A data collection method for insufficient participation and budget allocation in crowd intelligence networks

Abstract

The present invention discloses a data collection method for insufficient participation and budget allocation in a crowd intelligence network. It solves the problem of insufficient number of participants by hiring agents to introduce their social neighbors to participate in data collection, and takes into account that participants will be more engaged when performing tasks of interest, and the closer the relationship between participants and their agents, the stronger the restriction by the agents, and the higher the quality of task completion. The present invention quantifies these two factors as the basic interest and cooperation tightness attributes of social neighbors, estimates the value of social neighbors to the platform through these two attributes, thereby obtaining the influence value of the agents on the platform, and selects agents in non-increasing order of the influence value to expand the participant set. During the selection process, the present invention dynamically allocates the budgets for hiring agents and data collectors according to the real-time task completion rate. Compared with the existing inventions, the present invention considers the problems of insufficient number of participants and budget allocation, and improves the revenue of the platform and the utilization rate of the budget.

Description

Technical Field

[0001] The present invention belongs to the field of mobile crowdsensing, and particularly relates to a data collection method for insufficient participation in a crowd network and budget allocation. Background Art

[0002] Mobile crowdsensing is an emerging application of the Internet of Things, which includes three components: the platform of mobile crowdsensing, the set of participants, and the task requester. The task requester submits the task to be sensed and the budget willing to be paid to the platform. The platform recruits participants to execute the task within the limited budget according to the specific requirements of the sensing task. Through such a task publishing model, a large amount of sensing data is collected and constructed into various applications, and then released for users to use. This is an effective method for large-scale data processing. The success of mobile crowdsensing applications depends on the timely and sufficient data submitted by participants to ensure service quality. However, two key attributes have not been well considered, which affect the performance of the platform.

[0003] The first factor is the insufficient number of participants. In previous studies, the application of mobile crowdsensing relied on a large amount of data provided by participants. It was considered in previous studies that the number of participants was sufficient and could provide enough data. However, in reality, the number of participants in tasks is often insufficient. If the number of participants is insufficient, the tasks released by the platform cannot be well completed, and the platform's revenue is very small. Even in the initial stage of operation, there will be a cold start problem.

[0004] The second factor is the budget allocation problem. In previous strategies, the fixed budget allocation method was adopted. The platform's budget was divided into two fixed parts, one part was used to hire task executors, and the other part was used to hire agents to spread tasks to social neighbors. However, in the actual operation process, the arrival situation of participants is uncertain. Sometimes, it may not be necessary to allocate too much budget to agents, and sometimes, because the number of participants in the platform user library is too small, more budget needs to be allocated to agents. These situations are unpredictable. If a fixed budget allocation is used every time, it is inevitable that there will be problems such as low budget utilization or blocked selection of task executors due to a low number of participants. Summary of the Invention

[0005] The present invention discloses a data collection method for insufficient participation and budget allocation in a crowd intelligence network, which can increase the final platform revenue. Aiming at the problem of insufficient number of participants in the mobile crowd sensing network, the invention proposes a method of hiring agents to distribute tasks to social neighbors, and by calculating the value that the social neighbors of the agents can bring to the platform, tries to select agents with high value as much as possible. And in previous strategies, a fixed budget allocation method was usually adopted, and the problem brought by this method is the low utilization rate of the budget, or the low task completion rate caused by the low budget allocated to a certain selection process. Considering these two problems, the present invention proposes a data collection method for insufficient participation and budget allocation in a crowd intelligence network, solves the problem of insufficient number of participants, and adopts a flexible budget allocation method, so that the platform can more effectively and optimally select suitable agents and participants, thereby improving the task completion rate.

[0006] To solve the above technical problems, the present invention adopts the following technical solutions:

[0007] 1. A data collection method for insufficient participation and budget allocation in a crowd intelligence network, characterized by comprising the following steps:

[0008] (1) The platform publishes a set of data collection tasks For each task e b The task type is y b , and it can only be executed by one participant. The platform hires agents and data collectors with a given budget Q within the time limit T. When the tasks are published, only n participants apply to the platform to execute the data collection tasks, and they form a participant set The participants in arrive at the platform dynamically. Each W i has its own interested task types, and these types form a set W i All the social neighbors of form a set Each social neighbor A in j also has its own set of interested task types

[0009] (2) According to the multi-stage sampling reception process, the time T is divided into phases, T = {t1, t2,..., t τ}, and the time interval of the k-th phase is The budget Q is allocated to each phase according to the time proportion of each phase. Then the budget of the k-th phase

[0010] (3) For the budget Q of the k-th phase k , before the start of the k-th phase, according to the task completion rate p at this time k-1Split into two parts. One part is used to hire data collectors to perform tasks, denoted as B k and the other part is used to hire agents to spread tasks to their social neighbors, denoted as R k ;

[0011] (4) In the k-th stage, for the participants who have arrived at the platform, the platform selects appropriate participants to perform data collection tasks on the premise of not exceeding the budget B k and records the set of tasks that have not been assigned for execution as

[0012] (5) Meanwhile, the platform selects agents from the participants who have arrived at the platform according to the following method:

[0013] For participant W i , the platform obtains the social neighbor set of W i

[0014] For each social neighbor A in j , according to the set of task types that A j is interested in the set of task types that its agent W i is interested in and the set of tasks that have not been assigned for execution calculate the value of A to the platform j

[0015] Sum up the values of all social neighbors in the social neighbor set of W i to obtain the influence value of W i to the platform The calculation formula is

[0016] According to the influence value of W i and the bid d proposed by W i calculate the influence rate of W i i The calculation formula is

[0017] (6) The platform sorts the arrived participants in non-increasing order according to the influence rate and selects the participant W with the largest influence rate each time x , if the bid d x of W k does not exceed the agent selection budget R in the k-th stage k ​​​​​For the remaining part, the platform distributes d x as a reward to W x and hires W x as an agent. If the bid d x of W x exceeds the remaining part of R k then the selection of the agent stops;

[0018] (7) The selected agent spreads the task to its social neighbors, and these social neighbors are added to the participant set and initiate a data collection request to the platform when they arrive at the platform;

[0019] 2. A data collection method for insufficient participation and budget allocation in a crowd intelligence network according to claim 1, wherein the value j of the social neighbor A described in (5) is calculated as follows:

[0020] For each social neighbor A j , if the set of task types j that A is interested in has a higher degree of overlap with the set of task types of the tasks that have not been assigned for execution in the k-th stage j , then A is more interested in the remaining tasks, and the base value of the contribution value is higher. The platform calculates the degree of overlap using the Jaccard similarity coefficient and records it as the base interest j of A If the degree of overlap between the set of task types that A j is interested in and that of its agent W i is higher, then the constraining effect of W i on A j is stronger, and the growth value of the value of A j is higher. The platform calculates the degree of overlap between A i and the set of tasks that W is interested in using the Jaccard similarity coefficient and records it as the cooperation tightness j between A i and its agent W Then the base interest j and cooperation tightness of A are calculated as follows:

[0021]

[0022]

[0023] According to A j 's basic interest and cooperation tightness calculate the value of A j to the platform, the calculation formula is as follows: The calculation formula is as follows:

[0024]

[0025] Among them, the basic interest is 's basic value, the cooperation tightness controls 's growth range, and λ is a parameter given by the platform for adjusting 's maximum value of the basic value;

[0026] 3. A data collection method for addressing insufficient participation and budget allocation in a crowd intelligence network according to claim 1, wherein the method for allocating the budget for the k-th stage in (3) is as follows:

[0027] The budget allocation for the k-th stage is determined by the task completion situation in the k - 1 stage. The task set is transformed into an m×1 matrix. The element in the m-th row and 1st column is the task completion situation, where 1 represents that the task has not been executed and 0 represents that the task has been completed. Then, the matrix generated by the task set is a matrix of all 1s, while the matrix generated by the set of completed tasks in the k - 1 stage is a matrix with alternating 0s and 1s. According to and calculate the task completion rate p of the k - 1 stage through the Frobenius norm k-1 is:

[0028]

[0029] Compare the task completion rate p k-1 with the completion rate threshold given by the platform to determine the budget allocation coefficient ε. The increased part of the total budget for the k-th stage compared to the previous stage is Q k , and B k and R k are increased based on B k-1 and R k-1 . Then Q k will be allocated to B k and R k according to the budget allocation coefficient ε, and the update formula is as follows:

[0030] B k = B k-1 + Q k ·(1 - ε)

[0031] R k = R k-1 + Q k ·ε

[0032] Beneficial effects

[0033] The present invention provides a data collection method for insufficient participation in crowd intelligence networks and budget allocation, which considers the problem of insufficient number of participants and adopts a flexible budget allocation method. Starting from the task types of social neighbors' interests, the present invention calculates the overlap degree with the remaining uncompleted task types and the task types of their agents' interests, so as to estimate the contribution value of the social neighbor to the platform after entering the platform, and selects the agent that can bring the greatest benefit to the platform to spread the task. The newly added social neighbor can also be used as a new agent to spread the task to his social neighbors. In this process, the participant library of the platform is continuously expanded, and the choice of data collectors is more sufficient, thus solving the problem of insufficient number of participants. At the same time, the present invention adopts a flexible budget allocation method. In the process of selecting agents and data collectors, the budget for hiring agents and the budget allocation for hiring data collectors are adjusted according to the task completion rate in stages, effectively improving the utilization rate of the budget, and at the same time, the income of the platform also becomes higher. Brief description of the drawings

[0034] Figure 1 It is a schematic structural diagram of the present invention for hiring an agent to spread a task to a social neighbor.

[0035] Figure 2 It is the task completion rate of the present invention and the existing method under different total budgets and initial number of participants.

[0036] Figure 3 It is the amount of data obtained by the present invention and the existing method under different total budgets, initial number of participants and number of tasks.

[0037] Figure 4 It is a comparison chart of the task completion rate and the amount of data obtained by the present invention and the fixed budget allocation method under different total budgets and initial number of participants. Detailed implementation manners

[0038] Example:

[0039] In the data collection tasks published in the mobile crowd sensing network, it is impossible for all the tasks published by the platforms to have enough participants to execute the tasks. Moreover, in the dynamic mobile crowd sensing network, participants arrive dynamically. In the case where the total quantity is not optimistic, the dynamic random arrival makes the initial platform task execution situation even more disadvantaged. For some time-sensitive tasks, if no executor can be found to execute the tasks at the initial stage, the value of these time-sensitive tasks will decrease or even disappear. In applied statistics, there is a snowball sampling method. When sampling a sparse population, the number of surveyed objects can be increased by inviting the surveyed objects to invite other objects. In addition, the disadvantages of fixed budget allocation are also very obvious. In platforms with too few participants, it is reasonable to allocate more budgets to hire agents, while in platforms with more participants, it is unreasonable. Therefore, according to the actual situation of the platform, the present invention continuously adjusts the budget allocation during the operation process.

[0040] The present invention will be further described below with reference to the accompanying drawings.

[0041] Figure 1 It is a schematic structural diagram of the present invention for hiring agents to spread tasks to social neighbors. Agents can spread to social neighbors through multiple layers. For example, in the first layer in the figure, it is the first time that agents spread to social neighbors, and the newly added social neighbors can be selected again as agents to spread tasks. Through multiple-layer spreading, the platform can be quickly expanded.

[0042] Figure 2 It is the task completion rate of the present invention and the existing methods under different total budgets and initial participant numbers. Among them, OPT represents the optimal algorithm in the offline scenario, SocialRecruiter is an offline algorithm proposed based on the epidemic model, and MTSF is an online algorithm with only one-time spread from agents to social neighbors. It can be seen that as methods in the offline scenario, OPT and SocialRecruiter are greater than the present invention and MTSF in the online scenario. However, when the budget is sufficient, the task completion rate of the present invention is close to 70% and far greater than the MTSF method.

[0043] Figure 3 It is the data volume obtained by the present invention and the existing methods under different total budgets, initial participant numbers and task quantities. It can be seen that the present invention and OPT reach an approximation ratio of 57.5% on average, and the obtained data volume is 80.7% higher than the MTSF method.

[0044] Figure 4This is a comparison of the task completion rate and the amount of data obtained between the present invention and the fixed budget allocation method under different total budgets and initial numbers of participants. It can be seen that under different budgets and initial numbers of participants, the budget allocation method of the present invention is superior to other methods. For the fixed allocation method, the 8:2 allocation method is the optimal method among fixed allocations when the budget is very large, but it no longer has an advantage when the initial number of participants is very large. Moreover, for time-sensitive tasks, too little budget allocated to hired agents in the early stage may result in an insufficient number of participants. Therefore, adopting a flexible allocation method in the present invention is the optimal solution. Whether it is for the task completion rate and the amount of data, the flexible allocation method is superior to the fixed allocation algorithm, and the utilization efficiency of the budget is improved.

Claims

1. A data collection method for insufficient participation and budget allocation in a crowd intelligence network, characterized in that including the following steps: (1) The platform publishes a set of data collection tasks Each task e b has a task type of y b , and can only be executed by one participant. The platform hires agents and data collectors within the time limit T and the given budget Q. When the tasks are published, only n participants apply to the platform to execute the data collection tasks, and they form the participant set The participants in i arrive at the platform dynamically. Each W W i has its own set of task types of interest, and these types form a set All social neighbors of each social neighbor A in j also have their own sets of task types of interest (2) According to the multi-stage sampling reception process, time T is divided into phases, T = {t1, t2, …, t τ}, and the time interval of the k-th phase is The budget Q is allocated to each phase according to the time proportion of each phase. Then the budget of the k-th phase is (3) For the budget Q at stage k k , it is split into two parts according to the task completion rate p at this time before the start of stage k k-1 . One part is used to hire data collectors to execute tasks, denoted as B k , and the other part is used to hire agents to spread tasks to their social neighbors, denoted as R k ; (4) In the k-th stage, for the participants who have reached the platform, the platform selects appropriate participants to perform data collection tasks on the premise that the budget B is not exceeded, and records the set of tasks that have not been assigned for execution as k and records the set of tasks that have not been assigned for execution as (5) Meanwhile, the platform selects agents from the participants who have arrived at the platform according to the following method: For participant W i , the platform obtains the i social neighbor set of W For each social neighbor A in , according to the set of task types that A j is interested in, its agent W j is interested in the set of task types and the set of tasks that have not been assigned for execution i is interested in the set of task types and the set of task types in , calculate the value of A to the platform j Sum the values of all social neighbors in i the social neighbor set of W to obtain the impact value W i on the platform The calculation formula is According to W i influence value and the bidding price d i proposed by W i calculate the influence rate i of W The calculation formula is (6) The platform sorts the arrived participants in non-increasing order according to the influence rate and selects the participant W with the highest influence rate each time x . If the bid d x of W x does not exceed the remaining part of the agent selection budget R k in the k-th stage, then the platform distributes d x as the reward to W x and hires W x as an agent. If the bid d x of W x exceeds the remaining part of R k , then the selection of agents stops; (7) The selected agent propagates the task to its social neighbors, who are added to the set of participants. When they arrive at the platform, they initiate a data collection request to the platform.

2. A data collection method for insufficient participation and budget allocation in a crowd intelligence network according to claim 1, characterized in that The social neighbor A described in (5) j Value The calculation method is as follows: For each social neighbor A j , if the set of task types that A j is interested in has a higher overlap with the set of tasks that have not been assigned for execution in the k-th stage in terms of the set of task types , then A j is more interested in the remaining tasks, and the base value of the contribution value is higher. The platform calculates the overlap and using the Jaccard similarity coefficient and records it as the base interest j of A If, however, the overlap of the set of task types that A j is interested in with that of its agent W i is higher, then W i has a stronger constraint on A j , and the growth value of the value of A j is higher. The platform calculates the overlap between i the set of tasks that A is interested in and that of W using the Jaccard similarity coefficient and records it as the cooperation tightness j between A i and its agent W Then the base interest j and cooperation tightness of A are calculated as follows: According to A j basic interest and cooperation tightness calculate the value of A j to the platform The calculation formula is as follows: Among them, the basic interest is the basic value, and the cooperation tightness controls the growth range, and λ is a parameter given by the platform for adjusting the maximum value of the basic value.

3. A data collection method for insufficient participation in crowd intelligence networks and budget allocation according to claim 1, characterized in that (3) The method of allocating the budget for the k-th stage is as follows: The budget allocation for the k-th stage is determined by the task completion of the k-1 stage. The task set is transformed into an m×1 matrix. The element in the m-th row and 1st column is the task completion status, where 1 represents the task has not been executed and 0 represents the task has been completed. Then, the matrix generated by the task set is a matrix of all 1s, while the matrix generated by the completed task set of the k-1 stage is a matrix with alternating 0s and 1s. According to and , the task completion rate p of the k-1 stage is calculated through the Frobenius norm k-1 as follows: The task completion rate p k-1 is compared with the completion rate threshold given by the platform to determine the budget allocation coefficient ε. The increased part of the total budget in the k-th stage compared with the previous stage is Q k , and B k and R k are increased on the basis of B k-1 and R k-1 . Then Q k will be allocated to B k and B k according to the budget allocation coefficient ε, and the update formula is as follows: B k = B k-1 + Q k ·(1 - ε) R k = R k-1 + Q k · ε。

Images