CN112270348A

CN112270348A - User activation method, model training method, device, equipment and storage medium

Info

Publication number: CN112270348A
Application number: CN202011129803.9A
Authority: CN
Inventors: 顾文渊; 谭乾; 陈汉; 华锦芝
Original assignee: China Unionpay Co Ltd
Current assignee: China Unionpay Co Ltd
Priority date: 2020-10-21
Filing date: 2020-10-21
Publication date: 2021-01-26

Abstract

The application discloses a user activation method, a model training method, a device, equipment and a storage medium, and belongs to the field of data processing. The method comprises the following steps: acquiring user characteristic data and user service characteristic data of a tested user; inputting user characteristic data and user service characteristic data of a detected user into a pre-trained activation probability model to obtain a predicted activation probability of the detected user output by the activation probability model, wherein the activation probability model comprises a regression tree, and leaf nodes of the regression tree are obtained by training according to the user characteristic data and the user service characteristic data of a sample user to ensure that the difference between the predicted activation probability and the actual activation probability is within an expected value range; determining a target user based on the predicted activation probability of the detected user; and sending activation reach information to the target user, wherein the activation reach information is used for improving the probability of the increase of the activity of the service used by the target user. According to the embodiment of the application, the accuracy of predicting the user activation success can be improved.

Description

User activation method, model training method, device, equipment and storage medium

Technical Field

The present application relates to the field of data processing, and in particular, to a user activation method, a model training method, an apparatus, a device, and a storage medium.

Background

With the development of information technology, it is becoming an important trend for users to complete a certain service through an application program or a specific device. Due to the universality of the service, each service corresponds to a large number of users, and whether to take an activation measure for the user can be determined according to whether the user uses the service within a period of time, so that the activity of the user on the service is increased.

In order to activate the user, an activation measure may be taken for all users corresponding to the service. However, since the number of users is very large, for example, 500 to 1000 ten thousand, the activation measures are taken for all users, and a large amount of activation resources are wasted. At present, in order to avoid the waste of activated resources, part of users are randomly selected from a large number of users corresponding to the service, and activation measures are taken for the randomly selected part of users. However, there are more users who fail to activate among the randomly selected users, so that the accuracy of the user activation success is low.

Disclosure of Invention

The embodiment of the application provides a user activation method, a model training method, a device, equipment and a storage medium, and can improve the accuracy of predicting the success of user activation.

In a first aspect, an embodiment of the present application provides a user activation method, including: acquiring user characteristic data of a tested user and user service characteristic data of the tested user; inputting user characteristic data and user service characteristic data of a detected user into a pre-trained activation probability model to obtain a predicted activation probability of the detected user output by the activation probability model, wherein the activation probability model comprises a regression tree, and leaf nodes of the regression tree are obtained by training according to the user characteristic data of a sample user and the user service characteristic data of the sample user so as to enable the difference between the predicted activation probability output by the activation probability model and the actual activation probability to be within an expected value range; determining target users based on the predicted activation probability of the detected users, wherein the target users are at least part of the detected users; and sending activation reach information to the target user, wherein the activation reach information is used for improving the probability of the increase of the activity of the service used by the target user.

In a second aspect, an embodiment of the present application provides a model training method, including: acquiring sample data, wherein the sample data comprises user characteristic data of a sample user, user service characteristic data and an activation state; establishing a first model based on the sample data, wherein the first model comprises a regression tree, and the regression tree comprises leaf nodes determined according to the sample data; calculating an objective function of the first model according to the regression tree, wherein the objective function is used for representing the difference between the predicted activation probability and the actual activation probability output by the first model; and taking a first model of which the value of the objective function is within the range of expected values as a trained activation probability model.

In a third aspect, an embodiment of the present application provides a user activation device, including: the acquisition module is used for acquiring the user characteristic data of the tested user and the user service characteristic data of the tested user; the operation module is used for inputting the user characteristic data and the user service characteristic data of the detected user into a pre-trained activation probability model to obtain the predicted activation probability of the detected user output by the activation probability model, the activation probability model comprises a regression tree, and leaf nodes of the regression tree are obtained by training according to the user characteristic data of the sample user and the user service characteristic data of the sample user, so that the difference between the predicted activation probability output by the activation probability model and the actual activation probability is within an expected value range; the determining module is used for determining target users based on the predicted activation probability of the detected users, and the target users are at least part of the detected users; and the sending module is used for sending activation touch information to the target user, and the activation touch information is used for improving the probability of the activity rise of the service used by the target user.

In a fourth aspect, an embodiment of the present application provides a model training apparatus, including: the acquisition module is used for acquiring sample data, and the sample data comprises user characteristic data of a sample user, user service characteristic data and an activation state; the model establishing module is used for establishing a first model based on the sample data, the first model comprises a regression tree, and the regression tree comprises leaf nodes determined according to the sample data; the calculation module is used for calculating an objective function of the first model according to the regression tree, wherein the objective function is used for representing the difference between the predicted activation probability and the actual activation probability output by the first model; and the training module is used for taking the first model of which the value of the target function is within the range of expected values as the trained activation probability model.

In a fifth aspect, an embodiment of the present application provides a user activation device, including: a processor and a memory storing computer program instructions; the processor, when executing the computer program instructions, implements the user activation method of the first aspect.

In a sixth aspect, an embodiment of the present application provides a model training apparatus, including: a processor and a memory storing computer program instructions; the processor, when executing the computer program instructions, implements the model training method of the second aspect.

In a seventh aspect, an embodiment of the present application provides a computer storage medium, where the computer storage medium stores computer program instructions, and the computer program instructions, when executed by a processor, implement the user activation method of the first aspect or implement the model training method of the second aspect.

The embodiment of the application provides a user activation method, a model training method, a device, equipment and a storage medium. And determining the leaf nodes in the regression tree according to the sample data. And taking a first model with the value of the objective function within a desired range as a trained activation probability model. And the target function is obtained by calculation according to the regression tree and is used for representing the difference between the predicted activation probability and the actual activation probability output by the first model. And inputting the user characteristic data and the user service characteristic data into the activation probability model to obtain the corresponding predicted activation probability output by the activation probability model. The activation probability model is obtained by training according to a large amount of sample data, and the value of the target function of the activation probability model obtained by training is within an expected range, namely the accuracy of the corresponding predicted activation probability output by the activation probability model is higher, so that the accuracy of predicting the activation success of the user is improved. And determining the target user receiving the sent activation touch information according to the predicted activation probability with higher accuracy, wherein the probability that the activity of the service used by the target user is increased, namely the accuracy of successful activation of the user is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments of the present application will be briefly described below, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is a flow diagram of one embodiment of a model training method provided herein;

FIG. 2 is a diagram illustrating an example of a regression tree in a first model according to an embodiment of the present disclosure;

FIG. 3 is a flow chart of another embodiment of a model training method provided herein;

FIG. 4 is a flow chart of yet another embodiment of a model training method provided herein;

FIG. 5 is a flow chart of an embodiment of a user activation method provided herein;

FIG. 6 is a flow chart of another embodiment of a user activation method provided herein;

FIG. 7 is a flow chart of yet another embodiment of a user activation method provided herein;

FIG. 8 is a flow chart of yet another embodiment of a user activation method provided herein;

FIG. 9 is a schematic diagram illustrating an embodiment of a model training apparatus according to the present application;

FIG. 10 is a schematic structural diagram of another embodiment of a model training device provided in the present application;

FIG. 11 is a schematic structural diagram of a model training device according to another embodiment of the present disclosure;

FIG. 12 is a schematic diagram illustrating an embodiment of a user-activated device;

FIG. 13 is a schematic diagram illustrating another embodiment of a user-activated device;

FIG. 14 is a schematic diagram illustrating a structure of a user-activated device according to yet another embodiment of the present disclosure;

FIG. 15 is a schematic diagram illustrating an embodiment of a model training apparatus provided herein;

fig. 16 is a schematic structural diagram of an embodiment of a user activation device provided in the present application.

Detailed Description

Features and exemplary embodiments of various aspects of the present application will be described in detail below, and in order to make objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are intended to be illustrative only and are not intended to be limiting. It will be apparent to one skilled in the art that the present application may be practiced without some of these specific details. The following description of the embodiments is merely intended to provide a better understanding of the present application by illustrating examples thereof.

With the development of information technology, it is becoming an important trend for users to complete a certain service through an application program or a specific device. For example, a user may complete a transaction service by operating a transaction application. Since the number of users is very large, each service corresponds to a large number of users. But the activity of each user using the service is not the same. In some cases, a user who has not used a service for a period of time may be defined as an attrition user, and a user who has used a service for a period of time may be defined as an attrition-free user. The activity of the lost user and the activity of the non-lost user to the service are different, the activity of different lost users to the service can be different, and the activity of different non-lost users to the service is also different. In order to improve the activity of the user on the service, an activation measure can be taken for the user, for example, activation reach information is sent. However, at present, it is difficult to activate users more accurately and successfully under the condition of a large number of users. Here, "activating" refers to increasing the activity of a user on a service.

The embodiment of the application provides a user activation method, a model training method, a device, equipment and a storage medium, and can predict the activation probability of each detected user through an activation probability model obtained through training, and take activation measures according to the predicted activation probability of the detected user so as to improve the accuracy of user activation success. The tested users may include attrition users or non-attrition users, and are not limited herein. In this embodiment of the application, the sample user, the tested user, and the like may specifically refer to an account of the sample user and an account of the tested user, which is not limited herein.

Fig. 1 is a flowchart of an embodiment of a model training method provided in the present application. As shown in fig. 1, the model training method may include steps S101 to S104.

In step S101, sample data is acquired.

The sample data includes user characteristic data, user service characteristic data and activation status of the sample user. The sample users are users serving as training samples, and the number of the sample users may be set according to training scenarios and training requirements, which is not limited herein.

The user characteristic data is used to characterize the user. For example, the user characteristic data may include, but is not limited to, one or more of the following: age, gender, residence, occupation, income level. In some cases, various user characteristic data can be converted into data which is convenient for recognition of a training model according to a standardized rule. For example, sex includes male or female, and the conversion from "male" to "1" and "female" to "2" are not limited herein.

The user service characteristic data is used to characterize the service used by the user. For example, user traffic characteristic data may include, but is not limited to, one or more of the following: the service completion number, the service resource transfer amount, the service mode, the service resource receiving party number, the service association card number, the service completion date, the service scene, the service preference, the service operation number, each service function operation number, the service function number, the service operation date, the service operation preference, the service activity information, the service resource consumption rule information and the user history activation reach information. Specifically, the acquired data may be user service feature data within a predetermined time period. For example, user traffic characteristic data of sample users within one month is obtained.

In the process of using the service by the user, there may be a case that the user performs the service operation but does not complete the service finally. The number of completed services is the number of completed services for the user. The service resource transfer amount is an amount of resource transfer in a service performed by a user, for example, in a transaction scenario, the service resource transfer amount may specifically be a transaction amount. The business method is a method of performing business by the user, for example, in a transaction scenario, the business method may include, but is not limited to, an application transaction, a card swiping transaction, and the like. The number of service resource receivers is the number of receivers of resources in the service performed by the user, for example, in a transaction scenario, the number of service resource receivers may include the number of merchants transacting with the user. The number of the business-related cards is the number of resource cards related to the business performed by the user, for example, in a transaction scenario, the number of the business-related cards may be the number of bank cards on which the transaction occurs by the user. The service completion date is the date on which the service of the user is completed. The business scenario is a scenario of a business of the user, for example, the business scenario may include, but is not limited to, dining, traveling, shopping, and the like. The service preference is a service preferred by the user. The number of business operations is the total number of business operations performed by the user. The number of the service function operations is the number of the user operations corresponding to each function of the service, for example, the number of the operations of a certain function module by the user, and each function module realizes one function. The number of service functions is the number of functions in the service, for example, the number of function modules corresponding to the service. The business operation date is the date when the business operation occurs. The business operation preference is a user preferred business operation. The business activity information is information of an activity initiated in the business, for example, in a trading scenario, the business activity information may include marketing activity information. The service resource consumption rule information is a rule according to which the service resource is consumed, for example, in a transaction scenario, the service resource consumption rule information may be preference information. The user history activation reach information is activation reach information received by the user history.

For a sample user, the activation status includes activation success or activation failure. In some examples, activation success may be represented by "1" and activation failure may be represented by "0", which is not limited herein.

In step S102, a first model is built based on the sample data.

The first model includes a regression tree. The regression tree includes leaf nodes determined from the sample data.

Specifically, the regression trees can be continuously added according to the user characteristic data and the user service characteristic data in the sample data, and the construction of each regression tree can be completed by adjusting and splitting continuously according to the user characteristic data and the user service characteristic data. Each added regression tree, equivalent to learning a new function, fits the residuals including the previously constructed regression tree predictions. The number of regression trees in the first model is not limited herein.

And according to the constructed regression tree, leaf nodes obtained by splitting the user characteristic data and the user service characteristic data of the sample user can be obtained. The predicted activation probability corresponding to a sample user is the sum of the scores of the leaf nodes of the sample user in all the regression trees. For example, the corresponding predicted activation probability of the sample user can be obtained according to equation (1):

wherein the content of the first and second substances,

to predict activation probability, x_iFor the ith sample user, K is the number of regression trees in the first model, f_k(x_i) And the score of the leaf node corresponding to the ith sample user in the kth regression tree is obtained.

In step S103, an objective function of the first model is calculated from the regression tree.

The objective function is used for representing the difference between the predicted activation probability and the actual activation probability output by the first model.

In some examples, the objective function includes a sum of values of the loss function for each sample user. The penalty function is used to characterize the gap between the predicted activation probability and the actual activation probability. For example, the objective function can be obtained according to equation (2):

wherein, L is an objective function,

n is the number of sample users as a loss function for the ith sample user.

In other examples, the objective function further includes a regularization term. The regularization term is related to the number of leaf nodes and the score of the leaf nodes of each regression tree. The regularization term is used to prevent the objective function from over-fitting, thereby improving the accuracy of the objective function. For example, the objective function can be obtained according to equation (3):

wherein, L is an objective function,

is the loss function of the ith sample user, n is the number of sample users, K is the number of regression trees, Ω (f)_k) The regularization term for the kth regression tree.

The regularization term in equation (3) above can be obtained according to equation (4):

where Ω (f) is the regularization term for a regression tree, γ is a constant coefficient, T is the number of leaf nodes of the regression tree, λ is a constant coefficient, ω is the fraction of each leaf node in the regression tree, which can be expressed as a vector or matrix, | ω | is the norm of the fraction of each leaf node in the regression tree.

In the above embodiment, the predicted activation probability of the target sample user is the sum of the scores of the leaf nodes to which the target sample user belongs in each regression tree in the first model. The target sample user is any one of the sample users. For example, fig. 2 is a schematic diagram of an example of a regression tree in the first model provided in the embodiment of the present application. As shown in FIG. 2, the first model includes two regression trees, with the leaf nodes labeled the identity of the sample user and the leaf nodes labeled the scores below the leaf nodes. The predicted activation probability of the target sample user a is 0.3+0.6 ═ 0.6.

In step S104, a first model in which the value of the objective function is within the expected value range is used as the trained activation probability model.

The expected value range may be set according to training scenarios and training requirements, and is not limited herein. In some examples, a first model for which the objective function reaches a minimum value that can be reached may be used as the trained activation probability model. In other examples, the expected value range may be set in consideration of the resource consumption amount of the activation probability model, and the like, and is not limited herein. And inputting the user characteristic data and the user service characteristic data of the tested user into the trained activation probability model, and outputting the accurate predicted activation probability of the tested user by the trained activation probability model.

The embodiment of the application is based on a Gradient Boosting Decision Tree (GBDT) algorithm, and the training of the model is realized by taking Gradient Boosting as a framework. Specifically, the first model may be an integrated machine learning model based on a decision tree, such as an XGBoost (explicit Gradient Boosting) model or a LightGBM model, which is not limited herein, and can reduce occupied memory and improve operation speed.

In the embodiment of the application, a first model comprising a regression tree is established by using sample data comprising user characteristic data, user service characteristic data and an activation state of a sample user. And determining the leaf nodes in the regression tree according to the sample data. And taking a first model with the value of the objective function within a desired range as a trained activation probability model. And the target function is obtained by calculation according to the regression tree and is used for representing the difference between the predicted activation probability and the actual activation probability output by the first model. And inputting the user characteristic data and the user service characteristic data into the activation probability model to obtain the corresponding predicted activation probability output by the activation probability model. The activation probability model is obtained by training according to a large amount of sample data, and the value of the target function of the activation probability model obtained by training is within an expected range, namely the accuracy of the corresponding predicted activation probability output by the activation probability model is higher, so that the accuracy of predicting the activation success of the user is improved.

Through later evaluation on the effect of the activation probability model obtained in the embodiment of the application, the effect of the activation probability model obtained in the embodiment of the application is better. For example, the effect of the model can be evaluated by using the AUC (i.e., Area Under cutter) value, KS (i.e., Kolmogorov-Smirnov) value, and other evaluation parameters of the model. The AUC values represent the probability that the positive case is ranked before the negative case as judged by the model, and the higher the AUC value, the better the model effect. The KS value is used for evaluating the risk discrimination capability of the model, and the larger the KS value is, the stronger the risk discrimination capability of the model is. And improving the KS value of the model, namely improving the risk discrimination capability of the model. The AUC value of the activation probability model obtained in the embodiment of the application can reach 0.9, and the KS value can reach 0.65.

FIG. 3 is a flow chart of another embodiment of a model training method provided herein. Fig. 3 is different from fig. 1 in that the model training method shown in fig. 3 may further include step S105.

In step S105, the first data is normalized according to a preset normalization rule, and normalized first data is obtained.

The first data comprises user characteristic data and/or user traffic characteristic data. Because the types of the characteristic data of each user and the service characteristic data of each user are different, in order to facilitate the training of the model, the first data can be standardized into the data which meets the training requirement of the model. The first data that already meets the model training requirements need not be normalized. The normalization rule may be determined according to the type of the first data, and each item of the first data may correspond to a different specific normalization rule, which is not limited herein.

For example, for the first data of the date-class variable, such as the service completion date, the reciprocal of the time difference between the service completion date and a certain specified date may be calculated as the service completion date after the normalization processing. For example, if the specified date is 6/1/2020, if the service completion date of a certain sample user is 5/31/2020, the service completion date after the standardization process for that sample user is 1/1, and if the service completion date of that sample user is 5/30/2020, the service completion date after the standardization process for that sample user is 1/2, and so on. The normalization process can effectively process missing values. In some examples, if the service completion date is the service completion date of the latest service, for a user without service, the service completion date only needs to be set to 0, and the service completion date representing the latest service is infinite from the lost month of the user.

In some examples, a dispersion parameter for each item of first data may be determined, the dispersion parameter being used to characterize the degree of dispersion of an item of first data; carrying out first standardization processing on each item of first data with the dispersion parameter higher than the dispersion threshold value to enable the dispersion parameter of the first data to be lower than or equal to the dispersion threshold value; and obtaining normalized first data according to the first data with the dispersion parameter lower than or equal to the dispersion threshold value and the first data after the first normalization processing.

The first normalization process may be set according to the type of the dispersion parameter, and is not limited herein. Various items of first data with the dispersion parameter lower than or equal to the dispersion threshold value and various items of first data after the first standardization processing can be directly used as standardized first data; the first data items with the dispersion parameter lower than or equal to the dispersion threshold value and the first data items after the first normalization processing may also be processed to obtain normalized first data, which is not limited herein.

The KS value of the model can be improved by performing the first normalization processing by using the dispersion parameter and the dispersion threshold.

For example, the dispersion parameter may be a standard deviation coefficient. The higher the standard deviation coefficient, the greater the degree of dispersion of the first data representing this term. The lower the standard deviation coefficient, the smaller the degree of dispersion of the first data representing this term. The standard deviation coefficient can be obtained according to equation (5):

wherein, V_σIs the coefficient of standard deviation, σ is the standard deviation, and x is the mean.

The dispersion threshold may be 1, corresponding to the standard deviation coefficient. When the standard deviation coefficient is higher than 1, the discrete degree of the first data is determined to be too high, and the logarithm can be taken for the first data, and then the calculation is carried out. I.e. the first normalization process is a log taking process. By carrying out standardization processing by using the dispersion parameter and the dispersion threshold value, the KS value of the model can be improved by about 0.10.

In other examples, each item of first data exceeding the preset threshold range is subjected to second normalization processing, so that each item of first data exceeding the preset threshold range is updated to an endpoint value of the preset threshold range; and obtaining the standardized first data according to the first data within the preset threshold range and the second standardized first data. Specifically, each item of first data exceeding the preset threshold range is updated to an endpoint value of the preset threshold range closest to the first data.

The preset threshold range may be set according to a training scenario and a training requirement, and is not limited herein. Various items of first data within a preset threshold range and various items of first data after second standardization processing can be directly used as the standardized first data; or processing each item of first data within a preset threshold range and each item of first data after second standardization processing to obtain standardized first data; each item of first data of which the dispersion parameter is lower than or equal to the dispersion threshold in the above example and each item of first data after the first normalization processing may be subjected to a second normalization processing, so as to obtain normalized first data, which is not limited herein.

The second standardization processing is carried out by utilizing the preset threshold range, so that the condition that the possible extreme values of certain first data, such as the number of completed services and the service resource transfer quantity of part of sample users, are far higher than the average value can be prevented, and the deviation caused by model training is avoided, thereby improving the accuracy of the model training. The KS value of the model can also be improved by performing a second normalization process using a preset threshold range.

For example, the preset threshold range is between-5 and 5. And if the first data is less than-5, updating the first data to-5. And if the first data is larger than 5, updating the first data to be 5. By performing the second normalization process within the preset threshold range, the KS value of the model can be increased by about 0.05.

FIG. 4 is a flow chart of yet another embodiment of a model training method provided herein. Fig. 4 is different from fig. 1 in that the model training method shown in fig. 4 may include step S106 and step S107.

In step S106, the importance level of each item of first data is determined based on the influence of each item of first data on the predicted activation probability output by the activation probability model.

The influence of each item of first data on the predicted activation probability output by the activation probability model can be obtained in the model training process, and the larger the influence is, the higher the importance level is, and the higher the importance level is.

The first data includes user characteristic data and/or user service characteristic data, and specific contents may refer to relevant descriptions in the above embodiments, which are not described herein again.

In step S107, a weighting factor of each item of first data is determined according to the importance level of each item of first data.

The weight coefficient of the first data is positively correlated with the importance level of the first data, namely the higher the importance level of the first data is, the larger the weight coefficient of the first data is; the lower the importance level of the first data is, the smaller the weight coefficient of the first data is. The weight coefficient can be used for multiplying first data, and the product of the first data and the corresponding weight coefficient is used as input data of the input activation probability model; or, the first data is input into the activation probability model, the first data is multiplied by the weight coefficient in the activation probability model, and other operations in the activation probability model are carried out on the product.

And setting the weight coefficient of the first data according to the importance level of the first data, thereby realizing the optimization of the activation probability model and improving the accuracy of the predicted activation probability output by the activation probability model.

In some examples, the weight coefficient of the first data with the importance level lower than the level threshold may be set to 0, or the first data with the importance level lower than the level threshold may be discarded, so as to simplify the complexity of the activation probability model and improve the operation speed of the activation probability model.

The application also provides a user activation method, which can be realized by using the activation probability model obtained by training the model training method in the embodiment. Fig. 5 is a flowchart of an embodiment of a user activation method provided in the present application. As shown in fig. 5, the user activation method may include steps S201 to S204.

In step S201, user feature data of the user to be tested and user service feature data of the user to be tested are obtained.

The user to be tested is the user to be predicted, and the number of the users to be tested is not limited herein.

In some examples, the user traffic characteristic data includes one or more of: the service completion number, the service resource transfer amount, the service mode, the service resource receiving party number, the service association card number, the service completion date, the service scene, the service preference, the service operation number, each service function operation number, the service function number, the service operation date, the service operation preference, the service activity information, the service resource consumption rule information and the user history activation reach information.

For specific contents of the user characteristic data and the user service characteristic data, reference may be made to the relevant description in the foregoing embodiments, and details are not described herein again.

In step S202, the user characteristic data and the user service characteristic data of the detected user are input into a pre-trained activation probability model, so as to obtain the predicted activation probability of the detected user output by the activation probability model.

The activation probability model includes a regression tree. And training leaf nodes of the regression tree according to the user characteristic data of the sample user and the user service characteristic data of the sample user to enable the difference between the predicted activation probability output by the activation probability model and the actual activation probability to be within an expected value range.

For the details of the pre-trained activation probability model, the regression tree, the leaf nodes, and the like, reference may be made to the relevant description in the above embodiments, which is not repeated herein.

In step S203, a target user is determined based on the predicted activation probability of the detected user.

The target users are at least part of the users to be tested. In some examples, a user under test with a predicted activation probability above an activation probability threshold may be considered a predicted activation probability. In other examples, the users may be ranked according to the predicted activation probability from high to low, with the top m% of the users being targeted.

In step S204, activation reach information is sent to the target user.

The activation trigger information is used for improving the probability of the rising of the activity of the target user for using the service. That is, after the target user receives the activation touch information, the probability that the activity of the service associated with the activation touch information used by the target user increases. The activity of the target user using the service is increased to indicate that the target user is successfully activated.

The activation reach information may include, but is not limited to, business activity information, business resource consumption rules, and the like. For example, in a transaction scenario, the activation reach information may include, but is not limited to, offer business information, offer rules, and the like.

In the embodiment of the application, the user characteristic data and the user service characteristic data of the tested user are input into the pre-trained activation probability model, and the predicted activation probability of the tested user output by the activation probability model can be obtained. The activation probability model includes a regression tree. And training leaf nodes in the regression tree according to the user characteristic data and the user service characteristic data of the sample user to obtain the leaf nodes. The value of the target function of the activation probability model, which is used for representing the difference between the predicted activation probability and the actual activation probability, is within the expected range, namely the accuracy of the corresponding predicted activation probability output by the activation probability model is higher, and the accuracy of the user activation success prediction is improved. And determining the target user receiving the sent activation touch information according to the predicted activation probability with higher accuracy, wherein the probability that the activity of the service used by the target user is increased, namely the accuracy of successful activation of the user is improved.

The touch effect evaluation can be performed according to the user who sends the activation touch information by using the user activation method in the embodiment of the application. For example, the detected users predicted by the activation coverage rate, i.e., the activation probability model, are ranked according to the predicted activation probability from high to low, and the proportion of the users with successful activation in the top p% of the detected users to the total users with successful activation is considered, so as to evaluate the reach effect. Under the condition that the user activation is carried out by using the user activation method in the embodiment of the application, the users which are successfully activated in the activation touch information can be sent to 7% of the lost users, 88% of the users which are successfully activated are covered, and a good touch effect is shown.

Fig. 6 is a flowchart of another embodiment of a user activation method provided in the present application. Fig. 6 is different from fig. 5 in that the user activation method shown in fig. 6 may further include step S205.

In step S205, the first data is normalized according to a preset normalization rule, and normalized first data is obtained.

Wherein the first data comprises user characteristic data and/or user service characteristic data.

In some examples, a dispersion parameter of each item of first data can be specifically determined, and the dispersion parameter is used for representing the dispersion degree of one item of first data; carrying out first standardization processing on each item of first data with the dispersion parameter higher than the dispersion threshold value to enable the dispersion parameter of the first data to be lower than or equal to the dispersion threshold value; and obtaining normalized first data according to the first data with the dispersion parameter lower than or equal to the dispersion threshold value and the first data after the first normalization processing.

In other examples, the first data exceeding the preset threshold range may be subjected to a second normalization process, so that the first data exceeding the preset threshold range are updated to an endpoint value of the preset threshold range; and obtaining the standardized first data according to the first data within the preset threshold range and the second standardized first data.

The specific contents of the first data, the preset normalization rule, the normalization process, the dispersion parameter, the dispersion threshold, the first normalization process, the preset threshold range, the second normalization process, and the like can be referred to the relevant description in the model training method, and the difference from the model training method is that the first data includes user characteristic data and/or user service characteristic data of the user to be tested, and the description thereof is omitted.

Fig. 7 is a flowchart of another embodiment of a user activation method provided in the present application. Fig. 7 is different from fig. 5 in that step S204 in fig. 5 can be further detailed as step S2041 in fig. 7, and the user activation method shown in fig. 7 can further include step S206 and step S207.

In step S2041, the activation reach information in the selected reach scenario is sent to the target user in the selected reach manner.

The touch-up mode may be selected randomly or as desired, and is not limited herein. For example, the reach-through may include a short message, a phone call, a mail, an application push, etc.

The touchdown scene may also be chosen randomly or as desired, and is not limited herein. For example, a reach scenario is an associated field that activates reach information, which may include dining, transportation, shopping, and the like.

In step S206, the actual activation probability of the target user, the reach mode and the reach scene corresponding to the activation reach information received by the target user are obtained.

The method can record the touch and reach mode and the touch and reach scene corresponding to the target user under the condition that the activation touch and reach information under the selected touch and reach scene is sent to the target user through the selected touch and reach mode, and the actual activation probability of the target user is determined according to the subsequent service use condition of the target user. And receiving the touch and reach mode, the touch and reach scene and the actual activation probability fed back by the target user after sending the activation touch and reach information under the selected touch and reach scene to the target user through the selected touch and reach mode.

In step S207, the actual activation probability of the target user, the reach mode and the reach scenario corresponding to the activation reach information received by the target user are used as sample data of the activation probability model to train the activation probability model, so that the activation probability model is further used to output the predicted optimal reach mode and the predicted optimal reach scenario, and the activation reach information under the predicted optimal reach scenario is sent to the target user through the predicted optimal reach mode.

And performing optimization training on the activation probability model according to the actual activation probability, the reach mode and the reach scene fed back by the target user, so that the activation probability model can also output a predicted optimal reach mode and a predicted optimal reach scene. The predicted optimal reach mode is the reach mode with the highest probability of successful user activation predicted by the activation probability model. The predicted best reach scenario is the reach scenario with the highest probability of successful user activation predicted by the activation probability model. The activation reach information under the condition of predicting the optimal reach is sent to the target user through the optimal reach predicting mode, so that the probability of the activity rise of the user using the service can be further improved, and the accuracy of the user activation success is further improved.

Fig. 8 is a flowchart of a further embodiment of a user activation method provided in the present application. Fig. 8 is different from fig. 5 in that the user activation method shown in fig. 8 may further include step S208, step S209, and step S210.

In step S208, the probability that the user characteristic data, the user service characteristic data, the predicted activation probability, and the actual activation probability of the first detected user are selected as sample data is increased.

The first tested user comprises a tested user of which the activation state corresponding to the predicted activation probability is different from the activation state corresponding to the actual activation probability. And obtaining the activation state corresponding to the activation probability according to the activation probability and a preset state probability threshold. Specifically, the activation probability is higher than or equal to a preset state probability threshold, and the activation state corresponding to the activation probability is an activation success state; the activation probability is lower than a preset state probability threshold, and the activation state corresponding to the activation probability is an activation failure state. Namely, the first tested user includes a tested user whose activation state corresponding to the predicted activation probability is an activation success state but whose activation state corresponding to the actual activation probability is an activation failure state, and a tested user whose activation state corresponding to the predicted activation probability is an activation failure state but whose activation state corresponding to the actual activation probability is an activation success state.

The user characteristic data, the user service characteristic data, the predicted activation probability and the actual activation probability of the first tested user are increased to be sample data, the number of the data of the first tested user in the sample data can be increased, the activation probability model is optimally trained by using the sample data, the weakness of the activation probability model can be overcome, and the accuracy of the predicted activation probability output by the activation probability model is further improved.

In step S209, the probability that the user characteristic data, the user service characteristic data, the predicted activation probability, and the actual activation probability of the second user to be tested are selected as sample data is reduced.

And the second tested user comprises a tested user of which the activation state corresponding to the predicted activation probability is the same as the activation state corresponding to the actual activation probability. Namely, the second detected user includes a detected user whose activation state corresponding to the predicted activation probability is an activation success state and whose activation state corresponding to the actual activation probability is an activation success state, and a detected user whose activation state corresponding to the predicted activation probability is an activation failure state and whose activation state corresponding to the actual activation probability is an activation failure state.

The user characteristic data, the user service characteristic data, the predicted activation probability and the actual activation probability of the second tested user are reduced, the number of the data of the second tested user in the sample data can be reduced, the number of the data of the first tested user in the sample data is increased, the activation probability model is optimally trained by using the sample data, the weakness of the activation probability model can be made up, and the accuracy of the predicted activation probability output by the activation probability model is further improved.

The above steps S208 and S209 may be executed, or one of them may be executed, and are not limited herein.

In step S210, user characteristic data, user service characteristic data, predicted activation probability, and actual activation probability of at least part of the detected users are selected as sample data of the activation probability model, so as to perform optimization training on the activation probability model.

Under the condition that the number of detected users is large, namely under the condition that the amount of each item of data of the detected users is large, in order to reduce the operation amount of the optimization training of the activation probability model on the basis of ensuring the optimization training effect of the activation probability model, part of the data of the detected users can be selected as sample data to carry out optimization training on the activation probability model.

And under the condition that the number of the detected users is small, selecting all or part of the data of the detected users as sample data to carry out optimization training on the activation probability model.

The activation probability model after optimization training can be used for next activation probability prediction and activation touch information sending. The activation probability model can be continuously optimized by utilizing various data of the tested user, so that the activation probability model which is continuously iteratively optimized is generated, the accuracy of the predicted activation probability output by the activation probability model is continuously improved, the accuracy of the success of the user activation is improved, and the accuracy of the success of the user activation is improved.

The application also provides a model training device. Fig. 9 is a schematic structural diagram of an embodiment of a model training apparatus provided in the present application. As shown in FIG. 9, the model training apparatus 300 may include an obtaining module 301, a model building module 302, a calculating module 303, and a training module 304.

The obtaining module 301 may be used to obtain sample data.

The sample data includes user characteristic data, user service characteristic data and activation status of the sample user.

Model building module 302 may be used to build a first model based on sample data.

The calculation module 303 may be configured to calculate an objective function of the first model from the regression tree.

The training module 304 may be configured to use a first model with a value of the objective function within a range of expected values as the trained activation probability model.

In some examples, the objective function includes a sum of values of the loss function for each sample user. The penalty function is used to characterize the gap between the predicted activation probability and the actual activation probability.

In some examples, the objective function further includes a regularization term that is related to the number of leaf nodes and the score of the leaf nodes of each regression tree.

In some examples, the predicted activation probability of the target sample user is a sum of scores of leaf nodes to which the target sample user belongs in each regression tree in the first model. The target sample user is any one of the sample users.

Fig. 10 is a schematic structural diagram of another embodiment of the model training apparatus provided in the present application. FIG. 10 differs from FIG. 9 in that the model training apparatus 300 may further include a normalization module 305.

The normalization module 305 may be configured to perform normalization processing on the first data according to a preset normalization rule, so as to obtain normalized first data, where the first data includes user characteristic data and/or user service characteristic data.

Specifically, the normalization module 305 can be configured to determine a dispersion parameter of each item of the first data, where the dispersion parameter is used to characterize a dispersion degree of one item of the first data; carrying out first standardization processing on each item of first data with the dispersion parameter higher than the dispersion threshold value to enable the dispersion parameter of the first data to be lower than or equal to the dispersion threshold value; and obtaining normalized first data according to the first data with the dispersion parameter lower than or equal to the dispersion threshold value and the first data after the first normalization processing.

Specifically, the normalizing module 305 may be configured to perform a second normalizing process on each item of first data exceeding the preset threshold range, so that each item of first data exceeding the preset threshold range is updated to an endpoint value of the preset threshold range; and obtaining the standardized first data according to the first data within the preset threshold range and the second standardized first data.

Fig. 11 is a schematic structural diagram of a model training apparatus according to another embodiment of the present application. Fig. 11 is different from fig. 9 in that the model training apparatus 300 shown in fig. 11 may further include a rank determination module 306 and a weight determination module 307.

The rank determination module 306 may be configured to determine a rank of importance for each item of first data based on an effect of the item of first data on a predicted activation probability output by the activation probability model.

The first data comprises user characteristic data and/or user traffic characteristic data.

The weight determining module 307 may be configured to determine a weight coefficient of each item of first data according to the importance level of each item of first data.

The weight coefficient of the first data is positively correlated with the importance level of the first data.

The application also provides a user activation device. Fig. 12 is a schematic structural diagram of an embodiment of a user activation device provided in the present application. As shown in fig. 12, the user activation apparatus 400 may include an obtaining module 401, an operation module 402, a determination module 403, and a sending module 404.

The obtaining module 401 may be configured to obtain user feature data of a user to be tested and user service feature data of the user to be tested.

The operation module 402 may be configured to input the user feature data and the user service feature data of the detected user into a pre-trained activation probability model, so as to obtain a predicted activation probability of the detected user output by the activation probability model.

The determination module 403 may be used to determine the target user based on the predicted activation probability of the detected user.

The target users are at least part of the users to be tested.

The sending module 404 may be used to send activation reach information to the target user.

The activation trigger information is used for improving the probability of the rising of the activity of the target user for using the service.

Specifically, the sending module 404 may be configured to send activation reach information in the selected reach scenario to the target user through the selected reach mode.

In some examples, the obtaining module 401 may further be configured to obtain an actual activation probability of the target user, a reach mode and a reach scene corresponding to activation reach information received by the target user; the actual activation probability of the target user, the reach mode and the reach scene corresponding to the activation reach information received by the target user are used as sample data of the activation probability model to train the activation probability model, so that the activation probability model is further used for outputting a predicted optimal reach mode and a predicted optimal reach scene, and the activation reach information under the predicted optimal reach scene is sent to the target user through the predicted optimal reach mode by the sending module 404.

Fig. 13 is a schematic structural diagram of another embodiment of a user activation device provided in the present application. FIG. 13 differs from FIG. 12 in that the user-activated device 400 shown in FIG. 13 may also include a normalization module 405.

The normalization module 405 may be configured to normalize the first data according to a preset normalization rule, so as to obtain normalized first data.

In some examples, the normalization module 405 may be configured to determine a dispersion parameter for each item of first data, the dispersion parameter being used to characterize a degree of dispersion of the item of first data; carrying out first standardization processing on each item of first data with the dispersion parameter higher than the dispersion threshold value to enable the dispersion parameter of the first data to be lower than or equal to the dispersion threshold value; and obtaining normalized first data according to the first data with the dispersion parameter lower than or equal to the dispersion threshold value and the first data after the first normalization processing.

In other examples, the normalizing module 405 may be configured to perform a second normalizing process on each item of first data that exceeds the preset threshold range, so that each item of first data that exceeds the preset threshold range is updated to an endpoint value of the preset threshold range; and obtaining the standardized first data according to the first data within the preset threshold range and the second standardized first data.

Fig. 14 is a schematic structural diagram of another embodiment of a user activation device provided in the present application. Fig. 14 differs from fig. 12 in that the user-activated device 400 shown in fig. 14 may also include an adjustment module 406 and a selection module 407.

The adjusting module 406 may be configured to increase the probability that the user feature data, the user service feature data, the predicted activation probability, and the actual activation probability of the first measured user are selected as sample data, where the first measured user includes a measured user whose activation state corresponding to the predicted activation probability is different from the activation state corresponding to the actual activation probability; and/or reducing the probability that the user characteristic data, the user service characteristic data, the predicted activation probability and the actual activation probability of a second tested user are selected as sample data, wherein the second tested user comprises the tested users of which the activation states corresponding to the predicted activation probability and the activation states corresponding to the actual activation probability are the same

The selecting module 407 may be configured to select user feature data, user service feature data, predicted activation probability, and actual activation probability of at least some detected users as sample data of the activation probability model, so as to perform optimization training on the activation probability model.

The embodiment of the application also provides model training equipment. Fig. 15 is a schematic structural diagram of an embodiment of a model training apparatus provided in the present application. As shown in FIG. 15, the model training apparatus 500 includes a memory 501, a processor 502, and a computer program stored on the memory 501 and executable on the processor 502.

In one example, the processor 502 may include a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or may be configured to implement one or more Integrated circuits of the embodiments of the present Application.

The Memory 501 may include Read-Only Memory (ROM), Random Access Memory (RAM), magnetic disk storage media devices, optical storage media devices, flash Memory devices, electrical, optical, or other physical/tangible Memory storage devices. Thus, in general, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software comprising computer-executable instructions and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the model training methods in accordance with the present application.

The processor 502 runs a computer program corresponding to the executable program code by reading the executable program code stored in the memory 501 for implementing the model training method in the above-described embodiments.

In one example, model training device 500 may also include a communication interface 503 and a bus 504. As shown in fig. 15, the memory 501, the processor 502, and the communication interface 503 are connected to each other via a bus 504 to complete communication therebetween.

The communication interface 503 is mainly used for implementing communication between modules, apparatuses, units and/or devices in the embodiments of the present application. Input devices and/or output devices may also be accessed through communication interface 503.

Bus 504 includes hardware, software, or both to couple the components of model training apparatus 500 to each other. By way of example, and not limitation, Bus 504 may include an Accelerated Graphics Port (AGP) or other Graphics Bus, an Enhanced Industry Standard Architecture (EISA) Bus, a Front-Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) Bus, an InfiniBand interconnect, a Low Pin Count (LPC) Bus, a memory Bus, a Micro Channel Architecture (MCA) Bus, a Peripheral Component Interconnect (PCI) Bus, a PCI-Express (PCI-X) Bus, a Serial Advanced Technology Attachment (SATA) Bus, a Video Electronics Standards Association Local Bus (VLB) Bus, or other suitable Bus, or a combination of two or more of these. Bus 504 may include one or more buses, where appropriate. Although specific buses are described and shown in the embodiments of the application, any suitable buses or interconnects are contemplated by the application.

The embodiment of the application also provides user activation equipment. Fig. 16 is a schematic structural diagram of an embodiment of a user activation device provided in the present application. As shown in fig. 16, the user activated device 600 comprises a memory 601, a processor 602 and a computer program stored on the memory 601 and executable on the processor 602.

In one example, the processor 602 may include a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or may be configured to implement one or more Integrated circuits of the embodiments of the present Application.

The Memory 601 may include Read-Only Memory (ROM), Random Access Memory (RAM), magnetic disk storage media devices, optical storage media devices, flash Memory devices, electrical, optical, or other physical/tangible Memory storage devices. Thus, in general, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software comprising computer-executable instructions and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the user activation methods in accordance with the present application.

The processor 602 runs a computer program corresponding to the executable program code by reading the executable program code stored in the memory 601 for implementing the user activation method in the above-described embodiment.

In one example, user-activated device 600 may also include a communication interface 603 and bus 604. As shown in fig. 16, the memory 601, the processor 602, and the communication interface 603 are connected via a bus 604 to complete communication therebetween.

The communication interface 603 is mainly used for implementing communication between modules, apparatuses, units and/or devices in the embodiments of the present application. Input devices and/or output devices are also accessible through communication interface 603.

The bus 604 comprises hardware, software, or both to couple the components of the user-activated device 600 to one another. By way of example, and not limitation, Bus 604 may include an Accelerated Graphics Port (AGP) or other Graphics Bus, an Enhanced Industry Standard Architecture (EISA) Bus, a Front-Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) Bus, an InfiniBand interconnect, a Low Pin Count (LPC) Bus, a memory Bus, a Micro Channel Architecture (MCA) Bus, a Peripheral Component Interconnect (PCI) Bus, a PCI-Express (PCI-X) Bus, a Serial Advanced Technology Attachment (SATA) Bus, a Video Electronics Standards Association Local Bus (VLB) Bus, or other suitable Bus, or a combination of two or more of these. Bus 604 may include one or more buses, where appropriate. Although specific buses are described and shown in the embodiments of the application, any suitable buses or interconnects are contemplated by the application.

The present application further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the model training method and/or the user activation method in the foregoing embodiments can be implemented, and the same technical effects can be achieved. The computer-readable storage medium may include a non-transitory computer-readable storage medium, such as a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and the like, which is not limited herein.

It should be clear that the embodiments in this specification are described in a progressive manner, and the same or similar parts in the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. For apparatus embodiments, device embodiments, computer-readable storage medium embodiments, reference may be made in the descriptive section to method embodiments. The present application is not limited to the particular steps and structures described above and shown in the drawings. Those skilled in the art may make various changes, modifications and additions or change the order between the steps after appreciating the spirit of the present application. Also, a detailed description of known process techniques is omitted herein for the sake of brevity.

Aspects of the present application are described above with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions/acts specified in the flowchart and/or block diagram block or blocks. Such a processor may be, but is not limited to, a general purpose processor, a special purpose processor, an application specific processor, or a field programmable logic circuit. It will also be understood that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware for performing the specified functions or acts, or combinations of special purpose hardware and computer instructions.

It will be appreciated by persons skilled in the art that the above embodiments are illustrative and not restrictive. Different features which are present in different embodiments may be combined to advantage. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art upon studying the drawings, the specification, and the claims. In the claims, the term "comprising" does not exclude other means or steps; the word "a" or "an" does not exclude a plurality; the terms "first" and "second" are used to denote a name and not to denote any particular order. Any reference signs in the claims shall not be construed as limiting the scope. The functions of the various parts appearing in the claims may be implemented by a single hardware or software module. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

Claims

1. A user activation method, comprising:

acquiring user characteristic data of a tested user and user service characteristic data of the tested user;

inputting the user characteristic data and the user service characteristic data of the detected user into a pre-trained activation probability model to obtain the predicted activation probability of the detected user output by the activation probability model, wherein the activation probability model comprises a regression tree, and leaf nodes of the regression tree are obtained by training according to the user characteristic data of a sample user and the user service characteristic data of the sample user, so that the difference between the predicted activation probability output by the activation probability model and the actual activation probability is within an expected value range;

determining target users based on the predicted activation probability of the tested users, wherein the target users are at least part of the tested users;

and sending activation reach information to the target user, wherein the activation reach information is used for improving the probability of the increase of the activity of the service used by the target user.

2. The method of claim 1, further comprising, prior to said inputting user characteristic data and user traffic characteristic data of said user under test into a pre-trained activation probability model:

and according to a preset standardization rule, carrying out standardization processing on the first data to obtain the standardized first data, wherein the first data comprises the user characteristic data and/or the user service characteristic data.

3. The method according to claim 2, wherein the normalizing the first data according to the preset normalization rule to obtain the normalized first data comprises:

determining a dispersion parameter of each item of the first data, wherein the dispersion parameter is used for representing the dispersion degree of one item of the first data;

carrying out first normalization processing on each item of the first data with the dispersion parameter higher than a dispersion threshold value, so that the dispersion parameter of the first data is lower than or equal to the dispersion threshold value;

and obtaining the normalized first data according to the first data of each item of which the dispersion parameter is lower than or equal to a dispersion threshold value and the first data of each item after first normalization processing.

4. The method according to claim 2, wherein the normalizing the first data according to the preset normalization rule to obtain the normalized first data comprises:

carrying out second standardization processing on each item of the first data exceeding a preset threshold range, and updating each item of the first data exceeding the preset threshold range into an endpoint value of the preset threshold range;

and obtaining the standardized first data according to the first data within a preset threshold range and the first data after second standardization processing.

5. The method of claim 1, wherein sending activation reach information to the target user comprises:

and sending the activation reach information under the selected reach scene to the target user through the selected reach mode.

6. The method of claim 5, further comprising:

acquiring the actual activation probability of the target user, and a reach mode and a reach scene corresponding to the activation reach information received by the target user;

and taking the actual activation probability of the target user, the reach mode and the reach scene corresponding to the activation reach information received by the target user as sample data of the activation probability model, training the activation probability model, enabling the activation probability model to be further used for outputting a predicted optimal reach mode and a predicted optimal reach scene, and sending the activation reach information under the predicted optimal reach scene to the target user through the predicted optimal reach mode.

7. The method of claim 1, further comprising:

selecting at least part of the user characteristic data, the user service characteristic data, the predicted activation probability and the actual activation probability of the tested user as sample data of the activation probability model so as to carry out optimization training on the activation probability model.

8. The method of claim 7, further comprising:

increasing the probability that the user characteristic data, the user service characteristic data, the predicted activation probability and the actual activation probability of a first tested user are selected as the sample data, wherein the first tested user comprises the tested user of which the activation state corresponding to the predicted activation probability is different from the activation state corresponding to the actual activation probability;

and/or the presence of a gas in the gas,

and reducing the probability that the user characteristic data, the user service characteristic data, the predicted activation probability and the actual activation probability of a second tested user are selected as the sample data, wherein the second tested user comprises the tested user of which the activation state corresponding to the predicted activation probability is the same as the activation state corresponding to the actual activation probability.

9. The method of claim 1, wherein the user traffic characteristic data comprises one or more of:

the service completion number, the service resource transfer amount, the service mode, the service resource receiving party number, the service association card number, the service completion date, the service scene, the service preference, the service operation number, each service function operation number, the service function number, the service operation date, the service operation preference, the service activity information, the service resource consumption rule information and the user history activation reach information.

10. A method of model training, comprising:

acquiring sample data, wherein the sample data comprises user characteristic data, user service characteristic data and an activation state of a sample user;

establishing a first model based on the sample data, the first model comprising a regression tree comprising leaf nodes determined from the sample data;

calculating an objective function of the first model according to the regression tree, wherein the objective function is used for representing the difference between the predicted activation probability and the actual activation probability output by the first model;

and taking the first model of which the value of the objective function is within the range of expected values as the trained activation probability model.

11. The method of claim 10, wherein the objective function comprises a sum of values of a loss function for each of the sample users, the loss function characterizing a gap between the predicted activation probability and the actual activation probability.

12. The method of claim 11, wherein the objective function further comprises a regularization term, wherein the regularization term is related to a number of the leaf nodes and a score of the leaf nodes of each of the regression trees.

13. The method of claim 10, wherein the predicted activation probability for a target sample user is a sum of scores of the leaf nodes to which the target sample user belongs in each regression tree in the first model,

the target sample user is any one of the sample users.

14. The method of claim 10, further comprising, prior to said building a first model based on said sample data:

15. The method according to claim 14, wherein the normalizing the first data according to the preset normalization rule to obtain the normalized first data comprises:

16. The method according to claim 14, wherein the normalizing the first data according to the preset normalization rule to obtain the normalized first data comprises:

17. The method of claim 10, further comprising, after the training of the first model with the value of the objective function within the expected value range as the activation probability model,:

determining the importance level of each item of first data based on the influence of each item of first data on the predicted activation probability output by the activation probability model, wherein the first data comprises the user characteristic data and/or the user service characteristic data;

determining a weight coefficient of each item of first data according to the importance level of each item of first data, wherein the weight coefficient of the first data is positively correlated with the importance level of the first data.

18. The method of claim 10, wherein the user traffic characteristic data comprises one or more of:

19. A user activated device, comprising:

the acquisition module is used for acquiring user characteristic data of a tested user and user service characteristic data of the tested user;

the operation module is used for inputting the user characteristic data and the user service characteristic data of the detected user into a pre-trained activation probability model to obtain the predicted activation probability of the detected user output by the activation probability model, the activation probability model comprises a regression tree, and leaf nodes of the regression tree are obtained by training according to the user characteristic data of a sample user and the user service characteristic data of the sample user, so that the difference between the predicted activation probability output by the activation probability model and the actual activation probability is within an expected value range;

a determining module, configured to determine target users based on the predicted activation probability of the detected users, where the target users are at least part of the detected users;

and the sending module is used for sending activation touch information to the target user, wherein the activation touch information is used for improving the probability of the activity rise of the service used by the target user.

20. A model training apparatus, comprising:

the acquisition module is used for acquiring sample data, wherein the sample data comprises user characteristic data, user service characteristic data and an activation state of a sample user;

a model building module for building a first model based on the sample data, the first model comprising a regression tree including leaf nodes determined according to the sample data;

the calculation module is used for calculating an objective function of the first model according to the regression tree, wherein the objective function is used for representing the difference between the predicted activation probability and the actual activation probability output by the first model;

and the training module is used for taking the first model of which the value of the target function is within the range of expected values as the trained activation probability model.

21. A user activated device, comprising: a processor and a memory storing computer program instructions;

the processor, when executing the computer program instructions, implements the user activation method of any of claims 1 to 9.

22. A model training apparatus, comprising: a processor and a memory storing computer program instructions;

the processor, when executing the computer program instructions, implements the model training method of any one of claims 10 to 18.

23. A computer storage medium having computer program instructions stored thereon which, when executed by a processor, implement a user activation method as claimed in any one of claims 1 to 9 or implement a model training method as claimed in any one of claims 10 to 18.