algorithm. The raw sensing data are perturbed by adding a random number, noises, and exchanges, so that the other party cannot mine the raw sensing data and privacy policies. The main difficulty with data perturbation is how to strike a balance between data correctness, privacy, and security.

(3) Secure Multiparty Computation (SMPC). This technique integrates data encryption and multiple parties are involved in the computation and mining. Because none of the parties have access to complete data, the users’ privacy can be ensured. SMPC is now used for collaborative computing among a group of untrusted parties. Many researches have been carried out over the SMPC problem. In 2000, Lindell proposed the method of secure multiparty decision tree (ID3) to protect the data privacy of users [7]. Asharov proposed the threshold homomorphic encryption scheme to improve efficiency of the privacy protection algorithm [8]. In 2014, the threshold-based encryption of -means outsourcing computing proposed by Liu is a more efficient privacy protection algorithm [9].

Proper application of information technology and algorithm design are the two major concerns in privacy protection. However, the users’ attitudes towards privacy are generally neglected. A survey [10] indicates that 17% of the Internet users are still unwilling to provide their authentic information even under privacy protection; 56% of the Internet users are more willing to provide their authentic information in the presence of proper privacy protection; the remaining 27% of the Internet users do not particularly care about their privacy and will provide the authentic information with or without privacy protection. It is obvious that the users’ attitudes towards privacy affect their willingness to share the personal information. Users may react differently to the prospect of disclosing different personal information. But under some incentive mechanisms, the psychological response of the users to the disclosure of different sensitive information may vary.

This study constructed an MCS network-based privacy-preserving algorithm by reference to SMPC. The weight function of privacy preference was built by combining the analysis of the users’ sensitivity to the disclosure of privacy and classification of the privacy level of the sensing data. This proposed algorithm can effectively prevent the divulging of privacy information while achieving a maximal acquisition and analysis of the sensing data.

3. Variable Weight Privacy-Preserving Algorithm

3.1. Measure of User Privacy Sensing
3.1.1. User Multiattribute Assumption

Suppose there exists Euclidean space, in which dimensions represent solutions to one problem; denotes the attribute , and is a set of attributes, . denotes one solution, and is a set of one solution, . denotes the attribute value of the solution under attribute . denotes a decision-making matrix of solution under the attribute :

Considering the varying sensitivity to privacy, the users show different willingness to share their privacy in the MCS network. The influence factors of this willingness are divided into profit factors and risk factors, each of which is measured differently. Let be the set of the profit attributes, and be the set of risk attributes. The two sets are normalized by multiple attribute decision-making using the following formula:

After the transform, the synthetic matrix is .

3.1.2. Weight Determination of Privacy Perception Attributes

As the users differ in privacy perception, each attribute will carry the information of different user preferences. Therefore, the given user preference can be expressed as the weight of the individual, and the weight of each attribute is expressed as

The utility of each user is expressed as the sum of the weighted attributes. Hence, the user utility is

The utility analysis of users’ privacy perception will provide not only the weight parameters for the SMPC, but also some suggestions for the collection modes of the sensing data under the MCS network. For example, the privacy information sensitive to most users will be prevented and a reasonable incentive mechanism can be designed on this basis. This is very important for increasing the confidence and participation level of users with a lower utility of privacy perception.

3.2. Variable Weight SMPC-Based Privacy-Preserving Algorithm
3.2.1. SMPC-Based Algorithm

SMPC can be conceptualized by the following mathematical model: participants of the protocol jointly implement the function . is the set of input variables. The set of input variables provided by the participant is a subset of , which satisfies . It is required in the computing of the function that the input from any participant is not known to other participants .

The essence of SMPC is a data encryption algorithm using the encryption scheme so as to ensure data privacy. Rivest et al. [11] proposed the concept of fully homomorphic encryption in 1978, aiming to construct an encryption mechanism that supports ciphertext retrieval. Goldwasser [12] studied the strategies used by mobile attackers in the secure channel model. They generalized the threshold mechanism to the ordinary SMPC. The plaintext will be revealed only when at least participants are involved in the collaborative decryption. This effectively restricts the access to the final SMPC output and the participants will not disclose the data.

3.2.2. Weighted Threshold Secret Sharing Scheme Based on Mignotte Sequence

The weighted threshold secret sharing scheme refers to that each participant assumes a different role, based on which different weights are assigned. The conventional weighted threshold secret sharing schemes achieve only works on the premise of assigning more secret shares to those who are given special permission. However, this will increase the insecurity of key management and transmission. In this study, we adopted the weighted threshold secret sharing scheme based on Mignotte sequence. Regardless of the weight, each participant is only allowed one private key and there is no transmission of secret information between the participants and the dealer. Therefore, the cost of key transmission and storage is spared.

Mignotte sequence is defined as follows [13]:

Let . If the integer sequence satisfies(1);(2), where ;(3),

then sequence is called a ()-Mignotte sequence.

The weighted threshold secret sharing scheme based on the Mignotte sequence is designed.

(1) Parameter Configuration. In this scheme, the dealer assigns the weights to each participant using a digit with a length of large prime. The secret to be shared is determined and the relevant system parameters are configured. There are participants and they constitute the set . The weight vectors of the participants are correspondingly . The threshold is , and the secret to be shared is .

(2) Construction and Expansion of Mignotte Sequence. The dealer needs the system parameters to construct an expanded Mignotte sequence fit for the weighted threshold secret sharing scheme. Meanwhile, the converted scheme should be equivalent to the original scheme. A ()-Mignotte sequence is constructed as , which is expanded into

The above sequence is a sequence of primers, where which makes the sequence satisfy the following conditions:(a)The product of the last numbers is smaller than the product of the first numbers.(b).

Let .

From above, it can be known that sequence has the following property: ().


Thus sequence is the expanded Mignotte sequence, denoted as ()-Mignotte sequence. This sequence is revealed.

(3) Generation of Secret Shares. The dealer computes the secret shares of each participant according to the Mignotte sequence and the shared secret:

This is sent to the participant via the secret channel.

(4) Secret Restoration. Suppose there are participants who constitute the set , and restore the secret. The vector weights for each participant in constitute the set .

When the sum of the weight vectors of each participant in is above or equal to the threshold, that is, , the following congruence equations are constructed: , where and the solution is the shared secret .

4. Implementation and Deployment

MapReduce System was used for the high-efficiency parallel processing in the large-scale matrix multiplication in the weighted threshold secret sharing scheme. On the simple data center comprising 5 host machines, the Hadoop distributed storage and computing environment was deployed as a mimic of the sensing platform in the MCS network. One host machine was the Master node, which was deployed with the roles of NameNode and JobTracker for the management of distributed data and task decomposition; 4 host machines were the Slaver and were deployed with the roles of DataNode and TaskTracker for the distributed data storage and task execution. The implementation and deployment (Figure 2) are illustrated below.

() The initialization program at the data center would preset the system parameters. The threshold was determined. The weight vectors of each participant were initialized. The key management system as the trusted third party generated pairs of homomorphic public and private keys. The public keys hom_PK were the same, and the private keys were distributed to different participant nodes.

() The block function MR_Splitter( ) in the MapReduce System was responsible for dividing the sensing data files submitted by the clients in the MCS network into blocks. Each block was 64 M. The data blocks were encrypted using the public key hom_PK. The encrypted data block file is stored in the distributed file system of the DataNode.

() An intermediate pair was computed during the matrix multiplication in the privacy-preserving algorithm. The map nodes were allocated to each operation. Before the mapper output the pair, the ciphertext for each participant was generated using formulae (7).

() Reducer replicated the intermediate output of the corresponding division from the mapper output terminal to the local file system.

() At least participants were involved in the decryption of the ciphertext using the decryption algorithm in formulae (8). These participants would share the decrypted information with other participants. The information decrypted by the participants was then combined with the information decrypted by the remaining participants to obtain the final result.

For users in network society, we divide them into three groups according to the weights aforementioned in Sections 2 and 3, that is, privacy careless person (group A), practical privacy person (group B), and the group who protect their privacy strictly (group C). In group A, they are not so sensitive with privacy and willing to share their true information. In group B, they may share personal files while policies and regulations are carefully learned. In group C, they are not interested in any sharing information activities at any circumstances.

In a certain survey, the percentage results of groups A, B, and C are obtained as 33.1%, 57.4%, and 9.5% from 352 users on the Internet, and we can initiate the weights of sharing by 0.9, 0.5, and 0.1. These parameters are easily adjusted during privacy protection mechanism proposed here.

As is shown in Figure 3, the three groups in privacy iteration results are given. Group A indicates that since they are not concerned about their information, those provided data are true and the efficiency is acceptable. Group B is matching data on the condition that they believe the privacy is protected, so that their efficiency is not stable and high. Group C are not willing to share their information, and their provided information is not all correct, which influences the computing fundamentally.

5. Algorithm Performance Analysis

5.1. Security Analysis

The private information of each participant is randomly divided into fragments in a certain way. Each participant selects one fragment randomly and preserves it. The remaining fragments are randomly allocated to other participants. After the fragments are reallocated according to the protocol, each participant will own an equal amount of fragments. Each participant owns one fragment of his or her information plus one fragment transmitted from another participant. Therefore, even if participants and conspire, they can only infer the reallocated information of participant and do not know other private information . Any two conspiring participants can only infer the reallocated information of the third party. Then, combining with the information fragments owned by themselves, they can infer the private information of the third party. But when there are more than 3 participants, it will be very difficult to infer all information of the other participants by conspiracy. When there are more than 4 participants and when most participants are honest, the possibility of information leak will approach 0.

5.2. Complexity Analysis

Computational complexity: each round of computation consists of operations (different from the aforementioned), and rounds involve operations. Thus the computational complexity is expressed as , as shown in Figure 4.

Communication Complexity. Each participant needs to transmit fragments to other participants. Therefore, in the fragment transmission stage, communications will occur. In the computing stage, each participant needs to transmit the summation of some fragments to other participants over the ring structure. Therefore, each round of computation consists of communications, and rounds involve communications. The overall communication complexity of the algorithm is expressed as , as shown in Figure 5.

6. Summary and Forecast

To protect against privacy violation in the MCS network, we proposed a variable weight SMPC-based privacy-preserving algorithm. The weighted threshold secret sharing scheme based on Mignotte sequence was applied for the encryption of the sensing data and private key management. Considering the different attitudes of users towards the disclosure of the private information, the privacy of the information was graded. Thus the weight parameters of the privacy-preserving algorithm were determined based on the utility analysis of the users’ privacy perception. The proposed model was deployed in the Hadoop distributed environment to verify its effectiveness and validity. The implementation of the SMCP protocol requires several participants, among which communications are necessary. This will incur significant communication and computational costs. How to enhance the reliability of channel communication and to increase the efficiency of sensing data encryption are issues awaiting resolution.

Conflicts of Interest

The authors declare that there are no conflicts of interest regarding the publication of this paper and confirm that the mentioned received funding in the Acknowledgments did not lead to any conflicts of interest regarding the publication of this manuscript.


This work was supported by the Natural Science Foundation of Hainan Province (no. 20166216 and no. 617033) and Education and Reaching Research Project of Hainan University (no. hdjy1325) investigated by Jiezhuo Zhong; National Natural Science Foundation of China (no. 61661019), the Major Science and Technology Project of Hainan Province (no. ZDKJ2016015), the Natural Science Foundation of Hainan Province (no. 20156217), and the Higher Education Reform Key Project of Hainan Province (no. Hnjg2017ZD-1) by Chunjie Cao; National Science and Technology Support Program (no. 2015 BAH55F01-5) and Natural Science Foundation of Hainan Province (no. 614232) investigated by Wenlong Feng.