CN111639969B - Dynamic incentive calculation method, system, equipment and medium for crowdsourcing system - Google Patents

Abstract

The invention discloses a dynamic incentive calculation method and a system for a crowdsourcing system, wherein the method comprises the steps of obtaining task data of a demander on a crowdsourcing platform and historical task question-answer data of a participating user; assigning tasks to participating users; constructing a cyclic neural network model for each participating user; training a cyclic neural network model according to historical task question-answer data of the participating users; according to the prediction results of the participating users, the tasks and the cyclic neural network model, calculating the final benefit brought by different excitation values, and judging whether the current user is given an excitation value or not; the answers of all participating users are collected to obtain the results required by the demander. When the online profit calculation problem is solved, the invention also provides a simple and efficient algorithm. Simulation experiments prove the high efficiency and the robustness of the invention under complex conditions. Practical experiments on crowdsourcing platforms also show the high efficiency and superiority of the present invention over traditional methods.

Description

Dynamic incentive computing method, system, device and medium for crowdsourcing system

technical field

The present invention relates to the problem of data quality improvement, in particular to a dynamic incentive calculation method, system, equipment and medium for a crowdsourcing system.

Background technique

In practical applications, there are many problems that humans can easily complete but machines cannot directly complete. For example, given two pictures of different resolutions, humans can quickly and accurately distinguish them, but it is difficult for machines to recognize them. The same example is the clear judgment of natural language emotion. In this context, crowdsourcing platforms have received extensive attention and development.

Crowdsourcing platform refers to a network job distribution platform. Demanders publish various tasks on the platform, users browse and select tasks, and demanders give certain rewards according to the quality of work submitted by users.

With the continuous development of crowdsourcing platforms, how to give incentives to improve the quality of question answering data has become a key issue. Some studies have found that appropriate incentives (such as prestige, money, etc.) can improve the quality of questions and answers, thereby increasing the ultimate benefit of the demander. However, giving too much incentive value will reduce the total income of the demander, which in turn will reduce the overall quality of the answer data, and too little incentive value will discourage the enthusiasm of users to reply, and the task may not be completed. Under such circumstances, a good incentive mechanism not only requires reasonable incentives for users to obtain high-quality replies, but also enables demanders to obtain relatively high returns through users' answers.

Scholars at home and abroad have done some work on the incentive calculation problem on the crowdsourcing platform, but these works still have limitations: (1) the incentive model is single, and the overall data quality improvement effect is limited; (2) every Each user who completes the task gets the same benefits, without considering the impact of answer quality and user behavior.

Contents of the invention

Embodiments of the present invention provide a dynamic incentive calculation method, system, device, and medium for a crowdsourcing system to solve the problem that the traditional solution is single and does not consider the quality of the answer. Respondents are motivated to maximize the quality of task data.

In order to achieve the above object, the present invention adopts following technical scheme:

In the first aspect, an embodiment of the present invention provides a dynamic incentive calculation method for a crowdsourcing system, the method includes the following steps:

Obtain the task data of demanders on the crowdsourcing platform and the historical task question and answer data of participating users;

Assign tasks to participating users;

Construct a recurrent neural network model for each participating user;

According to the historical task question and answer data of participating users, train the recurrent neural network model;

Based on the prediction results of the participating users, tasks and the cyclic neural network model, calculate the final benefits brought by different incentive values and judge whether to give the current user an incentive value;

Collect answers from all participating users to get the results required by demanders.

Further, the acquired task data includes: the number of tasks, the number of tasks assigned in each round, and the total set of tasks requiring question and answer. The historical task question and answer data of participating users includes the number and quality of the user's past task questions and answers.

Further, the recurrent neural network model is a time series model, which is used to predict the quality level of the user's output answer when the incentive value is given. The recurrent neural network model is composed of multiple fully connected layers, and at each time node Receive the output of the previous time node to form a cyclic neural network.

Further, the main parameters of the recurrent neural network model are as follows:

1) The demander can decide whether to give the user an incentive value at each time node t, denoted as a _t , if a _t is 1, the incentive value is given, and if 0 is no;

2) The output of the cyclic neural network model is the variable y _t , which represents the probability that the user completes the task with high quality. The closer y _t is to 1, the higher the quality of the answer is, and the closer y t is to 0, the lower the quality of the answer;

3) There are multiple hidden states, and the transfer parameters between the input and hidden states, between hidden states, and between hidden states and outputs in the neural network are obtained by training the cyclic neural network model.

Further, the construction steps of the recurrent neural network model are as follows:

The training data set is the user's historical task question answering data sequence _{<a t , y t >} , and the model parameters are updated and optimized through the backpropagation of the recurrent neural network to optimize its performance on the training data set. After multiple iterations of training, the desired answer quality assessment model can be obtained.

Further, according to the prediction results of participating users, tasks and cyclic neural network, calculate the final benefit brought by different incentive values to judge whether to give the current user an incentive value, including:

For a certain user, given the corresponding incentive value, as the input of the recurrent neural network model, the predicted answer quality corresponding to the incentive value is obtained;

Construct the final reward function through the predicted answer quality, and the final reward function includes the reward for completing the next task and the reward for completing future tasks;

Solving the corresponding final benefit function can obtain the final benefit obtained by the corresponding incentive value, so as to decide whether to give the current user an incentive value.

Further, the solution to the corresponding final revenue function adopts a heuristic pruning strategy to reduce the computational complexity of the model.

In the second aspect, the embodiment of the present invention also provides a dynamic incentive computing system for a crowdsourcing system, including:

The acquisition module is used to acquire the task data of demanders on the crowdsourcing platform and the historical task question and answer data of participating users;

Assignment module for assigning tasks to participating users;

A model building block for building a recurrent neural network model for each participating user;

The model training module is used to train the recurrent neural network model according to the historical task question and answer data of the participating users;

The decision-making module is used to calculate the final benefits brought by different incentive values according to the prediction results of participating users, tasks and the cyclic neural network model, and judge whether to give the current user an incentive value;

The result output module is used to collect the answers of all participating users to obtain the results required by the demanders.

In a third aspect, the embodiment of the present invention also provides an electronic device, including:

one or more processors;

memory for storing one or more programs;

When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in the first aspect.

In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the method described in the first aspect is implemented.

According to the above technical solution, the method includes: assigning the task to the user, and establishing a recurrent neural network model for prediction through the user's historical task question and answer data; through the trained model, this method predicts whether to give the user the quality of the user's question and answer data with an incentive value , and calculate the final income from it. This method model not only considers the income of the next task, but also considers the impact of the completion of all future tasks on the final income. At the same time, it can dynamically decide whether to give the incentive value when the current user accepts the task, so as to maximize the demander’s income. benefit in the end. Simulation experiments confirm the efficiency and robustness of the invention in complex situations. The actual experiment on the crowdsourcing platform also shows the high efficiency and superiority of the present invention compared with the traditional method.

Description of drawings

The accompanying drawings described here are used to provide a further understanding of the present invention, and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention, and do not constitute improper limitations to the present invention. In the attached picture:

Fig. 1 is a flow chart of a dynamic incentive calculation method for a crowdsourcing system according to an embodiment of the present invention;

Fig. 2 is the model system block diagram in the embodiment of the present invention;

Fig. 3 is the schematic diagram of recurrent neural network (RNN) in the embodiment of the present invention;

Fig. 4 is a block diagram of a dynamic incentive computing system used in a crowdsourcing system according to an embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

On the contrary, the invention covers any alternatives, modifications, equivalent methods and schemes within the spirit and scope of the invention as defined by the claims. Further, in order to make the public have a better understanding of the present invention, some specific details are described in detail in the detailed description of the present invention below.

Embodiment one

Fig. 1 is a flow chart of a dynamic incentive calculation method for a crowdsourcing system according to an embodiment of the present invention. As shown in Fig. 1, the method includes the following steps:

Step S100: Obtain task data submitted by demanders on the crowdsourcing platform and historical task question and answer data of participating users. The task data includes the task set O that needs to be completed, the total number of tasks C, and the number c of tasks that need to be completed in each round.

Step S200: According to the above task information, assign the tasks of each round to the users.

Step S300: Construct a recurrent neural network model for each participating user for prediction according to the historical task question and answer data of the participating users.

Step S301: For each user, establish a corresponding recurrent neural network RNN model, the RNN model is a time series model, used to predict the quality level of the user's output answer when the incentive value is given. The RNN model is composed of multiple fully connected layers, and receives the output of the previous time node at each time node to form a recurrent neural network. Its main parameters are as follows:

1) The demander can decide whether to give the user an incentive value at each node t, denoted as a _t . When a _t is 1, an incentive value is given, and 0 means no.

2) The output of the cyclic neural network model is the variable y _t , which represents the probability that the user completes the task with high quality. The closer y _t is to 1, the higher the quality of the answer is, and the closer y t is to 0, the lower the quality of the answer;

3) There are multiple hidden states. The transmission parameters U, W, and V between the input and the hidden state, between the hidden states, and between the hidden state and the output in the neural network can be obtained by training the cyclic neural network model.

Step S400: Train the cyclic neural network model according to the historical answer data of the participating users.

Step S401: The steps of training the cyclic neural network model are as follows: the training data set is the user's historical task question answering data sequence _{<a t , y t >} , through the backpropagation of the cyclic neural network, the transfer parameters of the model are modified, and the training data The performance on the set is optimized. After many times of training, the required answer quality prediction model RNN can be obtained.

Step S500: According to the prediction results of participating users, tasks and the cyclic neural network model, calculate the final benefit brought by different incentive values and judge whether to give the current user an incentive value.

Step S501: First, obtain corresponding prediction results through the RNN model. The cyclic neural network RNN trained on the training set can be expressed as:

f _θ : a→y (1)

Where a and y are both L-dimensional vectors, representing whether the L tasks are given incentive values and the probability that the answer is a high-quality reply. Among them, the first L-1 dimensional data is the data of the user’s previous question and answer, which is defined as the current state s, and a _L is the input of the current task. Therefore, the RNN model can also be recorded as y(s, a), s is the current state, a is whether the current task is given an incentive value, and its output is the probability that the current user has a high-quality answer under the state and incentive value level.

Step S502: Second, construct a corresponding loss function based on the predicted result. This method dynamically decides whether to give incentive value by calculating the demander's online income expectation. Assuming that a user has completed several tasks and has several unfinished tasks, the utility function of the total revenue not only includes the revenue from completing the next task, but also considers the revenue from completing future tasks. When the current state s is given, the amount of work to be completed in the future t _n and whether the current incentive value a is given, the income function is recorded as E[U(s; a; t _n )] or E[U], defined as follows :

in

F(s,a)=[1-y(s,a)]w _l +y(s,a)[w _h -I(a)b] (3)

G(s' _{a, y} , t _n -1) = max _{a' ∈ {0, 1}} E[U(s' _{a, y} ; a'; t _n -1)] (4)

I(a) is 1 if and only if a=1, and 0 otherwise.

Step S503: Decide whether to give incentives by maximizing the corresponding revenue function, so as to realize the decision. With the above income function, we can formulate the definition of the decision-making problem, that is, when the current state s is known and the number of tasks to be completed in the future t _n is obtained:

a = arg max _{a ∈ {0, 1}} E[U(s; a; t _n )] (5)

When a=1, the incentive value will be given to the user, if it is 0, it will not be given.

The dynamic solution of the revenue function calculation problem can be solved directly by dynamic programming, but its time complexity is exponential. This scheme proposes an efficient heuristic algorithm to solve the above problems.

1) Beam-Search algorithm: At a certain moment, there may be a certain number of tasks to be completed, so it may be necessary to iterate too many times when seeking the final income at this moment, resulting in excessive computational complexity. Therefore, for a certain task at the current moment, no matter whether it is given an incentive value or not, we only consider the influence of the previous m maximum revenues when calculating the revenue function, where m is the search width and is an artificially determined parameter. Other parts of the income are ignored. Beam-Search is slightly different from the greedy algorithm. The algorithm considers the influence of the first m largest returns, while the greedy algorithm only considers the previous largest return influence. With the previous cyclic neural network model, the Beam-Search algorithm can calculate different benefits according to whether the user gives the incentive value, so as to make a decision.

Step S600: Collect the answers of the previous tasks, and obtain the results through the answers of all users.

Embodiment two

The present invention also provides an embodiment of a dynamic incentive calculation system for a crowdsourcing system, since the dynamic incentive calculation system for a crowdsourcing system provided by the present invention corresponds to the aforementioned example of a dynamic incentive calculation method for a crowdsourcing system , the dynamic incentive calculation system for the crowdsourcing system can realize the purpose of the present invention by executing the process steps in the specific implementation of the above method, so the explanation in the embodiment of the dynamic incentive calculation method for the crowdsourcing system, It is also applicable to the embodiments of the dynamic incentive computing system, which will not be repeated in the following embodiments of the present invention.

As shown in Figure 4, this embodiment provides a dynamic incentive computing system for crowdsourcing systems, including:

The obtaining module 91 is used to obtain the task data of the requester on the crowdsourcing platform and the historical task question and answer data of the participating users;

Assignment module 92, for assigning tasks to participating users;

Model construction module 93, is used for constructing a recurrent neural network model for each participating user;

The model training module 94 is used to train the recurrent neural network model according to the historical task question and answer data of the participating users;

The decision-making module 95 is used to calculate the final profit brought by different incentive values according to the prediction results of the participating users, tasks and the cyclic neural network model, and judge whether to give the current user an incentive value;

The result output module 96 is used to collect the answers of all participating users, so as to obtain the results required by the demanders.

The serial numbers of the above embodiments of the present invention are for description only, and do not represent the advantages and disadvantages of the embodiments.

In the above-mentioned embodiments of the present invention, the descriptions of each embodiment have their own emphases, and for parts not described in detail in a certain embodiment, reference may be made to relevant descriptions of other embodiments.

In the several embodiments provided in this application, it should be understood that the disclosed technical content can be realized in other ways. Wherein, the device embodiments described above are only illustrative. For example, the division of the units may be a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or may be Integrate into another system, or some features may be ignored, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of units or modules may be in electrical or other forms.

The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes. .

The above description is a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, some improvements and modifications can also be made, these improvements and modifications It should also be regarded as the protection scope of the present invention.

Claims (9)

1. A dynamic incentive computation method for a crowdsourcing system, characterized by: comprising the following steps:

acquiring task data of a demander on a crowdsourcing platform and historical task question-answer data of a participating user;

assigning tasks to participating users;

constructing a cyclic neural network model for each participating user;

training a cyclic neural network model according to historical task question-answer data of the participating users;

according to the prediction results of the participating users, the tasks and the cyclic neural network model, calculating the final benefit brought by different excitation values, and judging whether the current user is given an excitation value or not;

collecting answers of all participating users to obtain a result required by a demander;

according to the prediction results of the participating users, tasks and the cyclic neural network, calculating the final benefit size brought by different excitation values to judge whether the current user is given with the excitation value, including:

for a certain user, giving a corresponding excitation value as input of a cyclic neural network model, and obtaining the predicted answer quality corresponding to the excitation value;

constructing a final benefit function through the predicted answer quality, wherein the final benefit function comprises the benefit for completing the next task and the benefit obtained by completing the future task;

the final benefit obtained by the corresponding incentive value is obtained by solving the corresponding final benefit function, thereby deciding whether to give the current user incentive value.

2. A dynamic incentive computation method for a crowdsourcing system as claimed in claim 1 wherein: the acquired task data comprises: the number of tasks, the number of tasks allocated per round and the total set of tasks requiring questions and answers, and the historical task question and answer data of the participating users comprise the number and quality of the past task questions and answers of the users.

3. A dynamic incentive computation method for a crowdsourcing system as claimed in claim 1 wherein: the cyclic neural network model is a time series model for predicting the output answer quality level of a user at a given excitation value, and consists of a plurality of fully connected layers, and receives the output of the last time node at each time node to form a cyclic neural network.

4. A dynamic incentive computation method for a crowdsourcing system as claimed in claim 1 wherein: the main parameters of the cyclic neural network model are as follows:

1) The demander can decide whether to give the user an incentive value at each time node t, denoted as a _t ，a _t Giving an excitation value of 1, and judging whether 0 is negative;

2) The output of the cyclic neural network model is the variable y _t Representing the probability of the user completing the task to high quality, y _t The closer to 1 the higher the quality of the answer, the closer to 0 the lower the quality of the answer;

3) There are multiple hidden states, and the transmission parameters between the input and hidden states, between hidden states and output in the neural network are obtained by training a cyclic neural network model.

5. The method of dynamic incentive computation for a crowdsourcing system of claim 4 wherein: the construction steps of the cyclic neural network model are as follows:

the training data set is a historical task question-answer data sequence of the user<a _t ,y _t >And updating and optimizing model parameters through back propagation of the cyclic neural network, so that the performance of the model on a training data set is optimized, and a required answer quality evaluation model can be obtained after repeated iterative training.

6. A dynamic incentive computation method for a crowdsourcing system as claimed in claim 1 wherein: and solving the corresponding final benefit function by adopting a heuristic pruning strategy.

7. A dynamic incentive computing system for a crowdsourcing system, comprising:

the acquisition module is used for acquiring task data of a demander on the crowdsourcing platform and historical task question-answer data of a participating user;

the allocation module is used for allocating the tasks to the participating users;

the model construction module is used for constructing a cyclic neural network model for each participating user;

the model training module is used for training a cyclic neural network model according to historical task question-answer data of the participating user;

the decision module is used for calculating the final benefit brought by different excitation values according to the prediction results of the participating users, the tasks and the cyclic neural network model and judging whether the current user is given an excitation value or not;

the result output module is used for collecting answers of all the participating users so as to obtain a result required by a demander;

according to the prediction results of the participating users, tasks and the cyclic neural network, calculating the final benefit size brought by different excitation values to judge whether the current user is given with the excitation value, including:

for a certain user, giving a corresponding excitation value as input of a cyclic neural network model, and obtaining the predicted answer quality corresponding to the excitation value;

constructing a final benefit function through the predicted answer quality, wherein the final benefit function comprises the benefit for completing the next task and the benefit obtained by completing the future task;

the final benefit obtained by the corresponding incentive value is obtained by solving the corresponding final benefit function, thereby deciding whether to give the current user incentive value.

8. An electronic device, comprising:

one or more processors;

a memory for storing one or more programs;

the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1-6.

9. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method according to any of claims 1-6.

Images