CN109523237A

CN109523237A - Crowdsourcing task method for pushing and relevant apparatus based on user preference

Info

Publication number: CN109523237A
Application number: CN201811364711.1A
Authority: CN
Inventors: 黄夕桐; 李佳琳; 王健宗; 肖京
Original assignee: Ping An Technology Shenzhen Co Ltd
Current assignee: Ping An Technology Shenzhen Co Ltd
Priority date: 2018-11-15
Filing date: 2018-11-15
Publication date: 2019-03-26
Anticipated expiration: 2038-11-15
Also published as: CN109523237B; WO2020098251A1

Abstract

The embodiment of the invention discloses crowdsourcing task method for pushing and relevant apparatus based on user preference, wherein the described method includes: determining the task response any active ues of the first preset quantity for the task of goal task type；The history propelling data collection of each task response any active ues is obtained respectively；According to history propelling data collection, acceptance probability Logic Regression Models are established；Using acceptance probability Logic Regression Models, according to any active ues data and historic task data estimation historic task acceptance probability；According to historic task acceptance probability and response results label data, estimation tasks respond user's likelihood degree of any active ues；The corresponding user's likelihood degree of any active ues is responded according to the task of the first preset quantity and determines the first high preference target user from the task of the first preset quantity response any active ues, and the target of goal task type push task is pushed to the first high preference target user.Using the solution of the present invention, the delivery efficiency of crowdsourcing task can be improved.

Description

Crowdsourcing task pushing method based on user preference and related device

Technical Field

The invention relates to the field of computers, in particular to a crowdsourcing task pushing method and a related device based on user preference.

Background

In recent years, more and more enterprises have begun to attempt to delegate certain technical work tasks to external individuals or organizations through internet channels, and this emerging internet-based open collaborative innovation model is referred to as crowdsourcing. The crowdsourcing mode generally includes three bodies: the method comprises the steps that a subcontracting party, a crowdsourcing platform and a subcontracting party generally release crowdsourcing tasks on the crowdsourcing platform, and then the subcontracting party receives the tasks from the crowdsourcing platform, completes the tasks according to convention and obtains corresponding rewards. Common crowdsourcing tasks include design, management consultation, project planning, movie production, information collection, picture recognition, software development, and the like. When the crowdsourcing task is distributed, the crowdsourcing platform widely spreads the task requirement of a publisher, and meanwhile, talents with capabilities, interests and the like matched with the corresponding crowdsourcing task are found on the Internet for distribution, so that the crowdsourcing task can be smoothly and efficiently completed.

However, in the current crowdsourcing job scenario, when the task is distributed in most cases, the preference of the individual subcontractor for the task is not considered, so that the problems that after part of the task is dispatched, the user does not timely accept the task or does not accept the task, and the task needs to be redistributed are caused. The user has low response degree to the task, and the completion degree and the delivery efficiency of the task are directly influenced.

Disclosure of Invention

The invention provides a crowdsourcing task pushing method and a related device based on user preference, which can improve the delivery efficiency of crowdsourcing tasks.

In a first aspect, the present invention provides a method for pushing a crowdsourcing task based on user preferences, including:

determining a first preset number of task response active users aiming at the task of the target task type;

respectively acquiring a historical push data set of each task response active user, wherein the historical push data set comprises a second preset amount of historical push data, and the historical push data comprises active user data generated by historical push tasks of the types of target tasks pushed by the task response active users, historical task data corresponding to the historical push tasks and response result label data of whether the historical push tasks are accepted by the task response active users or not;

establishing a logical regression model of the receiving probability of the historical tasks relative to the historical task data and the active user data according to the respective historical pushing data sets of the first preset number of the task response active users, wherein the receiving probability of the historical tasks is the receiving probability of the task response active users to each historical pushing task;

estimating the historical task acceptance probability according to the active user data and the historical task data by utilizing the acceptance probability logistic regression model;

estimating the user likelihood degree of the response result corresponding to the historical pushing task and the response result label data of the historical pushing task of the task response active user according to the historical task receiving probability and the response result label data;

and determining a first high-preference target user from the first preset number of task response active users according to the user likelihood degree corresponding to the first preset number of task response active users, and pushing a target pushing task of a target task type to the first high-preference target user.

With reference to the first aspect, in a possible implementation manner, the historical task data includes at least one task attribute feature data of the historical push task, and the active user data includes at least one user attribute feature data of the task responding to an active user;

the step of establishing a logical regression model of the acceptance probability of the historical tasks with respect to the historical task data and the active user data according to the respective historical pushed data sets of the first preset number of the tasks in response to the active users comprises the following steps:

if the second preset number is m, and m is a positive integer, using the event A₁Indicating that the task responds to the active user u to accept the ith history push task in the m history push tasks, wherein i is a positive integer and belongs to [1, m ]]If the task response active user u receives the history task of the ith history push task, the probability that the history task is received by the task response active user u to the ith history push task is as follows:

wherein, the vector x_iA task characteristic value vector formed by task attribute characteristic data corresponding to the ith historical push task; vector w_uAnd counting user characteristic weight vectors formed by the task response active users u aiming at the user attribute characteristic data of the ith historical pushing task according to the user characteristic weight of each user attribute characteristic data in the user characteristic weight vectors and the response result label data of the historical pushing data sets corresponding to all the task response active users in the first preset number.

With reference to the first aspect, in a possible implementation, the estimating, according to the historical task acceptance probability and the response result tag data, a user likelihood degree of a response result corresponding to the historical push task and the response result tag data of the historical push task by the task response active user includes:

if the second preset number is m, m is a positive integer, a sampling sample set of the task response active user u is formed by the m historical pushing tasks pushed by the task response active user u and response results corresponding to the response result label data of the m historical pushing tasks, and the likelihood degree of the task response active user u with respect to the sampling sample set is as follows:

wherein ,P_u(A₁|x_uj) Representing the historical task acceptance probability of the j-th historical push task in the m historical push tasks by the task response active user u; y is_ujResponding response result label data of the active user u to the jth historical pushing task in the m historical pushing tasks for the task, wherein j is a positive integer, and j belongs to [1, m ∈]，y_uj∈{0,1}。

With reference to the first aspect, in one possible implementation, the determining, according to the user likelihood degrees corresponding to the first preset number of task response active users, a first high-preference target user from among the first preset number of task response active users includes:

sequencing the task response active users of the first preset number from high to low according to the likelihood degree of the corresponding users;

and determining a third preset number of task response active users which are ranked most front as first high-preference target users, wherein the third preset number is less than or equal to the first preset number and is greater than or equal to the number of target pushing tasks.

With reference to the first aspect, in one possible implementation, the method further includes:

deleting the first high-preference target users from the first preset number of task response active users, and determining the task response active users after deleting the first high-preference target users as secondary screening active users;

when a task rejection instruction of the first high-preference target user for the target pushing task is received, determining a second high-preference target user from the secondary screening active users according to the user likelihood degree corresponding to each secondary screening active user;

and pushing the target pushing task corresponding to the task refusing instruction to the second high-preference target user.

With reference to the first aspect, in one possible implementation, the pushing a target push task of a target task type to the first high-preference target user includes:

determining high preference acceptance time data of the first high preference target user according to the historical push data set of the first high preference target user;

and pushing the target pushing task of the target task type to the first high-preference target user at the moment corresponding to the high-preference receiving time data.

In a second aspect, an embodiment of the present invention provides a crowdsourcing task pushing apparatus based on user preferences, including:

the active user determining unit is used for determining a first preset number of task response active users aiming at the task of the target task type;

the data set acquisition unit is used for respectively acquiring historical push data sets of all task response active users, wherein the historical push data sets comprise a second preset amount of historical push data, and the historical push data comprise active user data generated by historical push tasks of target task types pushed by the task response active users, historical task data corresponding to the historical push tasks and response result label data of whether the historical push tasks are accepted by the task response active users or not;

the model establishing unit is used for establishing a logical regression model of the receiving probability of the historical tasks relative to the historical task data and the active user data according to the respective historical pushing data sets of the first preset number of the task response active users, wherein the receiving probability of the historical tasks is the receiving probability of the task response active users to each historical pushing task;

the estimation unit is used for estimating the historical task acceptance probability according to the active user data and the historical task data by utilizing the acceptance probability logistic regression model;

the estimation unit is further configured to estimate, according to the historical task acceptance probability and the response result tag data, a user likelihood degree of the response result corresponding to the historical push task and the response result tag data of the historical push task for the task response active user;

and the task pushing unit is used for determining a first high-preference target user from the first preset number of task response active users according to the user likelihood degree corresponding to the first preset number of task response active users and pushing a target pushing task of a target task type to the first high-preference target user.

With reference to the second aspect, in a possible implementation manner, the apparatus further includes:

the secondary screening unit is used for deleting the first high-preference target users from the first preset number of task response active users and determining the task response active users after the first high-preference target users are deleted as secondary screening active users;

the secondary screening unit is further configured to determine a second high-preference target user from the secondary screening active users according to user likelihood degrees corresponding to the respective secondary screening active users when a task rejection instruction of the first high-preference target user for a target push task is received;

the task pushing unit is further configured to push a target pushing task corresponding to the task rejection instruction to the second high-preference target user.

In a third aspect, an embodiment of the present invention provides another crowdsourcing task pushing device based on user preferences, including a processor, a memory, and a communication interface, where the processor, the memory, and the communication interface are connected to each other, where the communication interface is configured to receive and send data, the memory is configured to store program codes, and the processor is configured to call the program codes to execute any one of the foregoing first aspect and various possible implementation manners of the first aspect.

In a fourth aspect, an embodiment of the present invention provides a computer storage medium, where a computer program is stored, where the computer program includes program instructions, and the program instructions, when executed by a computer, cause the computer to perform any one of the above first aspect and each possible implementation manner of the first aspect.

In the embodiment of the invention, a logical regression model of historical task acceptance probability related to historical task data and active user data is established according to a historical pushing data set of each task response active user, a user likelihood degree of each task response active user related to respective historical pushing task and respective response result label data corresponding to respective response result label data is respectively estimated according to the historical task acceptance probability estimated by the logical regression model of the acceptance probability and the response result label data of each task response active user to respective historical pushing task, and then a target pushing task of a target task type is pushed to a first high-preference target user determined from the task response active users according to the user likelihood degree, so that a differential pushing crowdsourcing task can be carried out according to the preference of the user, the response degree of the user to the crowdsourcing task is improved, and the delivery efficiency of the crowdsourcing task is further improved.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.

Fig. 1 is a schematic flowchart of a crowdsourcing task pushing method based on user preferences according to an embodiment of the present invention;

fig. 2 is a schematic flowchart of another crowdsourcing task pushing method based on user preferences according to an embodiment of the present invention;

fig. 3 is a schematic structural diagram of a crowdsourcing task pushing device based on user preferences according to an embodiment of the present invention;

fig. 4 is a schematic structural diagram of another crowdsourcing task pushing device based on user preferences according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Referring to fig. 1, fig. 1 is a schematic flowchart of a crowdsourcing task pushing method based on user preferences according to an embodiment of the present invention, as shown in the figure, the method includes:

s101, determining a first preset number of task response active users aiming at the task of the target task type.

Specifically, the task mainly refers to a crowdsourcing task appearing on the internet, and relates to the fields of information retrieval, artificial intelligence, video analysis, knowledge mining, image quality assessment and the like. The task type of the task may be various, for example, a semantic judgment type, a voice recognition type, an information collection type, and the like, and the target task type may be one of them. The first preset number can be adjusted according to different tasks of the target task type, and it can be understood that the larger the first preset number is, that is, the larger the number of sampled task response active users is, the more accurate the first high-preference target user of the task of the target task type determined by the crowd-sourced task pushing method based on user preference is.

Optionally, the first preset number of task response active users may be determined by:

11) and respectively acquiring all historical users of the tasks aiming at the target task type.

12) And respectively calculating the target type task receiving proportion of each historical user, wherein the target type task receiving proportion is the ratio of the receiving times of each historical user to the task of the target task type to the times of pushing the target task type.

13) And sequencing the historical users according to the sequence of the target type task acceptance proportion from high to low, and determining a first preset number of historical users in the front of the sequence as the active users of the task response.

And S102, respectively acquiring historical pushing data sets of the users with active task responses.

Specifically, for the first number of task response active users determined in step S101, a history push data set of each task response active user is respectively collected, where the history push data set includes a second preset number of history push data, and the history push data includes active user data generated by a history push task of a type that the task response active user is pushed by a target task, history task data corresponding to the history push task, and response result tag data indicating whether the history push task is accepted by the task response active user, that is, a group of history push data is generated every time the task response active user pushes one history push task. It can be understood that the larger the second preset number is, that is, the more historical push data sampled by responding to active users for each task, the more accurate the first high-preference target user of the task of the target task type determined by the crowd-sourced task push method based on user preference is.

Optionally, the history push information sets of the users with active task responses are respectively obtained, and each history push information in the history push information sets is quantized to obtain the history push data set corresponding to the users with active task responses. For example, if one of the acquired historical push information sets of the user a responding to the active task is 17 points of the task receiving time, the historical push information may be quantized into the historical push data 17. For another example, one historical push information in the obtained historical push information set of the user B with the active task response is that the task receiving place is home, and in this crowdsourcing task push based on the user preference, home, company, commute, outdoor, and others in the task receiving place are respectively corresponding to the labels 1, 2,3, 4, and 5, and then the historical push information can be quantized into the historical push data 1.

S103, according to the respective historical pushed data sets of the first preset number of the tasks responding to the active users, a logical regression model of the historical task acceptance probability with respect to the historical task data and the active user data is established.

Here, the historical task acceptance probability is the acceptance probability of the task response active user to each historical push task. The acceptance probability logistic regression model is established according to historical pushing data sets of all the first preset number of task response active users, and the acceptance probability logistic regression model can be used for estimating the historical task acceptance probability of any one task response active user to any one historical pushing task.

Optionally, the historical task data includes at least one task attribute feature data of the historical push task, and the active user data includes at least one user attribute feature data of the task responding to the active user. If the second preset number is m, and m is a positive integer, using the event A₁Indicating that the task responds to the active user u to accept the ith history push task in the m history push tasks, wherein i is a positive integer and belongs to [1, m ]]Then, the historical task acceptance probability of the task response active user u to the ith historical pushing task is expressed by formula (1):

And S104, estimating the historical task acceptance probability according to the active user data and the historical task data by using the acceptance probability logistic regression model.

Specifically, by using the acceptance probability logistic regression model, the historical task acceptance probability of each historical active user for the historical push task when the historical active user is pushed by the historical active user when the historical push task is pushed by the historical active user is estimated according to the active user data and the historical task data included in the historical push data set of each active user responded by the first preset number of tasks.

For example, the historical push data set of the user a includes two sets of historical push data, the two sets of historical push data correspond to the historical push task of the target task type pushed twice by the user a, the user a generates active user data 1 when being pushed a first historical push task, the historical task data of the first historical push task is historical task data 1, the user a generates active user data 2 when being pushed a second historical push task, and the historical task data of the second historical push task is historical task data 2, then the step estimates, by using the acceptance probability logistic regression model, the historical task acceptance probability of the first historical push task when the user a is pushed the first historical push task once according to the active user data 1 and the historical task data 1, and estimates, according to the active user data 2 and the historical task data 2, the historical push task of the user a is pushed the second historical push task, a historical task acceptance probability for the second historical push task.

And S105, estimating the user likelihood degree of the task response active user about the historical pushing task and the response result corresponding to the response result label data of the historical pushing task according to the historical task receiving probability and the response result label data.

Specifically, according to the historical task acceptance probability of the task response active user to the historical push task that the task response active user has been pushed by the user when the task response active user is a historical push task that the user has been pushed by the user, and the response result label data of the historical push task, the user likelihood of the task response active user is estimated by using a likelihood function, where the user likelihood is the likelihood of the task response active user with respect to a sampling sample set of the user, and the sampling sample set includes the historical push tasks corresponding to all the historical task data included in the historical push data set of the task response active user and the response results corresponding to all the response result label data included in the historical push data set of the task response active user. Because the sampling sample set is a more active sampling sample set for the task response active user, that is, the number of history push tasks in the sampling sample set that have been accepted by the corresponding history active user is large, the user likelihood of the task response active user with respect to the sampling sample set can be used to approximate the preference degree of the task response active user for the task of the target task type.

Optionally, if the second preset number is m, and m is a positive integer, a sampling sample set of the task response active user u is formed for m historical pushing tasks pushed by the task response active user u and response results corresponding to response result tag data of the m historical pushing tasks, and a likelihood degree of the task response active user u with respect to the sampling sample set is represented by a formula (2):

S106, determining a first high-preference target user from the first preset number of task response active users according to the user likelihood degree corresponding to the first preset number of task response active users, and pushing a target pushing task of a target task type to the first high-preference target user.

Specifically, since the user likelihood degree may represent a preference degree of the task response active user for the task of the target task type, the task response active user with the higher user likelihood degree estimated in step S105 has a higher preference degree for the task of the target task type, so that a certain number or a certain proportion of task response active users with the highest user likelihood degree may be determined as the first high-preference target user, and when a target push task of the target task type is to be pushed, the target push task may be pushed to the first high-preference target user.

Optionally, the specific step of determining the first high-preference target user from the first preset number of task response active users according to the user likelihood degree corresponding to each of the first preset number of task response active users may be as follows:

21) and sequencing the active users responding to the tasks with the first preset number from high to low according to the likelihood degree of the corresponding users.

22) And determining a third preset number of task response active users with the most front ranking as first high-preference target users, wherein the third preset number is less than or equal to the first preset number and is greater than or equal to the number of target pushing tasks.

In an alternative embodiment, the method may further comprise the steps of:

31) and deleting the first high-preference target users from the first preset number of task response active users, and determining the task response active users after the first high-preference target users are deleted as secondary screening active users.

32) And when a task rejection instruction of the first high-preference target user for the target pushing task is received, determining a second high-preference target user from the secondary screening active users according to the user likelihood degree corresponding to each secondary screening active user.

33) And pushing the target pushing task corresponding to the task refusing instruction to the second high-preference target user.

Through the above steps 31), 32) and 33), when there is a user who sends a task rejection instruction among the first high-preference target users who have pushed the target push task, a second high-preference target user with a higher preference degree can be selected from users other than the first high-preference target user among the active users responding to the task, and the target push task corresponding to the task rejection instruction is pushed to the second high-preference target user, so that it is ensured that the target push task corresponding to the task rejection instruction is pushed by the user with the higher preference degree, and the first high-preference target user is prevented from being repeatedly pushed the target push task corresponding to the target task type, so that the preference degree of the first high-preference target user for the task of the target task type is reduced, and the exposure degree of the task of the target task type is increased, and pushing more second high-preference target users except the first high-preference target user to the task of the target task type, and improving the adhesion degree of the second high-preference target users and the task of the target task type.

In the embodiment of the invention, a logical regression model of the acceptance probability of the historical task related to the historical task data and the active user data is established according to the historical push data set of each task responding to the active user, respectively estimating the user likelihood degree of each task response active user about the respective historical push task and the corresponding response result of each response result label data according to the historical task receiving probability estimated by the receiving probability logistic regression model and the response result label data of each task response active user to the respective historical push task, since the user likelihood may represent a degree of preference of the task to respond to active users for tasks of the target task type, the first high-preference target user may be determined based on the user likelihood, and then pushing the target pushing task of the target task type to the first high-preference target user. The method and the device have the advantages that the crowdsourcing task can be pushed in a differentiated mode according to the preference of the user, the response degree of the user to the crowdsourcing task is improved, and the delivery efficiency of the crowdsourcing task is improved.

Referring to fig. 2, fig. 2 is a schematic flowchart of another crowdsourcing task pushing method based on user preferences according to an embodiment of the present invention, as shown in the figure, the method includes:

s201, all historical users of the tasks of the target task types are respectively obtained.

For example, if the target task type is an information collection type, all historical users pushed by the information collection type task are acquired.

S202, respectively calculating the target type task receiving proportion of each historical user, sequencing the historical users from high to low according to the respective target type task receiving proportions, and determining the historical users with the first preset number in the front sequencing as the active users for task response.

Here, the target type task acceptance ratio is a ratio of the number of times each historical user accepts a task of the target task type to the number of times the task is pushed by the historical user. For example, assuming that there are 100 history users determined in step S201 for the tasks of the information collection class, the number of times that one user a of the 100 history users has been pushed with the tasks of the information collection class is 15, and the number of times that the user a receives the pushed tasks of the information collection class is 10, the target type task acceptance ratio of the user a is 10/15-2/3, the target type task acceptance ratios of the 100 history users are sequentially calculated according to the above calculation method, and if the first preset number is 50, the 50 users with the highest target type task acceptance ratio of the 100 history users are determined as the active users responding to the tasks.

S203, respectively acquiring the historical pushing information sets of the task response active users, and carrying out quantization processing on each historical pushing information in the historical pushing information sets to obtain the historical pushing data sets corresponding to the task response active users.

Specifically, the history push information set of each task response active user includes a second preset number of groups of history push information, and each group of history push information includes active user information generated by a history push task of a type that the task response active user is pushed to a target task, history task information corresponding to the history push task, and a response result indicating whether the history push task is accepted by the task response active user. The historical task information comprises at least one task attribute characteristic information of the historical pushing task, and the active user information comprises at least one user attribute characteristic information of the task responding to the active user. And quantizing each task attribute feature information and each user attribute feature information to further finish the quantization processing of the historical push information set.

For example, if the target task type is an information collection type, the first preset number is 50, the second preset number is 30, and the 50 tasks obtained in step S203 respond to the history push information set of the user B in the active user, as shown in table 1:

TABLE 1

The collection time, the collection channel and the information type in table 1 are 3 task characteristic attribute information included by historical task information, and age, place when pushed and 3 user attribute information included by proficiency for active user information, and the above task characteristic attribute information and the user attribute information are respectively quantized, and the collection time is converted into a number corresponding to time, and the collection channel is quantized into: telephone collection, internet collection and field collection correspond to numbers 1, 2 and 3, respectively, and information types are quantized as follows: entertainment type, diet type and medical type correspond number 1, 2 and 3 respectively, quantify age with the figure that the age corresponds, quantify the place when being pushed place as: home, outdoor and company correspond to numbers 1, 2 and 3, respectively, and proficiency is quantified as: very skilled, more skilled, less skilled, and unskilled numbers 5, 4, 3, 2, and 1 respectively, quantize the response result acceptance as response result tag 1, and quantize the response result rejection or omission as response result tag 0, resulting in the historical pushed data set for user B, as shown in table 2:

TABLE 2

And S204, counting the user characteristic weight of each user attribute characteristic data in a user characteristic weight vector consisting of the user attribute characteristic data according to the user attribute characteristic data and response result label data contained in the respective historical push data sets of all the first preset number of the task response active users.

Specifically, the user characteristic weight represents an influence degree of corresponding user attribute characteristic data on the task response active user to receive the history push task, and the user characteristic weight may be calculated by statistical software such as SPSS (statistical product and Service Solutions) or SAS (statistical analysis System).

Optionally, in order to improve the calculation efficiency and reduce the calculation cost, the user attribute feature data with a smaller absolute value of the corresponding user feature weight in the user attribute feature data may be discarded, and the user attribute feature data with a larger user feature weight may be screened out to participate in the calculation. For example, if the statistical software SPSS is used to statistically calculate the user feature weights of gender, age, location and proficiency at which the user is pushed: 0.002, 0.335, 0.114 and 0.219, the user attribute feature data corresponding to gender may be discarded without participating in the calculation.

S205, according to the user characteristic weight, establishing a logical regression model of the acceptance probability of the historical task with respect to the historical task data and the active user data.

Based on the example in step S203, if the user feature weights of the age, the location when pushed, and the proficiency obtained in step S204 are: 0.335, 0.114 and 0.219, defining the collection time, collection channel, collection type are respectively represented by variables t, c and k, age, location when pushed, proficiency are respectively represented by variables a, s and l, and event A is represented by₁Indicating that the u-th user among the 50 task response active users receives the ith historical push task among the 30 historical push tasks, the historical task acceptance probability of the u-th user on the ith historical push task is expressed by formula (3):

wherein, the vector x_iTask attribute characteristic data acquisition time t corresponding to ith history push task_iAnd a collection channel c_iInformation type k_iConstructed vector of task eigenvalues, vector x_i＝(t_i,c_i,k_i) (ii) a Vector w_uUser attribute feature data age a for ith user for ith history push task_uThe place s when pushed_uProficiency level l_uFormed user feature weight vector, vector w_u＝(0.335*a_u,0.114*s_u,0.219*l_u)。

S206, estimating the historical task acceptance probability according to the active user data and the historical task data by using the acceptance probability logistic regression model.

Here, the historical task acceptance probability is a probability that the task responds to a historical push task corresponding to the active user and accepts the historical push task corresponding to the historical task data when the historical push task corresponding to the historical task data is pushed once.

Based on the example in step S203, if the user feature weights of the age, the location when pushed, and the proficiency calculated by the statistical software SPSS in step S204 are: 0.335, 0.114 and 0.219, according to the history push data set corresponding to table 2, the task characteristic value vector of the history push task with the label of 3 is x₃That is, (12,3,3), the user feature weight vector of the user B for the history push task with the index 3 is w_B(0.335 × 20,0.114 × 3,0.219 × 4), x₃ and w_BThe historical task acceptance probability P of the historical push task with the label 3 when the user B is pushed the historical push task with the label 3 once is calculated by substituting the formula (3) in the step S205_B(A₁|x₃). According to the method, when history push tasks with the labels of 1, 2,3, … and 30 are pushed by the user B respectively, the history task acceptance probability P of the history push tasks with the labels of 1, 2,3, … and 30 can be estimated in sequence_B(A₁|x₁)、P_B(A₁|x₂)、P_B(A₁|x₃)、…、P_B(A₁|x₃₀). And then the historical task acceptance probability of each task responding to the corresponding historical push task when the active user is pushed by the corresponding historical push task once can be estimated.

And S207, estimating the user likelihood degree of the task response active user about the historical pushing task and the response result corresponding to the response result label data of the historical pushing task according to the historical task receiving probability and the response result label data.

Based on the example in step S203, if the history push tasks corresponding to the respective push labels 1, 2,3, …, and 30 of the user B are obtained in step S206, the history task acceptance probability P for the history push tasks corresponding to the respective push labels 1, 2,3, …, and 30 is obtained_B(A₁|x₁)、P_B(A₁|x₂)、P_B(A₁|x₃)、…、P_B(A₁|x₃₀) The sampling sample set of the user B is [ (history push task corresponding to reference numeral 1, accept), (history push task corresponding to reference numeral 2, reject), (history push task corresponding to reference numeral 3, accept), (history push task corresponding to reference numeral 4, ignore), … …, (history push task corresponding to reference numeral 30, accept)]The user likelihood of the user B to the self sampling sample set can be calculated by using formula (4):

And sequentially calculating the likelihood degree of the self sampling sample set of 50 active task response users according to the method.

S208, the task response active users with the first preset number are ranked from high to low according to the corresponding user likelihood degrees, and the task response active users with the third preset number which are ranked most front are determined as first high preference target users.

Here, the third preset number is the number of the first high-preference target users, the value of the third preset number may be a preset fixed value, or may be a fixed percentage of the value of the first preset number, the fixed value and the fixed percentage may be set differently according to different attribute information of the target push task, such as an emergency, but the fixed value or the fixed percentage is set to ensure that the third preset number is less than or equal to the first preset number and is greater than or equal to the number of the target push task.

For example, if 50 task response active users in step S207 are ranked from high to low according to their respective user likelihood degrees, the resulting sequence is shown in table 3:

sequencing label	User' s	Degree of user likelihood
			1	User 20	0.0835
2	User 8	0.0762
			3	User 44	0.0754
4	User 6	0.0733
			5	User 29	0.0689
6	User 12	0.0687
			7	User 3	0.0644
8	User 1	0.0621
			9	User 46	0.0584
……	……	……
			50	User 39	0.0014

TABLE 3

If the third preset number is 6, determining the user 20, the user 8, the user 44, the user 6, the user 29, and the user 12 as the first high-preference target user.

S209, according to the historical push data set of the first high preference target user, determining high preference acceptance time data of the first high preference target user, and pushing the target push task of the target task type to the first high preference target user at a moment corresponding to the high preference acceptance time data.

Specifically, the time corresponding to the high-preference receiving time data is the time when the first high-preference target user has the maximum possibility of receiving the target pushing task in 24 hours of a day, and the pushing is performed at the time when the first high-preference target user has the maximum possibility of receiving the target pushing task, so that the possibility of receiving the target pushing task by the first high-preference target user can be further improved, and the delivery efficiency of the target pushing task is further improved.

Here, there may be a plurality of methods for determining high preference acceptance time data common to one of the first high preference target users, and collectively pushing the target push task to the first high preference target users at a time corresponding to the high preference acceptance time, or determining high preference acceptance time data of each of the first high preference target users, and pushing the target push task to the corresponding first high preference target users at a time corresponding to the high preference acceptance time data of each of the first high preference target users, respectively, and further determining the high preference acceptance time data of the first high preference target users.

In an alternative embodiment, the statistical calculation may be performed on the respective historical push data sets of all the first high-preference target users to obtain one acceptance peak time period with the largest number of accepted historical push tasks pushed at 0 point-1 point, 1 point-2 point, 2 point-3 point, … point, 23 point-24 point, and the acceptance peak time period may be determined as the high-preference acceptance time data, for example, if the statistical calculation results in the largest number of accepted historical push tasks pushed within a time period from 17 point to 18 point, the target push tasks may be pushed to the first high-preference target users at random time points between 17 point and 18 point.

Or respectively counting the historical push data sets of the first high-preference target users to obtain the peak receiving time periods of the first high-preference target users, and pushing the target push tasks to the corresponding first high-preference target users in the peak receiving time periods of the first high-preference target users.

In another optional implementation manner, a highest fourth preset number of historical task acceptance probabilities among the historical task acceptance probabilities of all first high-preference target users are obtained, the pushing time data of the historical pushing tasks corresponding to the highest fourth preset number of historical task acceptance probabilities is determined as the high-preference acceptance time data, the target pushing tasks of the target task type are allocated to fourth preset number of pushing batches, and the target pushing tasks of the corresponding pushing batches are respectively pushed to the first high-preference target users at times corresponding to the fourth preset number of high-preference acceptance time data. The push time data may be present in active user data included in the historical push data set, or may be present in historical task data included in the historical push data set.

For example, if the fourth preset number is 3, the highest 3 historical task acceptance probabilities among the historical task acceptance probabilities of all the first high-preference target users are the historical task acceptance probabilities P of the user a for the 3 rd historical pushing task of the target task type that has been pushed by the user a_A(A₁|x₃) The historical task acceptance probability P of the user M to the 12 th historical pushing task of the target task type which is pushed by the user M_M(A₁|x₁₂) The historical task acceptance probability P of the user G to the 29 th historical pushing task of the target task type which is pushed by the user G_G(A₁|x₂₉) Further, it may be determined that a time when the user a has been pushed the 3 rd historical push task is 17 points according to the historical push data set of the user a, that a time when the user M has been pushed the 12 th historical push task is 11 points according to the historical push data set of the user M, that a time when the user G has been pushed the 29 th historical push task is 15 points according to the historical push data set of the user G, and if there are 15 target push tasks, randomly distributing the 15 target push tasks to 3 push batches, for example, randomly distributing 5 target push tasks to each push batch respectively, and then pushing the target push tasks of the 3 push batches to the first high-preference target user at 11 points, 15 points, and 17 points respectively.

The highest historical task acceptance probability of each first high-preference target user can also be respectively obtained, the pushing time of the historical pushing task corresponding to the highest historical task acceptance probability of each first high-preference target user is further determined, and the target pushing task is pushed to the first high-preference target user at the pushing time of the historical pushing task corresponding to the highest historical task acceptance probability of each first high-preference target user.

Optionally, if a task rejection instruction of the first high-preference target user for the target push task is received, a second high-preference target user is determined for the target push task corresponding to the task rejection instruction, high-preference acceptance time data of the second high-preference target user is further determined, and the target push task corresponding to the task rejection instruction is pushed to the second high-preference target user at a time corresponding to the high-preference acceptance time data of the second high-preference target user.

Based on the example in step S208, after the target push tasks of 6 target task types are pushed to the users 20, 8, 44, 6, 29, and 12, and the task rejection instructions of the users 44 and 12 are received, the users 3 and 1 may be determined as second high-preference target users, and then the highest historical task acceptance probabilities of the users 3 and 1 are further obtained as the historical task acceptance probability of the user 3 for the 18 th historical push task of the target task type that has been pushed by the user 3 and the historical task acceptance probability of the user 1 for the 5 th historical push task of the target task type that has been pushed by the user 1, respectively, a time when the user 3 has pushed the 18 th historical push task is 3 points, a time when the user 1 has pushed the 5 th historical push task is 10 points, historical push tasks corresponding to the task rejection instructions sent by the users 44 and 12 are pushed to the user 3 and the user 1 at 3 and 10 points, respectively.

In the embodiment of the invention, firstly, a task response active user is determined from historical users of tasks of the target task type, then, a historical pushing data set of the task response active user is obtained by quantizing the historical pushing information set of the task response active user, the user likelihood degree of the task response active user is estimated according to the historical pushing data set and an established acceptance probability logistic regression model, a first high-preference target user is determined from the task response active user according to the user likelihood degree, and the target pushing task of the target type is pushed to the first high-preference target user at the moment corresponding to the high-preference acceptance time data of the first high-preference target user. The method and the device have the advantages that the crowdsourcing task can be pushed differentially according to the preference of the user, the task can be pushed at the moment when the user has high possibility of receiving the task, the response degree of the user to the crowdsourcing task is further improved, and the delivery efficiency of the crowdsourcing task is improved.

Referring to fig. 3, fig. 3 is a schematic structural diagram of a crowdsourcing task pushing device based on user preferences according to an embodiment of the present invention, as shown in the figure, the crowdsourcing task pushing device 30 based on user preferences may include:

an active user determining unit 301, configured to determine a first preset number of task responses to active users for the task of the target task type.

Specifically, the task mainly refers to a crowdsourcing task appearing on the internet, and relates to the fields of information retrieval, artificial intelligence, video analysis, knowledge mining, image quality assessment and the like. The task type of the task may be various, for example, a semantic judgment type, a voice recognition type, an information collection type, and the like, and the target task type may be one of them. The first preset number can be adjusted according to different tasks of the target task type, and it can be understood that the larger the first preset number is, that is, the larger the number of sampled task response active users is, the more accurate the first high-preference target user of the task of the target task type determined by the crowdsourcing task pushing device based on user preference is.

Optionally, the active user determining unit 301 may be specifically configured to: respectively acquiring all historical users of the task aiming at the target task type; respectively calculating a target type task receiving proportion of each historical user, wherein the target type task receiving proportion is the ratio of the receiving times of each historical user to the task of the target task type to the times of pushing the target task type by the historical user; and sequencing the historical users from high to low according to the respective target type task acceptance proportion, and determining a first preset number of historical users in the front of the sequence as the active users of the task response.

The data set obtaining unit 302 is configured to obtain a history push data set of each task responding to an active user, where the history push data set includes a second preset number of history push data, and the history push data includes active user data generated by a history push task of a type of a target task pushed by the task responding to the active user, history task data corresponding to the history push task, and response result tag data indicating whether the history push task is accepted by the task responding to the active user.

Specifically, for a first number of task response active users determined by the active user determining unit 301, a history push data set of each task response active user is respectively collected, where the history push data set includes a second preset number of history push data, and the history push data includes active user data generated by a history push task of a type that the task response active user is pushed by a target task, history task data corresponding to the history push task, and response result tag data indicating whether the history push task is accepted by the task response active user, that is, a group of history push data is generated every time the task response active user pushes one history push task. It can be understood that the larger the second preset number is, that is, the more historical push data sampled in response to active users for each task, the more accurate the first high-preference target user of the task of the target task type is determined by the crowd-sourced task pushing device based on user preference.

Optionally, the data set obtaining unit 302 may specifically obtain history push information sets of users with active task responses respectively, and perform quantization processing on each history push information in the history push information sets to obtain a history push data set corresponding to each user with active task responses.

A model establishing unit 303, configured to establish a logistic regression model of the historical task acceptance probability with respect to the historical task data and the active user data according to the respective historical pushed data sets of the first preset number of task response active users.

Here, the historical task acceptance probability is the acceptance probability of the task response active user to each historical push task. The acceptance probability logistic regression model is built by the model building unit 303 according to the historical push data sets of all the first preset number of task response active users, and the acceptance probability logistic regression model can be used to estimate the historical task acceptance probability of any one task response active user to any one historical push task.

Optionally, the historical task data includes at least one task attribute feature data of the historical push task, and the active user data includes at least one user attribute feature data of the task responding to the active user. If the second preset number is m, and m is a positive integer, using the event A₁Indicating that the task responds to the active user u to accept the ith history push task in the m history push tasks, wherein i is a positive integer and belongs to [1, m ]]Then, the model building unit 303 builds a logistic regression model of the acceptance probability, and the historical task acceptance probability of the task response active user u to the ith historical pushing task is represented by formula (5):

wherein, the vector x_iA task characteristic value vector formed by task attribute characteristic data corresponding to the ith historical push task; vector w_uResponding to a user feature weight vector formed by the active user u aiming at the user attribute feature data of the ith historical pushing task for the task, wherein the user feature weight of each user attribute feature data in the user feature weight vector, and the model establishing listThe element 303 is obtained by statistics according to the respective user attribute feature data of all the first preset number of the task response active users and the response result tag data of the history push data set corresponding to the task response active users.

An estimating unit 304, configured to estimate the historical task acceptance probability according to the active user data and the historical task data by using the acceptance probability logistic regression model.

Specifically, the estimating unit 304 estimates, by using the above acceptance probability logistic regression model, historical task acceptance probabilities for the historical push tasks when the historical push tasks that have been pushed by the historical active users are already pushed by the historical active users, according to the active user data and the historical task data included in the historical push data sets of the first preset number of task response active users.

The estimating unit 304 is further configured to estimate, according to the historical task acceptance probability and the response result label data, a user likelihood degree of a response result corresponding to the historical push task and the response result label data of the historical push task for the task response active user.

Specifically, the estimating unit 304 estimates, according to the estimated historical task acceptance probability of the task response active user for the historical push task that the task response active user has been pushed by the user at the time of the historical push task, and the response result label data of the historical push task, a user likelihood degree of the task response active user by using a likelihood function, where the user likelihood degree is a likelihood degree of the task response active user with respect to a self sampling sample set, and the sampling sample set includes the historical push tasks corresponding to all the historical task data included in the historical push data set of the task response active user and the response results corresponding to all the response result label data included in the historical push data set of the task response active user. Because the sampling sample set is a more active sampling sample set for the task response active user, that is, the number of history push tasks in the sampling sample set that have been accepted by the corresponding history active user is large, the user likelihood of the task response active user with respect to the sampling sample set can be used to approximate the preference degree of the task response active user for the task of the target task type.

Optionally, if the second preset number is m, where m is a positive integer, and a sampling sample set of the task response active user u is formed for m historical pushing tasks pushed by the task response active user u and response results corresponding to response result label data of the m historical pushing tasks, the estimating unit 304 expresses the likelihood of the task response active user u with respect to the sampling sample set by using a formula (6):

The task pushing unit 305 is configured to determine a first high-preference target user from the first preset number of task response active users according to the user likelihood degrees corresponding to the first preset number of task response active users, and push a target pushing task of a target task type to the first high-preference target user.

Specifically, since the user likelihood degree may represent a preference degree of the task response active user for the task of the target task type, the task response active user with the higher user likelihood degree estimated by the estimation unit 304 has a higher preference degree for the task of the target task type, so that a certain number or a certain proportion of task response active users with the highest user likelihood degree may be determined as the first high-preference target user, and when a target push task of the target task type is to be pushed, the target push task may be pushed to the first high-preference target user.

Optionally, the task pushing unit 305 may be specifically configured to: sequencing the task response active users of the first preset number from high to low according to the likelihood degree of the corresponding users; and determining a third preset number of task response active users which are ranked most front as first high-preference target users, wherein the third preset number is less than or equal to the first preset number and is greater than or equal to the number of target pushing tasks.

Further optionally, the crowd-sourced task pushing device 30 based on the user preference may further include:

the secondary screening unit 306 is configured to delete the first high-preference target user from the first preset number of task response active users, and determine the task response active user after the first high-preference target user is deleted as a secondary screening active user.

The secondary screening unit 306 is further configured to, when receiving a task rejection instruction of the first high-preference target user for the target push task, determine a second high-preference target user from the secondary screening active users according to user likelihood degrees corresponding to the respective secondary screening active users.

The task pushing unit 305 is further configured to push a target pushing task corresponding to the task rejecting instruction to the second high-preference target user.

When there are users who send task rejection instructions among the first high-preference target users who have pushed the target push tasks, the secondary screening unit 306 may screen out second high-preference target users with higher preference degrees from users other than the first high-preference target users among the task response active users, and the task pushing unit 305 pushes the target push tasks corresponding to the task rejection instructions to the second high-preference target users, so as to ensure that the target push tasks corresponding to the task rejection instructions are pushed by the users with higher preference degrees, and also prevent the first high-preference target users from repeatedly pushing the target push tasks corresponding to the target task types, so as to reduce the preference degree of the first high-preference target users for the tasks of the target task types, and also improve the exposure degree of the tasks of the target task types, and pushing more second high-preference target users except the first high-preference target user to the task of the target task type, and improving the adhesion degree of the second high-preference target users and the task of the target task type.

Further optionally, the task pushing unit 305 may be specifically configured to: determining high preference acceptance time data of the first high preference target user according to the historical push data set of the first high preference target user; and pushing the target pushing task of the target task type to the first high-preference target user at the moment corresponding to the high-preference receiving time data.

Specifically, the time corresponding to the high-preference receiving time data is the time when the first high-preference target user receives the target pushing task at the maximum probability within 24 hours of a day, and the task pushing unit 305 pushes the first high-preference target user at the time when the first high-preference target user receives the target pushing task at the maximum probability, so that the possibility that the first high-preference target user receives the target pushing task can be further improved, and the delivery efficiency of the target pushing task is further improved. Here, the task pushing unit 305 may determine high preference acceptance time data common to one first high preference target user, and collectively push the target pushing task to the first high preference target user at a time corresponding to the high preference acceptance time, or may determine respective high preference acceptance time data of each first high preference target user, and push the target pushing task to the corresponding first high preference target user at a time corresponding to the respective high preference acceptance time data of each first high preference target user.

In the embodiment of the invention, the model establishing unit establishes a logical regression model of the acceptance probability of the historical task with respect to the historical task data and the data of the active users according to the historical pushing data set of each task response active user, the estimating unit estimates the user likelihood degree of each task response active user with respect to each historical pushing task and the corresponding response result of each response active user with respect to each historical pushing task according to the historical task acceptance probability estimated by the logical regression model of the acceptance probability and the response result label data of each task response active user with respect to each historical pushing task, and since the user likelihood degree can represent the preference degree of the task response active user with respect to the task of the target task type, the first high preference target user can be determined according to the user likelihood degree, and then pushing the target pushing task of the target task type to the first high-preference target user. The method and the device have the advantages that the crowdsourcing task can be pushed in a differentiated mode according to the preference of the user, the response degree of the user to the crowdsourcing task is improved, and the delivery efficiency of the crowdsourcing task is improved.

Referring to fig. 4, fig. 4 is a schematic structural diagram of another crowdsourcing task pushing device based on user preferences according to an embodiment of the present invention, and as shown in the figure, the crowdsourcing task pushing device 40 based on user preferences includes a processor 401, a memory 402, and a communication interface 403. The processor 401 is connected to the memory 402 and the communication interface 403, for example, the processor 401 may be connected to the memory 402 and the communication interface 403 through a bus.

The processor 401 is configured to support the crowdsourcing task pushing device based on user preferences to perform corresponding functions in the crowdsourcing task pushing method based on user preferences described in fig. 1-2. The Processor 401 may be a Central Processing Unit (CPU), a Network Processor (NP), a hardware chip, or any combination thereof. The hardware chip may be an Application-Specific integrated circuit (ASIC), a Programmable Logic Device (PLD), or a combination thereof. The PLD may be a Complex Programmable Logic Device (CPLD), a Field-Programmable Gate Array (FPGA), General Array Logic (GAL), or any combination thereof.

The memory 402 is used to store program codes and the like. The memory 402 includes an internal memory that may include at least one of: volatile memory (e.g., Dynamic Random Access Memory (DRAM), Static RAM (SRAM), Synchronous Dynamic RAM (SDRAM), etc.) and non-volatile memory (e.g., one-time programmable read-only memory (OTPROM), Programmable ROM (PROM), Erasable Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM). memory 402 may also include external memory, which may include at least one of a Hard Disk (Hard Disk Drive, HDD) or a Solid-State Drive (SSD), flash drives, such as high-density flash (CF), Secure Digital (SD), micro SD, mini SD, extreme digital (xD), memory sticks, etc.

The communication interface 403 is used for receiving or transmitting data.

Processor 401 may call the program code to perform the following:

It should be noted that, the implementation of each operation may also correspond to the corresponding description of the method embodiment shown in fig. 1-2; the processor 401 may also be configured to perform other operations in the above-described method embodiments.

Embodiments of the present invention also provide a computer storage medium storing a computer program, the computer program comprising program instructions, which when executed by a computer, cause the computer to execute the method according to the foregoing embodiments, wherein the computer may be part of the above-mentioned crowd-sourced task pushing apparatus based on user preferences.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like.

The above disclosure is only for the purpose of illustrating the preferred embodiments of the present invention, and it is therefore to be understood that the invention is not limited by the scope of the appended claims.

Claims

1. A crowd-sourced task pushing method based on user preferences is characterized by comprising the following steps:

2. The method of claim 1, wherein the historical task data includes at least one task attribute feature data of the historical push task, and the active user data includes at least one user attribute feature data of the task response active user;

if the second preset number is m, and m is a positive integer, using the event A₁Indicating that the task response active user u accepts the ith history push task from the m history push tasks,i is a positive integer, i belongs to [1, m ]]If the task response active user u receives the history task of the ith history push task, the probability that the history task is received by the task response active user u to the ith history push task is as follows:

3. The method of claim 1, wherein the estimating, according to the historical task acceptance probability and the response result label data, a user likelihood degree of the task responding to the active user with respect to the historical pushing task and the response result corresponding to the response result label data of the historical pushing task comprises:

wherein ,P_u(A₁|x_uj) Representing the historical task acceptance probability of the j-th historical push task in the m historical push tasks by the task response active user u; y is_ujIs the said renThe response result label data of the business response active user u to the jth historical push task in the m historical push tasks, j is a positive integer, j belongs to [1, m ∈]，y_uj∈{0,1}。

4. The method of claim 1, wherein the determining a first high-preference target user from the first preset number of task response active users according to the user likelihood degree corresponding to each of the first preset number of task response active users comprises:

5. The method of claim 1, wherein the method further comprises:

6. The method of claim 1, wherein pushing the target push task of the target task type to the first high-preference target user comprises:

7. A crowdsourcing task pushing device based on user preferences, comprising:

8. The user-preference-based crowdsourcing task pushing device of claim 7, wherein the device further comprises:

9. A crowdsourcing task pushing device based on user preferences, comprising a processor, a memory and a communication interface, wherein the processor, the memory and the communication interface are connected with each other, wherein the communication interface is used for receiving and sending data, the memory is used for storing program codes, and the processor is used for calling the program codes and executing the method according to any one of claims 1-6.

10. A computer storage medium, characterized in that the computer storage medium stores a computer program comprising program instructions that, when executed by a processor, cause the processor to perform the method according to any one of claims 1-6.