Stackelberg dynamic heterogeneous task allocation method and system for crowd sensing

Abstract

The invention relates to the technical field of task allocation, in particular to a Stackelberg dynamic heterogeneous task allocation method and system oriented to crowd sensing. The method comprises the steps of obtaining task data of a crowdsourcing task, requester historical data and worker historical data; calculating a task allocation strategy index according to the task data, the requester historical data and the task historical data; balancing task allocation strategy indexes based on an improved fireworks algorithm; performing user task response distribution prediction according to the task distribution strategy index; and obtaining platform operation benefits based on user task response distribution prediction. Under the Stackelberg game model, super-multi-objective optimization of worker profit, requester cost, platform profit and QoS is realized, the worker profit is guaranteed on the premise of taking the platform profit as a first starting point, and meanwhile, the participation rate of the requester for the use task of the platform is also guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of task allocation, and in particular to a Stackelberg dynamic heterogeneous task allocation method and system for crowd intelligence perception. Background Art

[0002] With the widespread adoption of mobile devices and wireless communication technologies, mobile crowdsourcing (MCS), a data collection and processing model based on mass intelligent devices, has rapidly developed in recent years in a variety of fields, including traffic management, environmental monitoring, disaster response, and healthcare. By incentivizing individual users to participate in sensing tasks, MCS achieves efficient resource utilization and intelligent service deployment, possessing significant economic and social value. However, despite significant progress in both theory and application, current technologies still have numerous shortcomings in dynamically allocating heterogeneous tasks, hindering their further development and widespread adoption.

[0003] First, existing task allocation strategies generally fail to fully consider the heterogeneity of tasks and the dynamic changes in the environment. In actual MCS applications, task types vary significantly, their attributes are complex and diverse, and different tasks have different requirements for dimensions such as time, accuracy, and cost. However, traditional methods often assume that task attributes are uniform, ignoring the differences in scheduling of diverse tasks, resulting in allocation results that fail to meet actual needs. Furthermore, MCS environments are highly dynamic, with tasks and worker status constantly changing. Existing static or semi-dynamic strategies struggle to adapt to these changes in real time, and are prone to problems such as wasted resources and reduced efficiency.

[0004] Second, existing optimization models struggle to reconcile the conflicting interests of the platform, workers, and requesters. In MCS systems, the platform focuses on maximizing profits, requesters seek to complete tasks at the lowest cost, and workers aim for the highest rewards. Current optimization algorithms, such as genetic algorithms and particle swarm optimization, often prioritize the goals of one party and fail to construct effective game models to balance these multiple parties, resulting in a decline in task participation rates.

[0005] Furthermore, existing multi-objective optimization algorithms generally suffer from slow convergence, becoming trapped in local optima, and poor adaptability. Traditional algorithms exhibit significant limitations, particularly when dealing with high-dimensional, multi-objective, and dynamically changing problems like MCS. Their complex structures and parameter sensitivity make it difficult to achieve a global optimal solution within a limited timeframe, thus failing to meet the efficiency requirements of real-time task scheduling. Summary of the Invention

[0006] In order to solve the above-mentioned problems, the present invention provides a Stackelberg dynamic heterogeneous task allocation method and system for crowd intelligence perception.

[0007] In a first aspect, the present invention provides a Stackelberg dynamic heterogeneous task allocation method for crowd intelligence perception, which adopts the following technical solutions: A Stackelberg dynamic heterogeneous task allocation method for crowd intelligence perception includes: Obtain task data, requester history data, and worker history data for crowdsourcing tasks; Calculate task allocation strategy indicators based on task data, requester history data, and task history data; Balance task allocation strategy indicators based on improved fireworks algorithm; Predict user task response allocation based on task allocation strategy indicators; Get the task response assignment result.

[0008] Furthermore, the task allocation strategy indicators are calculated based on the task data, the requester's historical data, and the worker's historical data, including calculating the matching degree between the task and the worker based on the task data and the worker's historical data, wherein the project requirement coverage is calculated based on the worker's skill set and the project requirement set, and the matching degree between the worker and the task's geographic location is calculated using the Manhattan distance. For the worker's geographic location Lw = (xw, yw), the task's geographic location Lt = (xt, yt), and the Manhattan distance between the worker and the task are expressed as: d(Lw, Lt) = |xw-xt|+|yw-yt|, and the worker's historical data is evaluated, and the worker's historical performance is used as an important basis for the platform to assign tasks. Let , workers' history is manifested as: , and calculate the worker reputation value. Let Q represent the quality of task completion, g represent the worker's praise rate, that is, the proportion of positive evaluations. The worker reputation value is expressed as: .

[0009] Furthermore, the calculation of the task allocation strategy indicator based on the task data, the requester's historical data, and the worker's historical data also includes calculating the worker's task reward coefficient based on the task data and the worker's historical data, wherein the worker's reputation value is calculated based on the task completion quality and the worker's historical positive evaluation data; the worker's behavior response rate is calculated based on the task data and the worker's historical performance data; and the worker's reward coefficient is determined based on the worker's reputation value and the worker's behavior response rate. The reward coefficient is expressed as: , Here, Δ is a constant.

[0010] Furthermore, the calculation of the task allocation strategy indicator based on the task data, the requester's historical data, and the worker's historical data further includes calculating the requester's reputation value based on the task data and the requester's historical data, wherein the difficulty of the tasks published and completed by the requester is divided and different coefficients are assigned based on the task data to calculate the requester's experience value; and the requester's reputation value is calculated based on the requester's historical data on the platform, wherein PO represents the number of task rewards that the requester did not pay in time after the worker completed the task normally, PO1 represents the number of payments beyond the deadline, and PO0 represents the number of payments that were never made. Then the requester's reputation value is expressed as: .

[0011] Furthermore, the improved fireworks algorithm based on balancing the task allocation strategy indicators includes maximizing the matching degree between worker skills and tasks as the goal, using the value of the matching degree between worker skills and tasks as the fitness indicator, using a greedy algorithm to screen and calculate the average fitness, and obtaining the initial fireworks population and optimal weight through iteration. By setting the maximum and minimum amplitude of the fireworks explosion and the maximum and minimum number of sparks generated by the fireworks explosion, the formula for calculating the amplitude and number of explosion sparks of each firework is: A = A_min + (A_max - A_min) * (M - fmin) / (fmax - fmin) N_sparks = N_sparks_min + (N_sparks_max - N_sparks_min) * (M - fmin) / (fmax - fmin) The spark produced by the explosion is represented as Fij, where i represents the i-th firework and j represents the j-th spark produced by the explosion of the i-th firework, j∈[1, N_sparks]. Spark mutation is set to obey Gaussian distribution, and all data in the spark candidate pool are updated to the mutated sparks. The sparks in the candidate pool are selected and iterated. The final Fire obtained after the iteration is terminated is a set of optimal solutions for the matching degree between worker skills and tasks.

[0012] Furthermore, the prediction of user task response allocation based on the task allocation strategy indicator includes calculating the platform reputation value based on the worker historical data and the requester historical data, wherein the platform reputation value is calculated based on the worker's rating of the platform and the requester's rating of the platform, and the requester's risk perception coefficient is calculated based on the platform reputation value and the worker's reputation value, which is expressed as: , Where Rrisk is the risk perception coefficient of the requester, k1, k2 and k3 are adjustment parameters; the worker risk perception coefficient is then calculated using the platform reputation value and the requester reputation value, expressed as: , Furthermore, the user task response allocation prediction based on the task allocation strategy indicator also includes calculating the requester behavior response function based on the task allocation strategy indicator, which is expressed as: , in Indicates the urgency of the task. Represents the requester budget.

[0013] Furthermore, the user task response allocation prediction based on the task allocation strategy indicator also includes calculating the worker behavior response function based on the task allocation strategy indicator, which is expressed as: , in represents the worker skill and task adaptation ratio obtained by strategy 1, represents the worker's reward coefficient obtained in strategy 2, Similar to the requester behavior response function, Indicates the number of grids across which workers execute tasks.

[0014] Furthermore, the user task response allocation prediction based on the task allocation strategy indicator also includes predicting the optimal solution of the task allocation strategy based on the worker behavior response function and the requester behavior response function, and verifying the effect of the task allocation strategy using the requester risk perception coefficient and the worker risk perception coefficient.

[0015] The second aspect is a Stackelberg dynamic heterogeneous task allocation system for crowd intelligence perception, including: A data acquisition module is configured to acquire task data, requester history data, and worker history data of a crowdsourcing task; A strategy indicator module is configured to calculate a task allocation strategy indicator based on the task data, the requester history data, and the task history data; The balancing module is configured to balance the task allocation strategy indicators based on the improved fireworks algorithm; The prediction module is configured to predict the user task response allocation based on the task allocation strategy indicator; The allocation module is configured to obtain a task response allocation result.

[0016] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the method described.

[0017] In a fourth aspect, the present invention provides a terminal device comprising a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; and the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded by the processor and executing the described method.

[0018] In summary, the present invention has the following beneficial technical effects: 1. Maximizing the interests of multiple parties: Under the Stackelberg game model, we achieve multi-objective optimization of worker profits, requester costs, platform profits, and QoS. While prioritizing platform interests, we ensure worker profits while also guaranteeing the requester participation rate in platform tasks.

[0019] 2. Improving the rationality of platform strategies: This invention predicts the needs of workers and requesters in advance and optimizes conflicting objectives within the platform's strategies using an improved Fireworks algorithm, enabling the platform to achieve the optimal strategy while satisfying the needs of all three parties. Experiments have demonstrated that the improved Fireworks algorithm exhibits excellent stability and optimization performance in dynamic environments, particularly in scenarios with multiple optimization objectives.

[0020] 3. Effectively Guaranteed User Engagement: This paper conducted three rounds of simulation experiments using three datasets. The results showed that the model ultimately yielded an average probability of 89.14% for workers to accept tasks, 73.76% for requesters to publish tasks, and 39.5% for workers to cross-grid tasks. This demonstrates the positive impact of the dynamic reward mechanism on behavioral responses, while also effectively allocating tasks based on user preferences, thereby ensuring user engagement and improving platform usage.

[0021] 4. Effectively predict user behavior: By incorporating three theories of user personalized needs into a game theory model, this paper conducts an in-depth analysis of the impact of user historical behavior on decision-making and minimizes negative impacts. As shown by the average user usage probability, integrating the behavioral preference constraint model into the game theory framework can effectively improve user behavioral response rates and task participation rates. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] picture Figure 1 This is the MCS system operation architecture diagram of the present invention.

[0023] Figure 2 It is a flow chart of the Stackelberg game model of the present invention.

[0024] Figure 3 It is a grid division diagram for workers to perform tasks across grids according to the present invention.

[0025] Figure 4This is a comparison chart of the optimization effects of the platform strategy optimized by the present invention after improving the fireworks algorithm and other algorithms. Figure 4 (a) shows the comparison of the experimental results of worker credibility. Figure 4 (b) in the figure shows the comparison of the experimental results of worker skills and task matching.

[0026] Figure 5 This is a bar chart showing the impact of the dynamic reward mechanism and behavior preference constraint mechanism on task participation rate. Figure 5 (a) in the equation represents the requester’s positive selection probability under different conditions. Figure 5 (b) in the figure represents the positive selection probability of workers under different conditions.

[0027] Figure 6 It is a comparative analysis of the task market pricing obtained by the present invention and the market base price.

[0028] Figure 7 This is a comparative analysis of the algorithm optimization effects in three rounds of experiments simulating dynamic environments. Figure 7 (a) in the figure shows the comparison of the platform’s optimization effect on workers’ evaluation. Figure 7 (b) shows the comparison of task pricing optimization effects. Figure 7 (c) in the figure shows the comparison of the optimization effects of workers’ enthusiasm.

[0029] Figure 8 This is a schematic diagram of the calculation of the number of cross-grids in the present invention. DETAILED DESCRIPTION

[0030] The present invention will be further described in detail below with reference to the accompanying drawings.

[0031] Example 1 Reference Figure 1 , a Stackelberg dynamic heterogeneous task allocation method for crowd intelligence perception in this embodiment includes: Obtain task data, requester history data, and worker history data for crowdsourcing tasks; Calculate task allocation strategy indicators based on task data, requester history data, and task history data; Balance task allocation strategy indicators based on improved fireworks algorithm; Predict user task response allocation based on task allocation strategy indicators; Get the task response assignment result.

[0032] Specifically: This paper uses five real-world datasets: ① The Uber New York City ride dataset, containing approximately 18.8 million ride records and rental company data from 2014-2015; ② The Nice Ride shared bike dataset, containing attributes such as ride count and ride duration; ③ The credit card fraud dataset, containing 284,807 European credit card transactions, 492 fraudulent transactions, and PCA transformation; ④ The Shenzhen Stock Exchange Main Board daily chart dataset, containing three types of price data for 466 stocks; and ⑤ The US health insurance market dataset, covering information such as insurance coverage, types, and rates.

[0033] The first aspect of the present invention proposes a Stackelberg two-stage game model for the platform, workers and requesters, so as to maximize the interests of all three parties.

[0034] The first stage of the game: The platform formulates strategies based on the needs of workers and requesters. The platform first formulates corresponding rules based on the needs of the three parties and preliminarily finds the optimal platform strategy solution that maximizes platform benefits.

[0035] S1. The relevant indicators required for the platform to formulate strategies for workers are as follows: S1.1 Indicators related to matching workers with tasks. When selecting suitable workers, the platform primarily considers their skills, location, historical performance, and reputation. The higher the quality of these indicators, the more suitable the worker is for the task. The indicator calculation process is as follows: S1.1.1 Calculate the matching degree between worker skills and tasks. Assume that the worker skill set S = {s1, s2…sm} and the project requirement set N = {n1, n2…nk}, where m is the number of workers and k is the number of tasks. Therefore, the project requirement coverage is expressed as: (1) S1.1.2 Use Manhattan distance to calculate the matching degree between workers and task locations. Assume that the worker's location is =( , ), the mission's geographical location =( , ), then the Manhattan distance between workers and tasks is expressed as: (2) S1.1.3 Evaluate workers’ historical data. In order to improve the perceived quality of the worker group, it is necessary to use workers’ historical performance as an important basis for the platform to assign tasks. , so the history is: (3) At this point, we've determined the worker's skill-task match, the worker's geographic location match, and the worker's historical data. We perform a weighted analysis on these three pieces of data and optimize them using our improved Fireworks algorithm (explained later) to obtain the optimal worker-task match, i.e., the optimal triplet solution.

[0036] S1.2 Calculate the worker's reputation. Let Q represent the quality of task completion (Q∈[1,10]), and g represent the worker's positive rating, i.e., the percentage of positive reviews. The worker's reputation is expressed as: (4) S1.3 Indicators required for the platform’s incentive strategy for workers, including: S1.3.1 To ensure the selection of high-quality workers, the platform evaluates the comprehensive performance of workers. The platform's evaluation formula for workers is: (5) One of the key requirements is Must be strictly less than or equal to This means that tasks that are not completed on time will not be included in the evaluation item to ensure the accuracy and fairness of the incentive mechanism.

[0037] S1.3.2 Considering the worker's motivation to perform the task, the worker's behavioral response rate is quantified based on task pricing and historical performance data. The worker's behavioral response rate is expressed as: (6) Among them, P represents the platform's pricing strategy, and its specific decision-making process will be elaborated in detail in subsequent chapters.

[0038] Based on the above evaluation and quantification, the worker reward coefficient is determined to represent the additional reward index that the worker can obtain for the next task. The reward coefficient can be expressed as: (7) Where Δ is a constant, and workers can get more rewards when the number of tasks they complete is greater than the average level.

[0039] After weighting the two indicators and optimizing the algorithm, we can obtain the optimal worker reward coefficient. This coefficient is used to calculate the additional rewards that workers can receive in addition to their task remuneration.

[0040] S2. The relevant indicators required by the platform to formulate strategies for requesters are as follows: S2.1 Platform pricing coefficient for tasks, including: S2.1.1 Calculate the task complexity based on the task difficulty and task type. Let Th be the task difficulty level [1,10], Tl be the task technical level requirement type [1,10], then the task complexity is expressed as: (8) S2.1.2 Quantify the urgency of the task. : Time required for the task, : remaining time of the task, : Task release time, I: Task importance (I∈[1,10]). Therefore, the urgency is expressed as: (9) Among them ≠ hour, Indicates the ratio of the remaining time of the task to the total time. The reduction of The increase, and The value of will increase, indicating that the urgency of the task is increasing. = When the task is released on the same day, the urgency of the task is mainly affected by and the impact of I. The larger I is, the lower the urgency of the task is; the larger I is, the greater the importance of the task is, and the higher the urgency of the task is.

[0041] The above two indicators are weighted and optimized using the algorithm of the present invention to obtain the optimal pricing coefficient, which is used to calculate the task pricing in different scenarios.

[0042] S2.2 Indicators required by the platform’s incentive strategy for requesters. Including: S2.2.1 Calculate the requester's experience value by dividing the difficulty of the tasks that the requester has posted and completed and assigning different coefficients. Let the number of tasks that the requester has posted and completed be PF. In order to prevent the requester from blindly pursuing to increase the experience value and frequently posting easy tasks, the present invention will calculate the task by dividing it into two categories according to the difficulty of the task. The number of tasks with a difficulty between [1,5] is , the number of tasks with a difficulty of [6,10] is , the present invention assumes that the impact index is 0.32 and 0.68 respectively. Then the requester's experience value is: (10) S2.2.2 Calculate the requester's reputation value based on the requester's historical data on the platform (requesters with negative reputation values ​​cannot participate in the incentive mechanism). Let PO represent the number of task rewards that the requester has not paid in time after the worker has completed the task normally. For amounts paid beyond the deadline, is the amount that has never been paid. Then the requester's reputation value is expressed as: (11) Based on the above data, the requester reward coefficient is determined, which is used to represent the preferential index that the requester can obtain when publishing the task next.

[0043] All of the aforementioned algorithm optimization tasks utilize the Fireworks algorithm to optimize the relevant metrics. This paper presents improvements to this algorithm, which will be described later. The optimization objective prioritizes platform benefits, achieving the optimal solution that initially maximizes platform benefits.

[0044] The second stage of the game: Based on the optimal solutions for the platform's relevant indicators obtained above, the second stage of the game begins, namely the decision-making stage of workers and requesters.

[0045] In order to better reflect how to give reasonable pricing for tasks under different conditions, this paper uses a payment matrix to reflect the comparison of task pricing under various conditions. Taking one task type as an example, assume that its market base price is 10. Combined with the pricing coefficients obtained in the first stage, the payment matrix C is obtained: , The second aspect of the present invention constructs three sets of user historical behavior characteristic parameters, establishes a user behavior preference constraint model, and minimizes the impact of users' negative decision-making factors. The first is a risk quantification assessment model, which is used to analyze the impact of the platform's potential risks on user behavior responses. The smooth execution of tasks is inseparable from the comprehensive quality of workers and the reasonable operation of the platform. Therefore, the present invention sets up a platform reputation value evaluation mechanism to more accurately reflect the platform's ability to protect user interests. The platform reputation value calculation formula is: (12) in Indicates the workers’ rating of the platform, Indicates the rating given by the requester to the platform, represents the number of workers, Indicates the number of requesters.

[0046] The task allocation mechanism model, worker reputation value, and platform reputation value are combined to construct a risk perception model. The formula for the requester's risk perception coefficient is expressed as: (13) in It is the risk perception coefficient of the requester, whose value is between 0 and 1, indicating the requester's perception of risk, that is, the probability that the requester will give up the upload task this time. 、 and It is an adjustment parameter used to control the influence of each factor on the risk perception coefficient.

[0047] This formula uses the Sigmoid function ( ), which maps the input value to between 0 and 1, thus achieving a smooth threshold effect. The Sigmoid function is used here to convert the raw score of each factor into a fraction of the risk perception coefficient. Therefore, the formula can be expressed as: (14) In this formula, each Sigmoid function independently processes a factor and multiplies the result by its corresponding weight coefficient. In this way, each factor jointly affects the risk perception coefficient through the Sigmoid function and the weight coefficient.

[0048] Get the average risk perception coefficient , which is used to predict how many requesters out of every 100 requesters may abandon the upload task.

[0049] The worker risk perception coefficient is evaluated based on the platform reputation value and the requester reputation value, which is expressed as: (15) Similar to the requester risk perception coefficient, the formula can be rewritten as: , The risk perception coefficient is used to verify whether the decision results of workers and requesters meet the platform's expectations after the platform makes a profit maximization decision in the first stage, thereby verifying the feasibility of the model of the present invention.

[0050] in Indicates the reputation value of the requester.

[0051] Based on the indicators obtained in the first phase, a user behavior response function is constructed. In addition to being influenced by the risk perception coefficient and task urgency, the requester behavior response function is also related to expected cost. This is the second user personalized demand theory introduced in this paper—the expected utility theory. If the task price exceeds the requester's budget, the requester will assess the value of the excess. If increasing the budget can bring satisfactory utility, the requester may choose to increase the budget. Therefore, the requester behavior response function is expressed as: (16) in Indicates the urgency of the task. Represents the requester's budget. Indicates the budget limit of the requester. When the task price exceeds the requester's budget, the requester will give up publishing the task. The value is 0, otherwise it is 1.

[0052] Based on the worker-related indicators obtained above, a worker behavior response function is constructed. The function is expressed as: (17) in represents the worker skill and task adaptation ratio obtained by strategy 1, represents the worker's reward coefficient obtained in strategy 2, Similar to the requester behavior response function, if the reward is lower than the worker's expectation, then The value is 0, otherwise it is 1. Indicates the number of grids across which workers execute tasks, which is obtained based on the Manhattan distance in strategy 1.

[0053] It can be obtained by Manhattan distance The formula is: (18) in Represents the reward weight for workers crossing grids. The more grids a worker crosses, the more additional rewards he or she will receive.

[0054] After obtaining the behavioral response functions for requesters and workers, we can predict their decision outcomes, effectively obtaining the optimal strategy for each requester and worker. The previously calculated behavioral response functions are used to verify the ultimate effectiveness of our model—that is, prioritizing platform profit maximization without significantly influencing the decisions of workers and requesters.

[0055] At this point, the Nash equilibrium solution for the platform, requesters, and workers has been obtained. Therefore, we can construct a tripartite benefit function to obtain the platform profit, worker profit, and requester cost output by the model. The tripartite benefit function is expressed as: (19) in Indicates the additional reward a worker receives for being a team leader. This invention proposes the assignment of a team leader to coordinate project work. Team leader selection is based on the reputation ranking of workers. Workers with the highest reputation are invited to be team leaders first and receive additional rewards. Additional wages paid to workers by requesters ( ∈[0.1,10], in units of $), is the reward mechanism implemented in this invention, and also the third user personalized demand theory introduced in this invention—the dynamic reward feedback mechanism. This dynamic reward feedback mechanism, designed to coordinate the behavior of requesters and workers through additional wage compensation, promotes stable cooperation. It clarifies work quality and reward expectations, reduces misunderstandings, and improves efficiency. This invention embodies this theory through additional wages, enabling both parties to flexibly respond to challenges and achieve a win-win situation in the mobile crowdsourcing environment. This is the additional reward for the worker who serves as the team leader mentioned above, which is set to 10% of the task price. This is the proportion of the total payment value of the requester that the worker can obtain, which is set to 0.6.

[0056] The tripartite profit function is used to digitize the results of the proposed model, visually displaying the final calculated results of platform and worker profits and requester costs under real-world dataset simulation. Experiments show that the three parties' interests reach a Nash equilibrium in this model scenario.

[0057] Thirdly, this paper proposes an improved fireworks algorithm—the Greedy Slot Machine Fireworks Algorithm. This algorithm is implemented in three steps. This paper uses the matching degree between worker skills and tasks as an example to demonstrate the algorithm optimization process.

[0058] first step: The goal of this invention is to maximize the match between worker skills and tasks. First, 500 candidate data sets are randomly selected, each consisting of a triplet (cov, mat, H), denoted by F. Using the match between worker skills and tasks as the fitness metric, three randomly generated weights are used to calculate the fitness of these 500 data sets. Subsequently, a greedy algorithm is used to select the 200 data sets with the highest fitness, and the average fitness of these data sets is calculated. After 1000 iterations, the iteration with the highest average fitness is found, and the 200 data sets from this iteration are added to the initial fireworks population. To prevent the algorithm from converging to a local optimal solution and ensure solution diversity, the present invention further randomly selects 100 data sets from these 500 data sets and adds them to the initial fireworks population, thereby broadening the search range for the optimal solution.

[0059] Step 2: After determining the initial population, a set of optimal weight distributions is found by simulating the random lottery mechanism of a slot machine. The number of random draws set in the present invention is 1000 times. Similar to the first stage, the optimal weight distribution is determined by comparing the average fitness of the initial population obtained by each set of weight distribution, and the maximum fitness fmax and minimum fitness fmin in the initial population under the optimal weight are calculated.

[0060] Step 3: According to the first two stages, 300 initial fireworks and three optimal weights ω1, ω2, and ω3 have been obtained. This stage enters the core link of GGMFA. First, set the maximum and minimum amplitudes of the fireworks explosion, as well as the maximum and minimum number of sparks produced by the fireworks explosion. Let the explosion amplitude be A and the number of sparks produced by the explosion be N_sparks. Therefore, the maximum and minimum amplitudes are A_max and A_min respectively, and the maximum and minimum number of explosion sparks are N_sparks_max and N_sparks_min. The initial fireworks population is represented by Fire, Fire={F1, F2...Fi}. Assume that the fitness of fireworks and sparks produced by explosions are both M values. The calculation formula for the amplitude and number of explosion sparks of each firework is: A = A_min + (A_max - A_min) * (M - fmin) / (fmax - fmin) N_sparks = N_sparks_min + (N_sparks_max - N_sparks_min) * (M - fmin) / (fmax - fmin) The sparks generated by the explosion are denoted as Fij, where i represents the i-th firework, j represents the j-th spark generated by the explosion of the i-th firework, and j∈[1, N_sparks].

[0061] The present invention sets the fireworks explosion mode to coordinate axis explosion, and the coordinates of each firework are set to (cov, mat, H). The spark mutation is set to obey Gaussian distribution with a mean of 0.5 and a standard deviation of 0.1, and the value range of each attribute in the spark after mutation is limited to [0, 1]. The spark after mutation is represented as All data in the candidate pool of sparks is updated to the mutated sparks, and the sparks in the candidate pool are selected. This method adopts an elite retention strategy, that is, 20 sparks are selected in each round, the fitness of these 20 sparks is compared with the fireworks in the Fire, and the sparks with better fitness replace the sparks in the Fire. The explosion, mutation, and selection process is repeatedly iterated. The final Fire obtained after the iteration is terminated is the set of optimal solutions for the matching of worker skills and tasks.

[0062] The above describes the experimental process for a single dataset. Considering the dynamic nature of platform data, this paper employed three different datasets to simulate real-time platform data dynamics. Three sets of optimal solutions were obtained through these three rounds of experiments. These optimal solutions were then integrated to form a comprehensive optimal solution for matching worker skills with tasks. Furthermore, the weight coefficients for certain formulas with fixed values ​​were calculated using another optimization algorithm, the bandit algorithm. To illustrate the weight calculation process, this paper uses the requester's experience value as an example.

[0063] Requester experience value is expressed as , simplified to . First, the initial weight The two are both 0.5, and then several rounds of experiments are carried out. In each round, a set of data of AB is imported to observe the change of J value. The goal of this invention is to maximize the experience value of the requester, that is, to maximize the J value. Therefore, if the change of A is higher than the recent average level, the value of J is increased. (e.g. +0.01), if the change in A is lower than the recent average, reduce (e.g. -0.01). Repeat several rounds until the weights barely change (e.g. the adjustment is less than 0.01), then stop. The weight distribution obtained at this point is the optimal distribution.

[0064] Fourthly, the present invention introduces the Quality of Service (QoS) indicator to analyze the optimal data ultimately output by the model, ensuring that a high-quality group of workers is selected to provide services to requesters, guaranteeing task execution efficiency and completion quality, and helping to fundamentally safeguard the long-term stability of platform operations.

[0065] like Figure 1 , the system operation process is: the requester first publishes the project requirements to the platform, and the platform evaluates and classifies each task in the project, and then assigns it to the appropriate worker, and the worker chooses whether to perform the task. In order to ensure the overall completion efficiency of the project, the applicable scenario of the present invention is multi-worker multi-task, that is, each project published by the platform contains multiple tasks, each task corresponds to a worker, and under the leadership of the project leader, the workers work together to complete the project, and may interact with the requester during the period. After the task is completed, the data is transmitted back to the platform, and the platform then feeds back the results to the requester. In MCS, the needs of participants are different, leading to conflicts of interest. Therefore, the present invention adopts the Stackelberg game model, positioning the platform as a leader, predicting the needs of workers and requesters, and formulating operating strategies. Workers and requesters act as followers and adjust their responses according to platform strategies. Strategy formulation is divided into two stages, such as Figure 2 This paper focuses on the heterogeneous multi-project multi-task assignment problem (HMPMTA) in MCS. This problem is NP-hard and involves the diverse attributes of projects, tasks, teams, and participants. Although existing research completes tasks through team collaboration, it limits workers to only one task in a project, ignoring the lack of worker recruitment and workers' multi-skills. In addition, it does not consider worker priority when selecting project leaders, which may result in suboptimal team leaders. In summary, this paper proposes that the assignment process meets the following conditions: 1) Each project consists of multiple heterogeneous tasks, each of which has specific skill requirements and can only be completed by workers with corresponding skills.

[0066] 2) The platform conducts two rounds of task allocation. The first round aims to maximize task completion efficiency. Each worker can only take on one task per project, but can participate in multiple projects. The second round removes this restriction, allowing workers to participate in multiple tasks within the same project.

[0067] 3) To coordinate project work, propose the appointment of a team leader. Team leaders are selected based on the reputation of the workers. Workers with the highest reputation are invited to serve as team leaders first and receive additional rewards.

[0068] Based on the above description, the present invention will explain the game process of the constructed model, such as Figure 2 As shown. The initial stage of the game is the platform decision-making link, that is, the leader makes the rules. In view of the diverse needs of the participants, the present invention tailors corresponding platform strategies for workers and requesters respectively, and optimizes each strategy using the fireworks algorithm to obtain a set of optimal decision solutions. At the end of this stage, a set of optimal solutions that maximize the platform's interests and achieve an initial balance between the interests of the three parties will be obtained. The second stage is the follower (i.e., worker and requester) decision-making stage. First, a payment matrix is ​​generated to clearly display the payment amounts obtained for various decisions of each task type. Then, the three user personalized demand theories are introduced into the follower decision-making process to obtain the follower's behavioral response function and the follower's optimal decision prediction solution. At this point, the interests of the platform, workers, and requesters have reached Nash equilibrium. According to the obtained optimal decision of the three parties, the expected profits of the platform and workers and the requester's costs can be calculated according to the benefit calculation formula. The cross-grid reward calculation method in the worker's profit is as follows Figure 3 The present invention adopts the (10*10) grid division method to more intuitively reflect the location of workers and tasks, and measures the distance between workers and tasks by Manhattan distance. In addition, if workers and tasks are in the same grid, the distance between them is negligible. If workers need to cross grids to perform tasks, additional rewards are given according to the number of grids crossed. The incentive mechanism strategy of the present invention selects 200 workers' data for cross-grid calculation, such as Figure 8 As shown. The horizontal axis represents the number of workers across the grid, that is, d ( , ) rounded down. Experiments revealed that among these 200 workers, the majority crossed between 0 and 2 grids. Calculations showed that the probability of a worker crossing a grid was 39.5%, and the average cross-grid reward received was $0.81. This demonstrates that workers have geographic preferences and that the platform maintains a balanced task allocation. The diversity in the number of grids crossed by workers indicates that the incentive mechanism improves participation and task completion efficiency, demonstrating the feasibility and effectiveness of incentive strategies in promoting optimal resource allocation in an MCS environment.

[0069] Figure 4This is a comparison chart of the experimental results of the GGMFA algorithm proposed in this invention and other algorithms. Figure 4 (a) in the figure shows the comparison of the experimental results of workers’ reputation value. Figure 4 (b) in the figure shows the comparison of the experimental results of worker skills and task matching. Figure 4 (a) and Figure 4 (b) in the figure shows that both worker reputation and the degree of match between worker skills and tasks perform well under the optimization algorithm of our invention. The final calculated average worker reputation value is 7.54 (maximum value is 10), indicating a high level of overall worker credibility on the platform. This highly reputable group of workers, through their high-quality service, increases user participation and loyalty to the service. Secondly, the average worker skill-task match value is 0.87 (maximum value is 1), indicating a high degree of fit between workers' skills and the requirements of the assigned tasks. This close match not only improves the speed and efficiency of task assignment and completion, but also indirectly ensures the quality of task completion. Figure 4 This chart shows the changing trends of various strategies under the influence of different indicators. While the overall trend is upward, some points may experience declines. This localized decline may be due to the fact that while the indicator selected on the horizontal axis has the greatest relative influence, it can still experience declines between adjacent points under the influence of other indicators, which does not affect the overall upward trend.

[0070] Figure 5 The paper compares the positive selection probabilities (i.e., the probability of a task being successfully assigned and executed) based on whether the conditions meet user needs. Experimental results show that, through a two-stage Stackelberg game and the GGMFA algorithm to optimize platform strategies, and in the second stage, predicting the different responses caused by the different needs of workers and requesters and historical user behavior, the average probability of a task being executed by a worker is 89.14%, and the average probability of a requester posting a task and being matched with a suitable worker is 73.76%. Figure 5 (a) and Figure 5 Panel (b) shows that task pricing has little impact on the decisions of workers and requesters, demonstrating the platform's flexible pricing strategy, capable of real-time adjustments to meet the needs of both parties. Requesters' selection probability and task urgency remain largely unchanged, with a greater focus on completion time than price. Workers' selection probability increases slightly with additional rewards but remains generally stable, with salary expectations fluctuating within a certain range. Long-term employment provides workers with an accurate understanding of their market value, and stable employment opportunities and a favorable environment also influence their decisions.

[0071] Figure 6The left y-axis compares the base price and actual task pricing for different task types. The right y-axis shows the percentage by which the actual task pricing exceeds the base price. Using the base price as the cost, calculations show that actual task pricing is typically higher than the base price because the platform adjusts prices based on factors such as task difficulty, urgency, and worker skills. This price adjustment mechanism ensures platform profitability, incentivizes worker participation, and ensures fair compensation for workers.

[0072] The present invention is carried out in a dynamic environment and three experimental data sets are constructed from five real data sets to verify the stability of the GGMFA algorithm. Figure 7 As shown in Figure 2, the present invention selects three key indicators to verify the stability and adaptability of the GGMFA algorithm on three experimental data sets. Figure 7 (a) in the figure shows the comparison of the platform’s optimization effect on workers’ evaluation. Figure 7 (b) shows the comparison of task pricing optimization effects. Figure 7 (c) in Figure 3 compares the effectiveness of worker motivation optimization. Despite the differences in the experimental datasets, the three indicators fluctuated slightly across the three rounds of experiments, demonstrating that the GGMFA algorithm maintains a stable strategy despite the diversity and dynamics of the datasets.

[0073] Example 2 This embodiment provides a Stackelberg dynamic heterogeneous task allocation system for crowd intelligence perception, including: A data acquisition module is configured to acquire task data, requester history data, and worker history data of a crowdsourcing task; A strategy indicator module is configured to calculate a task allocation strategy indicator based on the task data, the requester history data, and the task history data; The balancing module is configured to balance the task allocation strategy indicators based on the improved fireworks algorithm; The prediction module is configured to predict the user task response allocation based on the task allocation strategy indicator; The allocation module is configured to obtain a task response allocation result.

[0074] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the method described.

[0075] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executing the method described.

[0076] The above are all preferred embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A Stackelberg dynamic heterogeneous task allocation method for crowd intelligence perception, characterized by: include: Obtain task data, requester history data, and worker history data for crowdsourcing tasks; Calculate task allocation strategy indicators based on task data, requester history data, and task history data; Balance task allocation strategy indicators based on improved fireworks algorithm; Predict user task response allocation based on task allocation strategy indicators; Get the task response assignment result.

2. The Stackelberg dynamic heterogeneous task allocation method for crowd intelligence perception according to claim 1 is characterized in that: The task allocation strategy indicators are calculated based on the task data, the requester's historical data and the worker's historical data, including calculating the matching degree between the task and the worker based on the task data and the worker's historical data, wherein the project requirement coverage is calculated based on the worker's skill set and the project requirement set, and the matching degree between the worker and the task's geographic location is calculated using the Manhattan distance. For the worker's geographic location Lw = (xw, yw), the task's geographic location Lt = (xt, yt), and the Manhattan distance between the worker and the task are expressed as: d(Lw, Lt) = |xw-xt|+|yw-yt|, and the worker's historical data is evaluated, and the worker's historical performance is used as an important basis for the platform to assign tasks. , workers' history is manifested as: , and calculate the worker reputation value. Let Q represent the quality of task completion, g represent the worker's praise rate, that is, the proportion of positive evaluations. The worker reputation value is expressed as: .

3. The Stackelberg dynamic heterogeneous task allocation method for crowd intelligence perception according to claim 2 is characterized in that: The task allocation strategy indicator is calculated based on the task data, the requester's historical data, and the worker's historical data, and further includes calculating the worker's task reward coefficient based on the task data and the worker's historical data, wherein the worker's reputation value is calculated based on the task completion quality and the worker's historical positive evaluation data; the worker's behavior response rate is calculated based on the task data and the worker's historical performance data; and the worker's reward coefficient is determined based on the worker's reputation value and the worker's behavior response rate. The reward coefficient is expressed as: ， Here, Δ is a constant.

4. The Stackelberg dynamic heterogeneous task allocation method for crowd intelligence perception according to claim 3 is characterized in that: The task allocation strategy indicator is calculated based on the task data, the requester's historical data, and the worker's historical data, and further includes calculating the requester's reputation value based on the task data and the requester's historical data, wherein the difficulty of the tasks published and completed by the requester is divided and assigned different coefficients based on the task data to calculate the requester's experience value; and the requester's reputation value is calculated based on the requester's historical data on the platform, wherein PO represents the number of task rewards that the requester did not pay in time after the worker completed the task normally, PO1 represents the number of payments beyond the deadline, and PO0 represents the number of payments that were never made. The requester's reputation value is expressed as: 。 5. The Stackelberg dynamic heterogeneous task allocation method for crowd intelligence perception according to claim 4 is characterized in that: The improved fireworks algorithm is based on balancing the task allocation strategy indicators, including maximizing the matching degree between worker skills and tasks as the goal, using the value of the matching degree between worker skills and tasks as the fitness indicator, using a greedy algorithm to screen and calculate the average fitness, and obtaining the initial fireworks population and optimal weight through iteration. By setting the maximum and minimum amplitude of the fireworks explosion and the maximum and minimum number of sparks generated by the fireworks explosion, the formula for calculating the amplitude and number of explosion sparks of each firework is: A = A_min + (A_max - A_min) * (M - fmin) / (fmax - fmin) N_sparks = N_sparks_min + (N_sparks_max - N_sparks_min) * (M - fmin) / (fmax - fmin) The spark produced by the explosion is represented as Fij, where i represents the i-th firework and j represents the j-th spark produced by the explosion of the i-th firework, j∈[1, N_sparks]. Spark mutation is set to obey Gaussian distribution, and all data in the spark candidate pool are updated to the mutated sparks. The sparks in the candidate pool are selected and iterated. The final Fire obtained after the iteration is terminated is a set of optimal solutions for the matching degree between worker skills and tasks.

6. The Stackelberg dynamic heterogeneous task allocation method for crowd intelligence perception according to claim 5 is characterized in that: The prediction of user task response allocation based on the task allocation strategy indicator includes calculating the platform reputation value based on the worker's historical data and the requester's historical data, wherein the platform reputation value is calculated based on the worker's rating of the platform and the requester's rating of the platform, and the requester's risk perception coefficient is calculated based on the platform reputation value and the worker's reputation value, which is expressed as: ， in is the risk perception coefficient of the requester, 、 and is an adjustment parameter; the worker risk perception coefficient is calculated by the platform reputation value and the requester reputation value, which is expressed as: 。 7. The Stackelberg dynamic heterogeneous task allocation method for crowd intelligence perception according to claim 6 is characterized in that: The user task response allocation prediction based on the task allocation strategy indicator also includes calculating the requester behavior response function based on the task allocation strategy indicator, which is expressed as: ， in Indicates the urgency of the task. Represents the requester budget.

8. The Stackelberg dynamic heterogeneous task allocation method for crowd intelligence perception according to claim 7 is characterized in that: The prediction of user task response allocation based on the task allocation strategy indicator also includes calculating the worker behavior response function based on the task allocation strategy indicator, which is expressed as: ， in represents the worker skill and task adaptation ratio obtained by strategy 1, represents the worker's reward coefficient obtained in strategy 2, Similar to the requester behavior response function, Indicates the number of grids across which workers execute tasks.

9. The Stackelberg dynamic heterogeneous task allocation method for crowd intelligence perception according to claim 8 is characterized in that: The user task response allocation prediction based on the task allocation strategy indicator also includes predicting the optimal solution of the task allocation strategy based on the worker behavior response function and the requester behavior response function, and verifying the effect of the task allocation strategy using the requester risk perception coefficient and the worker risk perception coefficient.

10. A Stackelberg dynamic heterogeneous task allocation system for crowd intelligence perception, characterized by: include: A data acquisition module is configured to acquire task data, requester history data, and worker history data of a crowdsourcing task; A strategy indicator module is configured to calculate a task allocation strategy indicator based on the task data, the requester history data, and the task history data; The balancing module is configured to balance the task allocation strategy indicators based on the improved fireworks algorithm; The prediction module is configured to predict the user task response allocation based on the task allocation strategy indicator; The allocation module is configured to obtain a task response allocation result.

Images