True value discovery method based on personalized privacy protection and trust

Abstract

The invention discloses a truth value discovery method based on personalized privacy protection and trust. The truth value discovery method is suitable for the truth value discovery problem considering privacy and existence of malicious user interference in a mobile crowd sensing system. Firstly, a personalized position privacy protection mechanism based on differential privacy is provided, a user is allowed to set a privacy budget according to own needs, then a truth value discovery mechanism based on trust evaluation is introduced, and the privacy budget is found by evaluating the closeness degree of a user perception result and an estimated truth value. And finally, a user selection and payment reward determination mechanism based on reverse auction is adopted, social welfare is maximized through an optimization model, and a heuristic algorithm is designed to find an optimal solution. According to the method, the personalized privacy protection requirement of the user can be met, the accuracy of sensing data can be improved while position privacy protection is achieved, and the method is particularly suitable for crowd sensing scenes with high requirements for privacy protection and data accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of privacy protection and truth discovery of mobile crowd intelligence perception, and in particular to truth value discovery based on personalized privacy protection and trust. Background Art

[0002] With the rapid development of sensor technology and wireless communication networks, mobile smart devices such as smart phones and smart bracelets are becoming increasingly popular, which has laid the foundation for the emergence and application of crowd sensing. As an emerging data perception paradigm, crowd sensing can achieve distributed and cross-temporal and spatial data collection of the target area by recruiting a large number of users to participate and using their smart devices and embedded rich sensors. Since crowd sensing allows ordinary people carrying mobile smart devices to participate in the process of data collection, it greatly expands the temporal and spatial scope of data collection and has broad application prospects. For example, in the smart city scenario, users can be recruited to collect city-scale environmental information, such as temperature and humidity data, noise pollution level, traffic congestion, etc., to assist urban management decisions. However, the widespread application of crowd sensing systems also faces challenges such as privacy protection. The tasks published in crowd sensing are usually closely related to location, requiring users to embed information such as geographic location coordinates when uploading perception data. For example, in environmental quality monitoring, air pollution conditions in different geographical areas are collected, and in road traffic monitoring, traffic congestion data on different sections of roads are collected. This information related to geographic location may leak the user's daily activity trajectory, living habits, social relationships and other private information. Once a privacy leak occurs, it will seriously threaten the user's personal and property safety, affect the user's trust in the crowd sensing system, lead to a decrease in user participation, and restrict the further development of crowd sensing. Therefore, providing effective location privacy protection is a key factor in promoting the sustainable development of crowd sensing systems.

[0003] In recent years, many privacy protection mechanisms have been applied to crowd sensing to protect the location privacy of participating users. Common methods include spatial hiding technology, location obfuscation, and traditional encryption algorithms. The spatial hiding technology-based method reduces the risk of accurate positioning by hiding the user's real location in a larger area; the location obfuscation method adds random noise or random offsets to the location uploaded by the user, making it difficult to directly infer the real location; the traditional encryption algorithm encrypts the location information to prevent sensitive information from being stolen. Although the academic community has conducted some research on location privacy protection in crowd sensing and proposed a variety of privacy protection schemes, the existing schemes still have the following shortcomings: First, most schemes adopt a non-discriminatory privacy protection strategy, treat the importance of all users' privacy equally, do not consider the personalized privacy needs of users, and ignore the differences in user privacy preferences in the design. Different users have different sensitivity and protection needs for privacy. Some users may be willing to sacrifice privacy to a certain extent in exchange for higher data quality and service experience, while others may require a higher level of privacy protection; second, the introduction of privacy protection mechanisms will inevitably lead to a decline in data quality. For example, fuzzy processing and random offset will reduce the accuracy of data, thereby affecting the actual application effect of data. How to find a balance between protecting user privacy and ensuring high data quality is an important challenge in current crowd perception research. In addition, in existing crowd perception networks, in the truth value discovery research involving privacy, malicious users are usually not considered in the process of estimating the truth value. In order to obtain greater benefits, these users do not make corresponding efforts, and their reported perception results often have large errors. This situation will have a greater negative impact on the aggregation results based on truth value discovery, thereby reducing the accuracy of estimation. Therefore, the present invention aims to propose a truth value discovery method based on personalized privacy protection and trust for crowd perception in view of the limitations of the prior art. The method can dynamically adjust the location privacy protection strength according to the user's privacy preference, combine differential privacy technology and individual preferences, and reduce the impact of privacy protection mechanism on data availability by introducing data improvement mechanism while meeting the user's privacy needs, thereby improving the service quality of crowd perception and promoting the further development and application of crowd perception. Summary of the invention

[0004] The purpose of the present invention is to provide a truth value discovery method based on personalized privacy protection and trust. By designing a personalized location fuzzy model and a data improvement function, users can customize the degree of privacy disclosure. In addition, a user selection mechanism that comprehensively considers privacy and trust is introduced to ensure high-quality data acquisition while protecting the user's location privacy, thereby achieving a balance between privacy protection and data availability.

[0005] To achieve the above purpose, the technical solution provided by the present invention is as follows:

[0006] The present invention provides a truth value discovery method based on personalized privacy protection and trust, the method comprising the following steps:

[0007] (1) Step 1: The requester requests the platform to publish a set of crowd-sensing tasks T = {t1,…,t j ,…t N}, the platform publishes this set of tasks and announces it to the user collection

[0008] (2) Step 2: Each user w i Submit a collection of tasks you are interested in and the bid value b i For the platform, i Reflect user w i The minimum acceptable remuneration to compensate for the cost expenditure i , in, It represents the cost of resources consumed by users to perform perception tasks. represents the privacy cost of the user, represents the unit cost of privacy, δ i represents the user's privacy budget;

[0009] (3) Step 3: The platform uses a heuristic mechanism to select users and determine payment rewards, and determines the selected user set as W * , the payment set is P * , this mechanism comprehensively considers the user's trust value Γ i and the privacy budget δ i , and determine the remuneration based on the reverse auction mechanism. The specific implementation method is as follows:

[0010] ①Determine the selected user set W * :First calculate each user w i The marginal social welfare i , φ is the quality-value conversion coefficient, Represents user w i Quality, Γ i Represents user w i The trust value, norm(δ i ) represents the normalized user privacy budget, ρ and (1-ρ) represent the weights of trust and privacy, respectively. For each user w i , if the value of its marginal social welfare is non-negative, then the user is added to the selected user set W *, and then calculate the currently selected user set W * The contribution of users in to each task is calculated, and then each task t is calculated. j The expected difference of the remaining quantity K′ j , Then the user selection cycle begins, and the cycle will continue until the number of tasks is expected to meet the requirements. In each cycle, the user w with the smallest absolute value of marginal social welfare and the smallest contribution to the number of users is selected. opt , Each time the selected user w opt Add to collection W * , and update the expected difference K′ of the remaining number of tasks j , K′ j =max{0,K′ j -1}, when all task quantity expectations meet the requirements, the loop ends and the selected user set W is returned * ;

[0011] ② Determine the platform payment reward set P * :First, for the selected user set W * For each user w l , the user selection process is performed without considering the participation of this user until user w l Interesting task collection Each task in satisfies the quantity expectation, that is, Thus, the currently selected user set W′ is obtained, and then W′ is compared with all user sets W whose marginal social welfare is non-negative. p If the former is greater than or equal to the latter, then it is necessary to traverse each user w z ,z∈W′\W p , calculate user w l Maximum bid price b′ l Make it possible to replace user w z , that is, satisfy Then b′ l The value of is represented as follows: All b′ calculated l The maximum value among them is taken as user w l The payment of p l , otherwise, the user who pays w l The reward is equal to its bid price, determine the set W * After each user pays the reward, the platform payment set P can be determined * ;

[0012] (4) Step 4: The platform notifies the selected user set W *The users in the platform perform tasks. According to the requirements of the tasks published by the platform, the users submit location information and perception data of the corresponding area. To conveniently represent the location information of users and tasks, the entire network is divided into L areas. Set W * For each user w i First, the real position Mapped to another fuzzy position l′ through the position mapping mechanism i , by reporting the fuzzy position l′ to the platform i To protect location privacy, the specific implementation is as follows: Each user w i According to the privacy budget δ i Customize your own location probability mapping model to determine the user's real location The probability of mapping to any position in L regions (including its own position) The calculation formula is as follows:

[0013]

[0014] Among them, e * is the perception data bias adjustment factor, which represents the maximum allowed true value difference, It represents user w i Since the reported fuzzy position l′ i The area deviates from the original true position The difference in the true value of the perceived data caused by the user selects the ambiguous position l′ according to the probability results of different position mappings i , and then based on the data improvement mechanism, the original real position Area-aware data Fit to the fuzzy position l′ i area, obtaining improved perception data The specific method of data improvement is as follows: the platform learns the data correlation between the areas before and after the fuzzy position based on historical perception data, and obtains a linear regression model f(*) to reduce the data quality loss caused by the fuzzy position to protect user privacy. The calculation formula is as follows: The user then sets the blurred position l i and improved sensory data Send it to the platform and receive a pre-determined reward from the platform;

[0015] (5) Step 5: After the platform receives the data submitted by the user, it conducts truth value discovery and trust evaluation on the user, and updates the user's trust value. The specific implementation method is as follows: For each task t j ,j∈[1,N], assuming that there are K users providing perception data i∈[1,K], initially assign an initial weight to each user is 0.5, and then the user data is weighted and summed according to the weight to obtain the estimated true value That is, data aggregation is performed, and the calculation formula is as follows:

[0016]

[0017] The platform then uses the data submitted by the user and the current estimated true value The difference between them updates the user's weight, and the calculation formula is as follows:

[0018]

[0019] in, Represents the perception data measured by user reports Estimated true value The distance, d(*) represents the function to measure the distance, F(*) represents a monotonically decreasing function, and the iterative process continues until the convergence condition is met, that is, the estimated true value If the change in two consecutive iterations is less than the set threshold, the user's trust value is updated. The calculation formula is as follows:

[0020]

[0021] Among them, Θ and represents the trust adjustment parameter, Θ∈[0,1], Depending on the allowed perception task error threshold, Represents user w i The distance between the reported perception data and the estimated true value,

[0022] (6) Step 6: The platform collects the data j∈[1,N] is fed back to the task requester and gets rewarded.

[0023] Beneficial Effects

[0024] The present invention provides a truth value discovery method based on personalized privacy protection and trust, which allows users to customize privacy budgets according to their own needs. By combining differential privacy technology and the concept of privacy budget, users can flexibly set the degree of privacy protection. The larger the privacy budget, the more relaxed the user's requirements for privacy protection and the more willing they are to disclose location information; conversely, the smaller the privacy budget, the higher the user's demand for privacy protection and the less location information disclosed. This method meets the personalized privacy protection needs of different users and provides a flexible and controllable crowd-sensing user location privacy protection mechanism.

[0025] While providing personalized location privacy protection, the present invention also takes into account the availability and quality of perceived data. Through the two stages of location blurring and data improvement, the impact on data quality is reduced while protecting privacy. The location blurring stage adopts an exponential-based location probability mapping mechanism. For fuzzy locations with smaller differences in perceived data, the greater the probability of user mapping, the similarity between the data at the fuzzy location and the real location is ensured. The data improvement stage utilizes the spatiotemporal correlation of perceived data, obtains a regression model by training historical perceived data, and fits the perceived data of the real location to the mapped fuzzy location, further reducing data quality loss.

[0026] On the basis of protecting the user's location privacy, the present invention also effectively addresses the interference of malicious users on the truth discovery results. By designing a heuristic-based user selection mechanism, the mechanism comprehensively considers the user's privacy budget and trust value, thereby optimizing the overall data quality while ensuring privacy protection. Specifically, users whose report results are closer to the estimated true value will obtain a higher trust value, and those users with high reliability and low data uncertainty will be selected to perform tasks first. In addition, the present invention introduces a reverse auction model to transform the user selection problem into an integer linear optimization problem to maximize social welfare. Therefore, while improving the accuracy of truth discovery, the user's location privacy protection needs are met. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart of the truth value discovery method based on personalized privacy protection and trust of the present invention;

[0028] Figure 2 This is a schematic diagram of comparing probability fuzzy matrices under different privacy budgets in an embodiment of the present invention;

[0029] Figure 3 Schematic diagram of the effect of residual standard error on fuzzy probability under different privacy budgets in an embodiment of the present invention;

[0030] Figure 4 A schematic diagram showing a comparison of mean absolute errors of finding the true value of user report data using different methods in an embodiment of the present invention; DETAILED DESCRIPTION

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments. Figure 1 This is a flow chart of the truth value discovery method based on personalized privacy protection and trust of the present invention, such as Figure 1 As shown, the main process of the present invention includes:

[0032] The requester requests the platform to publish a set of crowd-sensing tasks T = {t1,…,t j ,…t N}, the platform publishes and announces this set of tasks to the user set W = {w1,…,w i ,…wM};

[0033] Each user w i Submit a collection of tasks you are interested in and the bid value b i For the platform, i Reflect user w i The minimum acceptable remuneration to cover the cost i , in, It represents the cost of resources consumed by users to perform perception tasks. represents the privacy cost of the user, represents the unit cost of privacy, δ i represents the user's privacy budget;

[0034] The platform models the user selection problem based on reverse auction as the following integer linear optimization problem:

[0035]

[0036] Where φ is the quality-value conversion coefficient, Represents user w i The mass, x i Represents user w i Whether the task is selected or not, since the platform does not know the actual task and cost value that the user is interested in in advance, the user's bidding value is used in the target design. represents the marginal social welfare of the user based on the bid value, so the maximization objective function is designed as:

[0037] The platform uses a heuristic mechanism to select users and determine payment rewards, and determines the selected user set as W * , the payment set is P * , this mechanism comprehensively considers the user's trust value Γ i and the privacy budget δ i , and determine the remuneration based on the reverse auction mechanism. The specific implementation method is as follows: ① Determine the selected user set W * :First calculate each user w i The marginal social welfare i , Γ i Represents user w i The trust value, norm(δ i ) represents the normalized user privacy budget, ρ and (1-ρ) represent the weights of trust and privacy, respectively. For each user w i, if the value of its marginal social welfare is non-negative, then the user is added to the selected user set W * , and then calculate the currently selected user set W * The contribution of users in to each task is calculated, and then each task t is calculated. j The expected difference of the remaining quantity K′ j , K′ j = Then the user selection cycle begins, and the cycle will continue until the number of tasks is expected to meet the requirements. In each cycle, the user w with the smallest absolute value of marginal social welfare and the smallest contribution to the number of users is selected. opt , Each time the selected user w opt Add to collection W * , and update the expected difference K′ of the remaining number of tasks j , K′ j =max{0,K′ j -1}, when all task quantity expectations meet the requirements, the loop ends and the selected user set W is returned * ; ② Determine the platform payment reward set P * :First, for the selected user set W * For each user w l , the user selection process is performed without considering the participation of this user until user w l Interesting task collection Each task in satisfies the quantity expectation, that is, Thus, the currently selected user set W′ is obtained, and then W′ is compared with all user sets W whose marginal social welfare is non-negative. p If the former is greater than or equal to the latter, then it is necessary to traverse each user w z ,z∈W′\W p , calculate user w l Maximum bid price b′ l Make it a replacement for user w z , that is, satisfy Then b′ l The value of is represented as follows: All b′ calculated l The maximum value among them is taken as user w l The payment of p l , otherwise, the user who pays w l The reward is equal to its bid price, determine the set W * After each user pays the reward, the platform payment set P can be determined * ;

[0038] The platform notifies the selected user set W* The users in the platform perform tasks. According to the requirements of the tasks published by the platform, the users submit location information and perception data of the corresponding area. To conveniently represent the location information of users and tasks, the entire network is divided into L areas. Set W * For each user w i First, the real position Mapped to another fuzzy position l′ through the position mapping mechanism i , by reporting the fuzzy position l′ to the platform i To protect location privacy, the specific implementation is as follows: Each user w i According to the privacy budget δ i Customize your own location probability mapping model to determine the user's real location The probability of mapping to any position in L regions (including its own position) The calculation formula is as follows: Among them, e * is the perception data bias adjustment factor, which represents the maximum allowed true value difference, It represents user w i Since the reported fuzzy position l′ i The area deviates from the original true position The difference in the true value of the perceived data caused by the user selects the ambiguous position l′ according to the probability results of different position mappings i , and then based on the data improvement mechanism, the original real position Area-aware data Fit to the fuzzy position l′ i area, obtaining improved perception data The specific method of data improvement is as follows: the platform learns the data correlation between the areas before and after the fuzzy position based on historical perception data, and obtains a linear regression model f(*) to reduce the data quality loss caused by the fuzzy position to protect user privacy. The calculation formula is as follows: The user then sets the blurred position l′ i and improved sensory data Send it to the platform and receive a pre-determined reward from the platform;

[0039] After receiving the data submitted by the user, the platform aggregates the data and conducts trust evaluation on the user, and updates the user's trust value. The specific implementation method is as follows: for each task t j ,j∈[1,N], assuming that there are K users providing perception data i∈[1,K], initially assign an initial weight to each user is 0.5, and then the user data is weighted and summed according to the weight to obtain the estimated true value That is, data aggregation is performed, and the calculation formula is as follows: The platform then uses the data submitted by the user and the current estimated true value The difference between them updates the user's weight, and the calculation formula is as follows: in, Represents the perception data measured by user reports Estimated true value The distance, d(*) represents the function to measure the distance, F(*) represents a monotonically decreasing function, and the iterative process continues until the convergence condition is met, that is, the estimated true value If the change in two consecutive iterations is less than the set threshold, the user's trust value is updated. The calculation formula is as follows: Among them, Θ and represents the trust adjustment parameter, Θ∈[0,1], Depending on the allowed perception task error threshold, Represents user w i The distance between the reported perception data and the estimated true value,

[0040] The platform will gather data j∈[1,N] is fed back to the task requester and gets rewarded.

[0041] In order to verify the feasibility and effectiveness of the truth value discovery method based on personalized privacy protection and trust proposed in this paper, an experimental simulation is carried out to verify the method. Each group of perception tasks contains 10 regional subtasks. Figure 2 to Figure 3 The experimental results lead to the following conclusions:

[0042] 1. Figure 2 The experimental results of the probability fuzzy matrix under different privacy budgets are given, indicating that user w i The privacy budget is δ i =0.1 and δ i =5, the probability of different location areas being mapped to other locations. According to sub-graphs (a) and (b), it can be observed that user w i The more likely it is that the ambiguous position is concentrated in the real position, that is, the mapping probability of the position in the diagonal area is higher. According to the position probability mapping function, users tend to choose the difference between the perceived data and the true value. The smaller the location. As the privacy budget δ i As δ increases, this trend becomes more obvious. i = 0.1, the probability of users choosing to map to different locations is close, so it is difficult for the adversary to infer the real location based on the fuzzy location reported by the user, which better protects the user's location privacy.

[0043] 2. Figure 3 The experimental results of probability blurring under different privacy budgets under different residual standard errors are given, indicating the impact of different residual standard errors on the blurring probability of different areas of the user. Considering the selection of the real location area 1, sub-graph (a) shows the data quality loss caused by the user blurring the real location to different location areas. If it is mapped to the original location area 1, there is no data quality loss. Comparing sub-graphs (b) and (c), we can see the probability of the user selecting different blurred locations. The smaller the residual standard error, the greater the probability of the user selecting the blurred location. When δ i When =5, the probability that the user chooses to report the real location is the highest. Due to the large privacy budget cost, it means that the user is more willing to disclose the location privacy.

[0044] 3. Figure 4 The mean absolute error (MAE) of the truth-finding aggregation results of user-reported data under different methods is compared, where the number of candidate users is 300 and the number of recruited users is 8. Without considering privacy, the data quality of the truth-finding aggregation results is the best, because each user reports the perception data of the real location, so the mean absolute error is the smallest, but the privacy of the user is not protected. Among the other methods compared, random selection, trust-based selection method (only trust) and privacy-based selection method (only privacy), the data quality is reduced due to the fuzzy location submitted to the platform. Compared with other privacy-preserving methods, the truth-finding aggregation results obtained by the optimized user selection mechanism proposed in this invention are of better quality and have the lowest mean absolute error. In particular, for monitoring tasks in different location areas, the proposed mechanism reduces the mean absolute error by 23.74% to 31.90% and 15.60% to 23.23% compared with random selection and trust-based selection method (only trust), respectively. Although the mean absolute error is similar to that of the selection method based only on privacy protection, the optimized user selection mechanism takes both trust value and privacy cost into consideration and can obtain higher quality perception data aggregation results.

Claims

1. A truth value discovery method based on personalized privacy protection and trust, characterized in that: The steps include: (1) Step 1: The requester requests the platform to publish a set of crowd-sensing tasks T = {t1,…,t j ,…t N }, the platform publishes this set of tasks and announces it to the user collection (2) Step 2: Each user w i Submit a collection of tasks you are interested in and the bid value b i For the platform, i Reflect user w i The minimum acceptable remuneration to cover the cost i , in, It represents the cost of resources consumed by users to perform perception tasks. represents the privacy cost of the user, represents the unit cost of privacy, δ i represents the user's privacy budget; (3) Step 3: The platform uses a heuristic mechanism to select users and determine payment rewards, and determines the selected user set as W * , the payment set is P * , this mechanism comprehensively considers the user's trust value Γ i and the privacy budget δ i , and determine the remuneration based on the reverse auction mechanism. The specific implementation method is as follows: ①Determine the selected user set W * :First calculate each user w i The marginal social welfare i , φ is the quality-value conversion coefficient, Represents user w i Quality, Γ i Represents user w i The trust value, norm(δ i ) represents the normalized user privacy budget, ρ and (1-ρ) represent the weights of trust and privacy, respectively. For each user w i , if the value of its marginal social welfare is non-negative, then the user is added to the selected user set W * , and then calculate the currently selected user set W * The contribution of users in to each task is calculated, and then each task t is calculated. j The expected difference of the remaining quantity K′ j , Then the user selection cycle begins, and the cycle will continue until the number of tasks is expected to meet the requirements. In each cycle, the user w with the smallest absolute value of marginal social welfare and the smallest contribution to the number of users is selected. opt , Each time the selected user w opt Add to collection W * , and update the expected difference K′ of the remaining number of tasks j , K′ j =max{0,K′ j -1}, when all task quantity expectations meet the requirements, the loop ends and the selected user set W is returned * ; ② Determine the platform payment reward set P * :First, for the selected user set W * For each user w l , the user selection process is performed without considering the participation of this user until user w l Interesting task collection Each task in satisfies the quantity expectation, that is, So as to obtain the currently selected user set W ′ , and then by comparing W ′ and the set W of all users whose marginal social welfare is non-negative p If the former is greater than or equal to the latter, then it is necessary to traverse each user w z ,z∈W′\W p , calculate user w l Maximum bid price b′ l Make it possible to replace user w z , that is, satisfy Then b′ l The value of is represented as follows: All b′ calculated l The maximum value among them is taken as user w l The payment of p l , otherwise, the user who pays w l The reward is equal to its bid price, determine the set W * After each user pays the reward, the platform payment set P can be determined * ; (4) Step 4: The platform notifies the selected user set W * The users in the platform perform tasks. According to the requirements of the tasks published by the platform, the users submit location information and perception data of the corresponding area. To conveniently represent the location information of users and tasks, the entire network is divided into L areas. Set W * For each user w i First, the real position Mapped to another fuzzy position l′ through the position mapping mechanism i , by reporting the fuzzy position l′ to the platform i To protect location privacy, the specific implementation is as follows: Each user w i According to the privacy budget δ i Customize your own location probability mapping model to determine the user's real location The probability of mapping to any position in L regions (including its own position) The calculation formula is as follows: Among them, e * is the perception data bias adjustment factor, which represents the maximum allowed true value difference, It represents user w i Since the reported fuzzy position l′ i The area deviates from the original true position The difference in the true value of the perceived data caused by the user selects the ambiguous position l′ according to the probability results of different position mappings i , and then based on the data improvement mechanism, the original real position Area-aware data Fit to the fuzzy position l′ i area, obtaining improved perception data The specific method of data improvement is as follows: the platform learns the data correlation between the areas before and after the fuzzy position based on historical perception data, and obtains a linear regression model f(*) to reduce the data quality loss caused by the fuzzy position to protect user privacy. The calculation formula is as follows: The user then sets the blurred position l′ i and improved sensory data Send it to the platform and receive a pre-determined reward from the platform; (5) Step 5: After the platform receives the data submitted by the user, it aggregates the data and conducts trust evaluation on the user, and updates the user's trust value. The specific implementation method is as follows: For each task t j ,j∈[1,N], assuming that there are K users providing perception data Initially assign an initial weight to each user is 0.5, and then the user data is weighted and summed according to the weight to obtain the estimated true value That is, data aggregation is performed, and the calculation formula is as follows: The platform then uses the data submitted by the user and the current estimated true value The difference between them updates the user's weight, and the calculation formula is as follows: in, Represents the perception data measured by user reports Estimated true value The distance, d(*) represents the function to measure the distance, F(*) represents a monotonically decreasing function, and the iterative process continues until the convergence condition is met, that is, the estimated true value If the change in two consecutive iterations is less than the set threshold, the user's trust value is updated. The calculation formula is as follows: Among them, Θ and represents the trust adjustment parameter, Θ∈[0,1], Depending on the allowed perception task error threshold, Represents user w i The distance between the reported perception data and the estimated true value, (6) Step 6: The platform collects the data Feedback to task requester and get paid.

Images