Privacy protection task coverage and excitation method based on double-cloud architecture in mobile crowdsourcing

Abstract

The invention discloses a privacy protection task coverage and excitation method based on a double-cloud architecture in mobile crowdsourcing. The method aims at solving the problems of position information leakage, low high-difficulty task coverage rate, high excitation cost and the like in traditional crowdsourcing task allocation. A double-cloud server framework is constructed, privacy protection distance calculation between a worker and a task is achieved through a confusion circuit protocol, and leakage of position information of the worker and the task is avoided; meanwhile, a task coverage incentive mechanism based on a target gradient effect is designed, the workers are stimulated to continuously participate in the tasks by setting a gradually-propelled reward target, and the acceptability of the workers to the remote tasks difficult to cover is improved. The key technology of the invention is as follows: (1) a privacy protection distance calculation mechanism is provided, a real distance is jointly calculated in a double-cloud environment by disturbing and splitting position data, and both precision and privacy are considered; and (2) a nonlinear incentive mechanism based on behavioral economics is designed, so that the acceptance willingness of workers to high-cost tasks is remarkably improved, and a higher task coverage rate is realized on the premise of not remarkably increasing the incentive cost. Experimental results show that compared with an existing method, the method has the advantages that the task coverage rate is increased by 14.25% at most, and the excitation cost is reduced by 15.17% on average.

Description

Technical Field

[0001] This invention relates to the fields of mobile crowdsourcing and privacy computing, and in particular to a task coverage and incentive method based on a dual-cloud server architecture, under the condition that the locations of both task participants and task requesters must be kept confidential. Background Technology

[0002] Mobile crowdsourcing (MCS) is a new paradigm for collecting sensor data in urban spaces. Platforms incentivize mobile device users (such as smartphone or wearable device owners) located in various locations to complete data collection, uploading, and feedback by issuing tasks. This model is widely used in many areas of smart city construction, such as traffic congestion monitoring, noise heat mapping, and air quality sensing. Compared to traditional fixed sensor networks, mobile crowdsourcing offers advantages such as flexible deployment, low cost, and strong scalability, and has become one of the important infrastructures for urban intelligence.

[0003] In a typical mobile crowdsourcing system, the task requester (hereinafter referred to as the "requester") is the data consumer, whose task is usually described as "collecting certain information at a certain location and time," such as collecting traffic flow information at a certain intersection in a city during a certain time period. Users with perception and mobility capabilities act as task executors (hereinafter referred to as "workers"), using their own devices (such as taking photos, recording audio, and measuring environmental data) to complete the task at a specific location and submit the data to the platform. The platform, acting as an intermediary, is responsible for task posting, worker recruitment, data quality verification, and result aggregation, and finally provides the final data feedback to the requester.

[0004] Completing tasks typically involves time consumption, relocation costs, and equipment usage, especially when the task location is far from the worker's current location, significantly reducing their willingness to participate. Therefore, the platform needs to provide appropriate compensation for task performers and construct a reasonable incentive mechanism to attract and motivate worker participation. Incentive strategies not only affect platform expenditures but also directly relate to the spatial coverage of tasks and the quality of data collection, thus impacting the overall system efficiency. However, in practical deployment, the incentive design for mobile crowdsourcing faces two key challenges:

[0005] First, there's the issue of location information privacy. To achieve efficient task allocation, platforms typically need to know the spatial distance between workers and tasks to prioritize assigning tasks to workers closer to them, thereby reducing worker movement costs and saving on platform incentive expenditures. However, requiring workers and task requesters to directly upload their real location information would seriously threaten their privacy. Most existing research can only protect location privacy on one side (worker or requester). A few bilateral location privacy protection schemes employ differential privacy, region segmentation, and matrix encryption, which, while able to hide worker and task location information to some extent, usually reduce the accuracy of distance calculations, causing task allocation results to deviate from the optimal path. Therefore, how to achieve accurate and efficient task allocation without exposing the real locations of workers and tasks to third-party platforms remains a crucial challenge.

[0006] Secondly, there are issues of incentive failure and high coverage costs. Even if the platform can accurately calculate the distance between workers and tasks while protecting privacy and prioritize assigning tasks to geographically closer workers, it still cannot guarantee the final execution of the tasks. This is because the distribution of tasks in mobile crowdsourcing systems typically exhibits significant spatial non-uniformity. Some tasks are located in remote areas, and executing these tasks incurs high time or economic costs for workers, making it difficult to attract their active participation; these are often referred to as "high-difficulty tasks." Such tasks face problems of low coverage and insufficient data collection. To incentivize workers to complete these tasks, previous studies have mostly adopted linear incentive mechanisms, where the compensation paid to workers is proportional to their travel distance. However, behavioral economics research shows that workers have a "loss aversion" tendency: when the travel distance required to perform a task exceeds their psychological expectations, workers' sensitivity to travel losses increases significantly, and their demand for compensation grows exponentially. Therefore, traditional linear incentive models are difficult to effectively drive workers to complete high-difficulty tasks. Although platforms can increase task acceptance rates by introducing exponential incentive mechanisms, the incentive costs also rise sharply, potentially exceeding the value of the task itself, resulting in resource waste. This shows that existing methods are insufficient to effectively cover high-difficulty tasks while controlling costs.

[0007] In summary, current technologies lack a comprehensive solution that can protect the location privacy of both the task provider and the worker, effectively incentivize workers to perform challenging tasks, and simultaneously control incentive costs. Therefore, achieving a cost-effective task coverage incentive mechanism while protecting bilateral privacy has become a pressing issue in mobile crowdsourcing systems. Summary of the Invention

[0008] This invention provides a privacy-preserving task incentive and coverage method based on a dual-cloud architecture, aiming to solve a key problem that has not yet been properly addressed in mobile crowdsourcing systems: how to achieve efficient task coverage and a reasonable incentive mechanism design under the premise that both task location and worker location must be kept confidential and not known to third-party platforms.

[0009] The technical solution of this invention includes the following two core modules:

[0010] Firstly, there is a privacy-preserving distance calculation mechanism. This invention is based on a dual-cloud architecture, with the platform consisting of two independent cloud servers, denoted as Server SA and Server SB. The worker's equipment undergoes a predefined perturbation decomposition process on its actual location, and the two perturbated location data parts are sent to Server SA and Server SB respectively. The task requester performs the same perturbation decomposition process on its task location and sends the data to both servers. Server SA and Server SB collaboratively perform Euclidean distance calculation based on a garbled circuit protocol. The perturbation scheme is designed to automatically cancel noise during the distance calculation process, ensuring that the obtained Euclidean distance result is consistent with the plaintext data calculation.

[0011] Second, a task incentive mechanism based on the goal gradient effect. Behavioral economics research has found that an individual's motivation level significantly increases as they get closer to a goal; this phenomenon is known as the goal gradient effect. The platform sets phased completion goals for workers on challenging tasks, and as the task completion progresses closer to the goal, it guides workers to continuously complete more challenging tasks to obtain the final reward by controlling the rate of progress increase for each task. This design fully utilizes the psychological principle of "the closer the goal, the stronger the motivation" in the goal gradient effect, avoiding one-time high incentives and reducing the average task coverage cost.

[0012] The overall mechanism execution flow is as follows: ① Location Noise Addition: Workers and requesters respectively perturb their own locations and split them into two parts, which are then uploaded to two independent cloud servers. ② Privacy-Preserving Distance Calculation: The two cloud servers accurately calculate the Euclidean distance between each pair of workers and tasks based on the obfuscated circuit protocol. ③ Task Difficulty Assessment and Classification: Based on the calculated worker-task distance, the platform estimates the coverage difficulty of each task and classifies tasks into high-difficulty and low-difficulty categories. ④ Reward Mechanism Release: The platform announces the additional reward rules for high-difficulty tasks to all workers, incentivizing them to continuously complete such tasks to obtain cumulative rewards. ⑤ Worker Selection Motivation Estimation and Task Allocation: Based on each worker's current progress in completing high-difficulty tasks, the platform estimates their selection motivation due to the goal gradient effect, calculates the minimum incentive reward required for task execution, and allocates workers to tasks according to incentive costs from low to high. ⑥ Task and Reward Push: The platform pushes task details and corresponding reward information to the selected workers. ⑦ Task Execution and Data Reporting. Workers decide whether to accept a task based on the task and reward information; after acceptance, they go to the designated location to collect sensory data and upload the results to the platform. ⑧ Remuneration and Progress Update. The platform distributes corresponding remuneration to workers who complete the task. For workers who perform high-difficulty tasks, the platform updates their cumulative progress. ⑨ Data Aggregation and Result Return. The platform aggregates and analyzes the sensory data submitted by multiple workers, generates the final task result, and returns it to the requester.

[0013] The method provided by this invention can accurately calculate the distance between a task and a worker while protecting the privacy of both task and worker locations, and design a reasonable incentive strategy based on the difficulty of task coverage, thereby achieving efficient task coverage and incentive allocation. Compared with existing technologies, this invention has the following advantages: First, it achieves bilateral privacy protection for both task and worker locations, preventing platforms or other third parties from maliciously inferring the user's real location, thus improving system security and user participation; second, it ensures the accuracy of the worker-task distance calculation upon which task allocation depends, avoiding the impact of noise on matching quality; third, it designs a behavior-driven incentive mechanism, effectively improving the coverage of difficult tasks without significantly increasing costs; and fourth, it can operate in a distributed environment without relying on a trusted center, adapting to the crowdsourcing task allocation needs in edge computing.

[0014] The method of this invention is widely applicable to mobile crowdsourcing platforms that require the protection of user and task location information, and is especially suitable for location-sensitive application scenarios involving public space perception, urban management, etc., and has good practicality and promotion value. Attached Figure Description

[0015] Figure 1 This is a schematic diagram illustrating the operating principle of the model in an embodiment of the present invention.

[0016] Figure 2 This is a comparison chart of task coverage using the method of this invention and the general method in this embodiment of the invention.

[0017] Figure 3 This is a comparison chart of the average task coverage cost using the method of this invention and the general method in this embodiment of the invention.

[0018] Figure 4 This is a comparison chart of task coverage using the method of the present invention with different numbers of workers in an embodiment of the present invention.

[0019] Figure 5 This is a comparison chart of the average task coverage cost under different numbers of workers using the method of the present invention in this embodiment of the invention. Detailed Implementation

[0020] The specific steps of the embodiments of the present invention are described below with reference to the accompanying drawings. Figure 1 As shown, the detailed interaction process between the worker, the requester, and the platform employing a dual-cloud architecture, as involved in this invention, is as follows.

[0021] Step 1: Add noise to the worker's and task's positions. (Worker w) i Randomly generate decomposition coefficients α i ,β i ∈(0,1), place it in position L i =(x i ,y i The linear decomposition consists of two parts:

[0022]

[0023] The task requester requests task t k Position L k =(x k ,y k Similarly, the decomposition coefficients α are randomly generated. k ,β k Decompose:

[0024]

[0025] Subsequently, the workers will (x) i,A ,y i,A (x) is sent to server SA. i,B ,y i,B ) is sent to SB; the task requester will send (x) k,A ,y k,A )Sent to SA, (x k,B ,y k,B Send to SB.

[0026] Step 2, worker-task distance calculation. Assume server SA is the obfuscator, holding the location pair (x... i,A ,y i,A ),(x k,A ,y k,A Server SB is the computing party, holding (x) i,B ,y i,B ),(x k,B ,y k,B First, SA constructs a Boolean circuit to represent the objective function:

[0027]

[0028] SA obfuscates the logic gates in the circuit, generating an obfuscated circuit. And input i A =(x i,A ,y i,A ,x k,A ,y k,A Encoding in tag form Together It is sent together to SB. SB obtains its input I from SA via an unintentional transfer protocol. B =(x i,B ,y i,B ,x k,B ,y k,B ) tag SB then used and For confusion circuits Perform door-to-door decryption to obtain the distance d i,k .

[0029] Step 3: Task coverage difficulty assessment and classification. The platform acquires workers and calculates the distance between each task (t). k Coverage difficulty φ k The calculation formula is:

[0030]

[0031] Where e is the natural constant. For all workers, calculate the classification threshold φ. th The coverage difficulty φ is the average of the coverage difficulty for each task. k Exceeding the threshold φ th The tasks are divided into high-difficulty tasks, forming a set.

[0032] Step four, release the reward mechanism. A special reward mechanism will be set up for high-difficulty tasks. The platform will provide a reward to each worker. i Record its cumulative progress g in performing high-difficulty tasks iFor each highly difficult task a worker performs, their cumulative progress g i Both will increase. If the worker can make g before the deadline τ i Achieve target g th In this case, a large reward of R can be obtained in one go. The platform announced the above reward rules to the workers and also made public a set of high-difficulty tasks. Large reward amount R, progress target g th The current progress g of each worker i And the current time l. Attracted by the large reward R, workers will generate interest in target g. th The pursuit of goals motivates workers to choose more challenging tasks. Furthermore, according to the goal gradient effect, the closer a worker's current progress is to the goal, and the closer the current time is to the deadline, the stronger their motivation to choose and execute more challenging tasks.

[0033] Step 5: Workers select power estimation and task allocation. The platform calculates the power estimation and task allocation for each worker based on their individual performance. i Current progress g i Based on the remaining time between the current time l and the deadline τ, estimate worker w according to formula (5). i For task t k Choice of power

[0034]

[0035] Where the parameter η i Describes worker w i The degree of desire for large rewards R. At the same time, the platform determines workers' motivation based on their desire for large rewards R. i With task t k The distance d between i,k This allows us to estimate the minimum incentive r required for workers to agree to perform the task. i,k Workers' need for incentives is influenced by their psychological expectation threshold regarding the distance they need to travel. Unit movement cost μ i and sensitivity λ to additional travel distance i The combined effects are calculated as follows:

[0036]

[0037] Subsequently, according to r i,k and Platform computing should enable workers w i Accept the task t k Minimum remuneration p required i,k ,

[0038]

[0039] The platform assigns workers to tasks based on the required compensation, from lowest to highest. The platform then assigns tasks to workers. k The set of workers assigned is denoted as

[0040] Step Six: Task and Reward Assignment. The platform informs the workers of the task allocation results; that is, for each task t... k The platform will t k Information and corresponding compensation information p i,k Push to collection each worker w i .

[0041] Step 7: Task Execution and Data Reporting. For each worker w i Given the platform's known cumulative progress reward rules for high-difficulty tasks and the task execution time (t), k Expected reward p i,k After receiving the information, decide whether to accept and execute the task. k The actual task t will be executed. k The set of workers is denoted as Workers in the process of completing the task k It collects sensing data at designated locations and reports the sensing results to the platform.

[0042] Step 8: Reward Distribution and Progress Updates. The platform distributes rewards and updates to those who have completed the task. k workers w i Pay the corresponding remuneration p i,k Meanwhile, according to the cumulative progress reward rules, for workers who have performed high-difficulty tasks, the platform updates their cumulative progress using the following formula:

[0043]

[0044] in It is a worker w i The cumulative progress at the current moment l, This represents the updated cumulative progress. th Let τ be the cumulative progress target that the workers need to achieve, and τ be the length of the reward period. This represents the increment size for a single growth step. Represents task t k It belongs to a collection of high-difficulty tasks Indicator parameter o i,k =1 represents worker w i Task t was executed k Otherwise o i,k=0. The reward step size function f controls the rate of increase of the cumulative progress, and the worker's current cumulative progress. The closer to the target g th The smaller the progress increment a worker gains from performing the same task, the less motivated they are to continue performing more challenging tasks due to the sunk cost effect, even though the rate of cumulative progress growth slows down. Workers cannot abandon their near-achieved goals. If the current time *l* has reached the reward cycle's deadline *τ*, the platform checks each worker's cumulative progress *g*. i Has the target value g been reached? th And a large bonus R is given to workers who meet the standards.

[0045] Step nine, data aggregation and result return. For each task t k The platform's set of executors Chinese workers w i The submitted perception data is aggregated to generate a truth inference for the task, and the task result is returned to the corresponding task requester.

[0046] In summary, this invention proposes a privacy-preserving task coverage and incentive method based on a dual-cloud architecture suitable for mobile crowdsourcing systems. Addressing the problems of privacy leakage, insufficient coverage of high-difficulty tasks, and excessively high incentive costs in existing solutions, it constructs an incentive scheme based on obfuscated circuits and target gradient effects. Compared with general methods, this invention effectively improves task coverage, reduces incentive costs, and dynamically adapts to different task coverage requirements while ensuring the non-disclosure of worker and task location information. Based on experimental results, the advantages of this invention are summarized as follows:

[0047] (1) It has a higher task coverage capability. Figure 2 The experiment presents comparative results of task coverage using the method of this invention and traditional methods. In the experiment, the traditional method uses a strategy based on a linear incentive model, while the present invention utilizes the goal gradient psychological effect from behavioral economics, which can more effectively motivate workers to participate in high-difficulty tasks. Figure 2 As can be seen, in the 30 comparative experiments conducted, the method of the present invention is significantly better than the traditional method in terms of task coverage, with a minimum improvement of 6.31% and a maximum improvement of 14.25%.

[0048] (2) It can significantly reduce the average task coverage cost. For example... Figure 3 The diagram illustrates a comparison between the method of this invention and conventional methods in terms of average task coverage cost. Because this invention employs a phased, progressive incentive strategy, workers need to complete a large number of highly challenging tasks to receive rewards, thus significantly reducing the average incentive cost per task while maintaining coverage effectiveness. Compared to traditional methods, this invention reduces average coverage cost by 15.17%.

[0049] (3) It has good scalability and adaptability. Figure 4 and Figure 5 The trends of task coverage rate and coverage cost under different numbers of workers are shown respectively. The results show that as the number of workers in the system increases, the method of the present invention can make full use of the distribution advantages of the location and price of the new workers to achieve higher coverage rate and lower incentive cost, demonstrating good scalability. At the same time, the task coverage incentive mechanism in the present invention can be flexibly adjusted according to the urgency and scarcity of tasks (for example, by adding corresponding parameters to the task coverage difficulty calculation formula (4)), so as to better adapt to diverse practical application needs.

Claims

1. A method for privacy-preserving task coverage and incentives based on a dual-cloud architecture in mobile crowdsourcing, characterized in that, Includes the following steps: Step 1: The worker and the task requester respectively add random noise and perform linear decomposition on their respective location information, and send the decomposed coordinates to two non-cooperative servers respectively. Step two: The two servers collaboratively calculate the Euclidean distance between the worker and the task using a confused circuit protocol to prevent the location information from being leaked. Step 3: The platform calculates the coverage difficulty of the task based on the worker-task distance and classifies tasks with high coverage difficulty into the high-difficulty task set; Step 4: The platform announces the set of challenging tasks and their corresponding target progress reward mechanism to all workers, including the reward amount, progress target, current progress and deadline; Step 5: The platform estimates the worker's motivation to choose a challenging task based on the worker's current progress, time information, and incentive preferences, and calculates the minimum incentive and estimated reward required to complete the task based on the behavioral model. Step 6: The platform assigns workers to tasks based on the reward ranking from low to high, and pushes the task and reward information to the assigned workers. Step 7: Workers decide whether to perform tasks based on their own compensation and progress rules, and workers who perform tasks send the perception data back to the platform. Step 8: The platform pays the workers who have completed the tasks and updates their cumulative progress on high-difficulty tasks based on their performance. Step nine: The platform aggregates and processes the perception data for each task, generates task results, and feeds them back to the requester.

2. The method according to claim 1, characterized in that: The location noise addition and decomposition described in step one specifically refers to: worker w i Randomly generate decomposition coefficients α i With β i The position coordinates (x i ,y i According to the following formula, decompose into (x) i,A ,y i,A ) and (x i,B ,y i,B ), respectively sent to servers SA and SB, Regarding task t, the requester of the task... k The decomposition coefficients α are also randomly generated. k With β k For task coordinates (x) k ,y k The same method is used to decompose and send data to servers SA and SB.

3. The method according to claim 1, characterized in that: The worker-task distance calculation described in step two is accomplished in the following way: The server SA constructs the objective function circuit: And generate a confusion circuit The server (SB) decrypts the obfuscated circuit using the input tag to calculate the distance.

4. The method according to claim 1, characterized in that: Step 3, task coverage difficulty φ k The calculation method is as follows:

5. The method according to claim 1, characterized in that: In step five, the worker selects the power source for their task. Calculate using the following formula.

6. The method according to claim 1, characterized in that: In step five, the minimum incentive r required for the worker to perform the task. i,k Calculate using the following formula. in This represents the worker's psychological expectation threshold for the distance they need to travel, μ i For unit movement cost, λ i This indicates a worker's sensitivity to additional distance traveled.

7. The method according to claim 1, characterized in that: In step five, the platform estimates the return p. i,k Calculated as follows 8. The method according to claim 1, characterized in that: Accumulated progress g in step eight i The update formula is: Among them o i,k For workers w i Execute task t k The indicator variable, g th Let τ be the schedule target, τ be the deadline, and γ1 and γ2 be parameters that control the progress increment.

Images