Worker recruitment and task matching method based on bidirectional preference degree verification mechanism

Abstract

The invention discloses a worker recruitment and task matching method based on a bidirectional preference degree verification mechanism so as to improve data quality and platform income of a crowd sensing network. In order to solve the problem that workers may falsely report preferences to obtain higher remuneration, a worker preference matrix is established through dynamic interaction, a multi-arm selection machine model is introduced, exploration and utilization are balanced through a greedy strategy, and the real preference degree of the workers is gradually found. And the platform calculates a worker preference quotation ratio in each round of matching, and optimizes worker selection and avoids a local optimal trap in combination with the number of successful matching times, an exploration index and historical preference performance. According to the method, the task matching satisfaction degree is effectively improved, low-quality data caused by false preferences is reduced, and the long-term efficiency and sustainability of a crowd sensing system are improved.

Description

Technical Field

[0001] The present invention belongs to the field of data collection in crowd intelligence networks, and particularly relates to a method for recruiting workers with high preference degrees and realizing stable matching between tasks and workers in crowd intelligence networks. Background Art

[0002] Crowdsensing network is a new data collection and application paradigm that relies on public participation to complete data collection tasks. The system platform will issue collection tasks, including the location, content, and corresponding reward information of data collection. Data collection is mainly completed by workers using mobile phones or other sensing devices. They sense the data of the surrounding environment through the devices and submit it to the platform to obtain rewards. Due to the large number of participants, the crowdsensing network can achieve long-term, large-scale, and low-cost data collection, providing efficient and reliable data support for various applications. Therefore, a suitable solution for crowdsensing lies in selecting high-quality workers to maximize the quality of the collected data while minimizing the sensing cost. However, in practical applications, workers and tasks have different preferences respectively. Workers may be more inclined to tasks with high rewards, low costs, or that match their own capabilities; tasks may require specific regions, times, or high-quality data. The mismatch of preferences may lead to the following situations: the generated tasks cannot meet the needs of both workers and task parties; workers are dissatisfied with the tasks, which may reduce their willingness to participate in the future. Therefore, by selecting suitable workers for tasks, the satisfaction of the matching between tasks and workers can be maximized, thereby improving the overall efficiency of the system.

[0003] In the crowd-sensing network, it is generally believed that workers have different preferences for different tasks, and their preferences are a partial order relationship. If a worker has a higher preference for a task, the worker will overcome the difficulties in completing the task, which will increase the task completion rate. In addition, if a worker is assigned a task with a high preference, the worker's enthusiasm will be higher, so that more workers will participate in the crowd intelligence sensing, and thus large-scale tasks can be completed with high quality. Therefore, although preference is actually an incentive without material cost, it plays a very important role in encouraging employees to complete tasks. In addition, tasks also have preferences for employees, and such preferences are also manifested in many aspects. First, choosing different workers for the same task will result in different task qualities, which makes the quality of the final synthesized application on the platform different. For example, for multimedia data sensing, the multimedia data sensed in different places should be different types of data, so that after platform integration, rich and various types of data can be obtained. If most of the sensed data is homogeneous, some types of data will be lost from the final data, which will affect the final quality of the application. Therefore, the matching between workers and tasks should be considered when assigning tasks to workers. Therefore, the matching solution based on the two-way preferences of workers and tasks can help participants obtain satisfactory task assignments; otherwise, there will be unhappy participants, which will affect the sustainable development of the crowd-sensing network.

[0004] However, the preference-based matching method still faces a thorny problem: workers claim false preferences to obtain illegal rewards. Since the preference degree of a worker for a task is claimed by the worker himself, some dishonest workers will set the tasks that are beneficial to themselves and can obtain greater rewards as higher preferences, so as to increase their probability of being selected. In fact, workers may submit false data or not execute, thus interfering with the platform's selection of other workers and resulting in low platform quality. It can be seen that in the actual crowd-sensing network, it is necessary to verify the authenticity of the preferences claimed by workers. However, all current studies assume that the preference degrees claimed by workers are true, which obviously cannot be applied to the actual crowd-sensing network. Therefore, a mechanism for verifying true preference degrees is still needed to ensure high-quality task matching and the healthy development of the platform. For example, when a worker applies for a task and marks his preference for the task, he will mark a higher preference for the tasks that are close to his movement trajectory, because for such tasks, the distance the worker needs to move when performing the task is small, so the cost he spends is low and he has a competitive advantage. However, since the number of workers in the crowd-sensing network is very large, there are often multiple workers claiming high preferences for the same task. Therefore, the optimization between the cost of verifying true preference degrees and the platform's revenue is also an important issue worthy of research. How to strategically select as few preferential workers as possible for testing to obtain their true preferences, so that the workers selected by the platform are those close to the maximum preference degree, is of great significance for reducing costs and making the platform's revenue approach the optimal value. Summary of the Invention

[0005] The present invention discloses a worker recruitment and task matching method based on a two-way preference degree verification mechanism. The inventive method provides an effective solution for dynamic worker recruitment in a crowd network to maximize the long-term benefits of the platform. The platform recruits workers and interacts with them, thereby being able to obtain a set of preference degree rankings of tasks for workers, and this set is dynamically updated. For the preference degrees claimed by workers, the preference degree of a worker for a task in each round is calculated according to the initial preference quotation ratio, and the cumulative verification times in the dynamic recruitment process can be calculated by calculating the number of preference degree matches in each round. However, if only the greedy selection is made to calculate the combination with the highest matching satisfaction in each round, the method will fall into a local optimal solution, that is, only the workers with a certain known preference degree will be selected, and some unknown workers will never be selected, or there are some workers with a generally high preference degree who are ignored by the platform for a long time due to large fluctuations in the quality of submitted tasks in the initial stage of the algorithm. The inventive method introduces the model of a multi-armed bandit, maintains a set of exploratory workers through the μ parameter, and realizes the maximization of the long-term interests of the platform by balancing exploration and exploitation.

[0006] The technical solution of the invention is as follows:

[0007] 1. A worker recruitment and task matching method based on a two-way preference verification mechanism, characterized by including the following steps:

[0008] (1) Calculate the satisfaction list P of tasks for workers based on the candidate information submitted by workers j with the true preference list for the number of matches to achieve two-way preference verification. The model is as follows: The platform constructs an application The program can be constructed in rounds, The task set is defined as The worker set is defined as Workers submit a candidate task set to the platform to apply for data collection, where represents the set of task options submitted by worker w i in the t-th round, represents the k-th task option information submitted by worker w i in the t-th round, that is, each task option contains three elements: represents the k-th task number that worker w i is willing to participate in in the t-th round, Worker w i The preference degree for the k-th task in the t-th round, represents the quote of the k-th task that worker w i is willing to participate in in the t-th round;

[0009] (2) Initialization: Initialize the high-exploratory worker set as the low-exploratory worker set as The probability value μ of randomly selecting workers from the high-exploratory worker set is 0.5;

[0010] (3) Recruit workers through the following methods:

[0011] First, construct a preference matrix of tasks for workers For each task option reorder according to the preference degree from high to low. For each task in the perceived task set and workers, an initial matrix of can be constructed, and all preference values are initialized to 0; for each in each If Then the preference value is assigned to the corresponding task, and the column value of each task corresponding to the worker in the matrix is updated; finally, the preference matrix of tasks for workers is constructed; similarly, the bidding matrix of tasks for workers is calculated by the preference bidding ratio claimed by the worker, that is:

[0012]

[0013] where α i,j is the satisfaction degree of task t j for worker w i ; is the element value of the matrix in the t-th round, the i-th row and the j-th column, is the element value of the matrix in the t-th round, the i-th row and the j-th column. If worker w i does not claim the preference value for task t j , then the corresponding α i,j is 0; the satisfaction degrees of each column corresponding to task t j in the matrix are sorted to obtain the preference list P j of task t j ;

[0014] If is not empty, a part of the workers are randomly selected from the set of highly exploratory workers at a ratio of μ, and a part of the remaining workers are selected from the preference degree list P j of tasks for workers at a ratio of 1 - μ, and the workers with the highest satisfaction degree α i,j are selected; then when the value of μ is greater than 0.1, the value of μ is decreased, such as decreasing by 0.02 per round; if is empty or the selection is completed, the workers with the highest preference degree are selected from the preference degree list P j of tasks for workers;

[0015] (4) The platform calculates the cumulative verification times through the following method:

[0016] After the platform obtains the true preference degree list of tasks for workers through interaction with workers , the degree of consistency between the task preferences claimed by the worker in a certain round and the true preferences calculated by the platform can be calculated, and the number of matches is used as a measurement standard. The following gives the calculation formula for the number of matches:

[0017]

[0018] is the list P of workers sorted by satisfaction j with the true preferences calculated by the platform for the number of matching validations, is the binary matrix The value of the element in, indicating whether the k-th task submitted by worker w i matches the true preference. The matching task is counted as 1, and the non-matching one is counted as 0; when obtaining the number of matching validations of the worker in each round, the cumulative validation count of worker w i in the t-th round can be updated:

[0019]

[0020] n i,j (t) represents task t j for worker w i the cumulative validation count in the preference order in the t-th round. When worker w i the currently selected task is in the cumulative validation count n i,j (t) increases, otherwise the validation count remains unchanged;

[0021] (5) After obtaining the cumulative validation count n i,j (t), the exploration degree Q i,j (t) of the worker can be calculated according to the validation frequency:

[0022]

[0023] where indicates that the k-th candidate task option submitted by worker w i in the t-th round exactly comes from the tasks in the perceived task set; then according to the exploration degree Q i,j (t), the cumulative validation count n i,j (t) and the true preference degree calculate the comprehensive exploration index of the worker:

[0024]

[0025] Establish a set of highly exploratory workers in the perceived task set a where the comprehensive exploration index S i,j (t) of the worker is greater than the high threshold In the set of highly exploratory workers sort in descending order according to the comprehensive exploration index S i,j (t) of the worker; another set of low-exploratory workers in the perceived task set a where the comprehensive exploration index Si,j (t) is lower than the low threshold The classification formulas for highly exploratory and lowly exploratory workers are as follows:

[0026]

[0027] Among the set of highly exploratory workers sort in descending order according to the comprehensive exploration index S i,j (t) of the workers;

[0028] (6) Calculate the matching satisfaction in the following way:

[0029]

[0030] θ(t) represents the comprehensive satisfaction of the matching between the tasks in the perceived task set in this t-th round and the workers. For a specific worker w i , calculate the satisfaction improvement ratio of worker w i , and the formula is as follows:

[0031]

[0032] represents the satisfaction improvement ratio of task t after selecting worker w i in the current t-th round of perceived task set k , represents the total bid amount put forward by all workers w i for all tasks. According to the greedy principle, allocate k the worker w with the largest value to task t i :

[0033]

[0034] This formula is used to select the final task-worker assignment relationship in the current worker set and store it in the relationship set R. Match the optimal worker w i of each task with each task in the perceived task set a in turn until all tasks are successfully matched.

[0035] Beneficial effects

[0036] The present invention discloses a method for improving data quality by selecting participants based on trust. The basic idea of the inventive method is as follows: In the initial stage, the platform establishes a preference matrix and a quotation matrix for workers. Through the matrices, the preference degree ranking of different tasks claimed by different workers can be obtained. During the interaction process between the platform and the workers, the true preference degree of the workers is gradually discovered. A combination of high-preference workers can be maintained through the preference degree of the workers' long-term performance. After obtaining the true preference degree of the workers, the inventive method can select workers according to the true preference degree. At the same time, the platform will measure the comprehensive exploration index based on the preference degree of the workers, the number of successful matches, and the degree of exploration. The number of successful matches reflects the authenticity of the preferences claimed by the workers in this round. And the degree of exploration Q i,j (t) The higher it is, the lower the frequency of the worker being selected, and the greater the uncertainty. Therefore, it is included in the consideration criteria of the comprehensive exploration index. The preference degree reflects the comprehensive performance of the workers during the long-term interaction process with the platform. Through the comprehensive exploration index, a set of exploratory workers is maintained for small-probability exploration. When a worker with higher quality is found, the worker is added to the set of high-preference workers. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Shows the task completion rate and satisfaction of different methods when the number of workers changes;

[0038] Figure 2 Shows the task completion rate and satisfaction of different methods when the number of tasks changes;

[0039] Figure 3 Shows the performance of different methods in different operating cycles;

[0040] Figure 4 Shows the performance of different application programs of the method of the present invention in different operating cycles; DETAILED DESCRIPTION OF THE INVENTION

[0041] For the convenience of understanding the present invention, the following will describe the present invention more comprehensively and meticulously in conjunction with the accompanying drawings of the specification and preferred embodiments. However, the protection scope of the present invention is not limited to the following specific embodiments.

[0042] Unless otherwise defined, all professional terms used hereinafter have the same meaning as commonly understood by those skilled in the art. The professional terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the protection scope of the present invention.

[0043] Unless otherwise specifically stated, various raw materials, reagents, instruments, and equipment used in the present invention can be obtained through market purchases or can be prepared by existing methods.

[0044] Embodiment:

[0045] In the smart city environment, it is necessary to monitor key indicators such as temperature, noise, and traffic flow in real time, and upload the sensed data to the platform for processing, so as to timely release urban environmental information and provide travel reference and decision-making support for citizens. Since there are a large number of data sensing workers in the city, they can sense and upload data anytime and anywhere. However, there may be untrusted workers among them who deliberately submit false or malicious data, affecting the reliability of the platform. Therefore, there is an urgent need for an efficient and real-time trust evaluation mechanism to ensure that the platform can obtain high-quality, true and reliable data, thereby improving the accuracy and practicality of smart city applications. The method of the present invention develops a simulated data set for the taxi general contract data. In the simulated data set, we set the quotes of workers to complete tasks to be in the range of [0.1, 1]. The workers claim that their preference degrees for each task are random, while the actual preference degrees of tasks for workers are obtained through actual sensing. This configuration method effectively simulates the diversity of the interaction between workers and the platform in the real world, and helps to study the performance of the algorithm under different actual situations. At the same time, we also set different methods for comparison to demonstrate the superiority of the method of the present invention. The following is an explanation of different methods: The method of selecting workers according to the minimum quote is that the platform greedily selects the worker with the lowest quote in each round; the method of matching tasks according to the worker's preference is that the platform selects workers according to the claimed preferences of the workers and assigns tasks to the worker with the highest claimed preference; the method of maximizing the quote preference ratio is to select the worker with the highest quote preference ratio in each round, that is, select according to the method of step (3) in each round, without using the comprehensive exploration index to explore unknown workers.

[0046] The experimental results of the inventive method are given below.

[0047] From Figure 1 It can be seen that as the number of workers increases, the satisfaction and task completion rate of the method of the present invention are higher than those of other methods. This rule shows that this method can more effectively discover workers with high preference degrees, and then optimize the matching of tasks and workers. At the same time, as the number of workers increases, the satisfaction and task completion rate of the system also increase. This is because when the number of workers increases, the platform can greedily select more high-quality workers for task allocation, thereby improving the satisfaction and task completion rate.

[0048] From Figure 2It can be seen that when the number of tasks increases, choosing the method of the present invention can more significantly improve the task completion rate and satisfaction of the platform. Specifically, the average task completion rate and satisfaction of the method of the present invention are 3.05% and 19.82% higher than those of the method based on maximizing the preference bid ratio. This is because in the case of a large number of tasks, this method can collect more worker recruitment interaction data through the multi-armed bandit mechanism, improving the accuracy of task allocation.

[0049] Figure 3 Shows the performance of different methods in different running cycles. It can be seen from the figure that μ = 0.3 performs best in improving the task completion degree in the initial stage, but fluctuates greatly in the later stage. This is because when the platform has already explored the highly preferred workers in the worker group relatively fully, continuing to explore at a high frequency will instead introduce fluctuations.

[0050] Figure 4 Describes the relationship between the satisfaction degree of workers and the number of iterations. Generally, the satisfaction degree increases relatively quickly under the configuration of μ = 0.3, but fluctuates greatly, while the growth is relatively stable in the case of μ = 0.1. This is because the higher the greediness, the more the method tends to explore unknown workers and the set of trustworthy workers when selecting workers. Therefore, it can obtain the true preference degrees of more workers faster, thus improving the satisfaction degree faster. However, the strategy with high greediness tends to select the same batch of highly preferred workers when exploring the preferences of workers, ignoring the preferences of other workers, which may lead to uneven satisfaction degrees among some worker groups, thus causing fluctuations. In practical applications, the greediness μ can be appropriately adjusted according to the actual needs of the platform to find the best balance between meeting the rapid improvement of satisfaction and overall matching stability.

Claims

1. A worker recruitment and task matching method based on a two-way preference verification mechanism, characterized in that The following steps are involved: (1) Calculate the worker satisfaction list P based on the candidate information submitted by the worker j With the real preference list The number of matches is used to achieve two-way preference verification. The model is as follows: Platform builds application Program can be divided Wheel build, The task set is defined as The worker set is defined as Workers submit candidate task sets to the platform Apply for data collection, including Represents worker w i The set of task options submitted in round t, Represents worker w i The kth task option information submitted in the tth round, That is, each task option contains three elements: Represents worker w i The kth task number that the user is willing to participate in in round t, Worker i The preference for the kth task in round t, Represents worker w i The bids for the kth task in round t; (2) Initialization: Initialize a set of highly exploratory workers for Low exploratory worker set for From a collection of highly exploratory workers The probability value μ of randomly selecting a worker is 0.5; (3) Recruit workers through the following methods: First, construct the task-to-worker preference matrix For each task option According to preference Rearrange from high to low, for each perception task set tasks and workers, can build The initial matrix, all preference values ​​are initialized to 0; for each Each of if Then Preference value Assign to the corresponding tasks, update the worker column value corresponding to each task in the matrix; finally complete the task preference matrix for workers Construction; Similarly, we can get the quotation matrix of tasks to workers The satisfaction between tasks and workers is calculated by the worker's stated preference-bid ratio, i.e.: where α i,j It is the task j To workers i satisfaction, is a matrix The value of the element in row i and column j in round t, is a matrix The value of the element in row i and column j in round t, if worker w i Undeclared for task t j The preference value of i,j is 0; for each column in the matrix corresponding to task t j Sort the satisfaction of task t j Preference list P j ; if is not empty, then a portion of μ is drawn from the set of highly exploratory workers Workers are randomly selected from the task preference list P at a ratio of 1-μ. j Select the satisfaction α among the remaining workers i,j The highest worker; then when the μ value is greater than 0.1, reduce the μ value, such as reducing it by 0.02 per round; if If it is empty or selected, select the task from the worker preference list P j Select the worker with the highest preference; (4) The platform calculates the cumulative number of verifications using the following method: Get a list of the real preferences of tasks for workers by interacting with them on the platform After that, we can calculate the degree to which the worker’s claimed task preferences in a round are consistent with the actual preferences calculated by the platform, and use the number of matches as a metric. The calculation formula for the number of matches is given below: is a list of workers P sorted by satisfaction j Real preferences for platform computing The number of matching verifications, is a binary matrix The value of the element in represents the worker w i Whether the kth task submitted matches the real preference, the matching task is counted as 1, and the unmatched task is counted as 0; when the number of matching verifications of the worker in each round is obtained, the worker w can be updated i The cumulative number of verifications in the tth round: n i,j (t) represents task t j To workers i The preference order of the worker w is verified in round t. i Currently selected task exist When the cumulative number of verifications is n i,j (t) increases, otherwise the number of verifications remains unchanged; (5) After getting the cumulative number of verifications n i,j (t), the worker exploration degree Q can be calculated based on the verification frequency i,j (t): in Represents worker w i The kth candidate task option submitted in round t happens to come from the perception task set. tasks; then according to the degree of exploration O i,j (t), cumulative number of verifications n i,j (t) and true preference Calculate the worker's composite exploration index: A set of highly exploratory workers in the perceived task set a is established based on the comprehensive exploration index The worker comprehensive exploration index S i,j (t) greater than the high threshold In the high exploratory worker set According to the comprehensive exploration index S of workers i,j (t) from high to low; another set of low exploratory workers in the perception task set a The worker comprehensive exploration index S i,j (t)Below the low threshold The classification formula for high-exploration and low-exploration workers is as follows: In the high exploratory worker set According to the comprehensive exploration index S of workers i,j (t) is sorted from high to low; (6) Calculate the matching satisfaction in the following way: θ(t) represents the perception task set in the tth round The overall satisfaction of matching tasks and workers in , for a specific worker w i , calculate worker w i The satisfaction improvement ratio is as follows: Represents the current t-round perception task set After selecting the worker w i Afterwards, task t k The satisfaction improvement ratio represents all workers w i The sum of bids for all tasks, according to the greedy principle, is k distribute The largest worker i : This formula is used to select the current set of workers and finally store the task and worker assignment relationship in the relationship set R, and assign the optimal worker w for each task in turn. i Match each task in the perception task set a until all tasks are successfully matched.

Images