CN111882420A

CN111882420A - Generation method of response rate, marketing method, model training method and device

Info

Publication number: CN111882420A
Application number: CN202010455846.XA
Authority: CN
Inventors: 宋亚南; 李谦; 林亚臣
Original assignee: Mashang Consumer Finance Co Ltd
Current assignee: Mashang Consumer Finance Co Ltd
Priority date: 2020-05-26
Filing date: 2020-05-26
Publication date: 2020-11-03
Anticipated expiration: 2040-05-26
Also published as: CN111882420B

Abstract

A generation method, a marketing method, a model training method and a device of response rate are provided, wherein the generation method comprises the following steps: extracting behavior records of a target user related to a preset service in a service database; determining a target user classification to which the target user belongs according to the behavior record of the target user; selecting a target response rate model corresponding to the target user classification from response rate models which are established in advance and correspond to different user classifications; and generating a response probability value of the target user according to the target response rate model, wherein the response probability value is used for representing the response probability of the target user to a preset interaction behavior. The embodiment of the invention can improve the generation efficiency of the response rate and improve the accuracy of the response rate result.

Description

Generation method of response rate, marketing method, model training method and device

Technical Field

The invention relates to the technical field of information processing, in particular to a response rate generation method, a marketing method, a model training method and a model training device.

Background

The response rate refers to a probability of response to a particular behavior. The response rate reflects the user's acceptance of a particular behavior, and may in turn be used to guide the production and development process to provide the user with a product/service that is easier to accept, improving the user experience of the product/service.

At present, one way to obtain the response rate is to collect the response result of the user to the specific behavior in a manual manner, and then obtain the response rate of the user to the specific behavior through manual analysis and calculation. The method depends on manual analysis and calculation, the calculation efficiency is low, and in addition, the method is difficult to generate more accurate response rate.

Disclosure of Invention

At least one embodiment of the invention provides a response rate generation method, a marketing method, a model training method and a model training device, which can improve the response rate generation efficiency and the accuracy of response rate results.

According to an aspect of the present invention, at least one embodiment provides a method for generating a response rate, including:

extracting behavior records of a target user related to a preset service in a service database;

determining a target user classification to which the target user belongs according to the behavior record of the target user;

selecting a target response rate model corresponding to the target user classification from response rate models which are established in advance and correspond to different user classifications;

and generating a response probability value of the target user according to the target response rate model, wherein the response probability value is used for representing the response probability of the target user to a preset interaction behavior.

According to another aspect of the present invention, at least one embodiment further provides a marketing method for a lending client, where the method described above is used to obtain a probability value of a response rate of a target client, determine a first value interval where the probability value of the response of the target client is located, and select a first trigger policy corresponding to the first value interval from a target trigger policy library as a trigger policy of the target client.

According to another aspect of the present invention, at least one embodiment further provides a method for training a response rate model, including:

according to the collected behavior records of the plurality of users related to the preset service, carrying out user classification on the plurality of users to obtain users under each user classification;

according to the response results of the users to the preset interactive behaviors, respectively marking labels on the users, wherein the labels are used for indicating whether the users respond to the preset interactive behaviors;

and aiming at each user classification, respectively generating training data comprising the user attribute characteristics of each user under the user classification and the corresponding label, and training a response rate model by using the training data of the user classification to obtain the response rate model corresponding to the user classification.

According to another aspect of the present invention, at least one embodiment further provides a response rate generation apparatus, including:

the first record extraction module is used for extracting behavior records of a target user related to a preset service in the service database;

the first classification determining module is used for determining the classification of the target user to which the target user belongs according to the behavior record of the target user;

and the response probability acquisition module is used for selecting a target response rate model corresponding to the target user classification from response rate models which are established in advance and correspond to different user classifications, and generating a response probability value of the target user according to the target response rate model, wherein the response probability value is used for expressing the response probability of the target user to a preset interaction behavior.

According to another aspect of the present invention, at least one embodiment further provides a device for training a response rate model, including:

the second record extraction module is used for carrying out user classification on the plurality of users according to the collected behavior records of the plurality of users related to the preset service to obtain the users under each user classification;

a second classification determining module, configured to respectively mark a tag on each of the multiple users according to a response result of the multiple users to a preset interaction behavior, where the tag is used to indicate whether the user responds to the preset interaction behavior;

and the model training module is used for respectively generating training data comprising the user attribute characteristics of each user under the user classification and the corresponding label aiming at each user classification, and training a response rate model by using the training data of the user classification to obtain the response rate model corresponding to the user classification.

According to another aspect of the present invention, there is also provided a processing apparatus comprising: a processor, a memory and a program stored on the memory and executable on the processor, which when executed by the processor implements the steps of the method as described above.

According to another aspect of the invention, at least one embodiment provides a computer readable storage medium having a program stored thereon, which when executed by a processor, performs the steps of the method as described above.

Compared with the prior art, the response rate generation method, the marketing method, the model training method and the model training device provided by the embodiment of the invention classify the users according to the behavior records of the users, and calculate the response rate of the users under the user classification by using the response rate models corresponding to different user classifications, so that a targeted response rate model is provided based on the user classification of the users, the pertinence of response rate calculation can be improved, and the accuracy of response rate results is further improved. In addition, as the response rate is calculated through the response rate model generated in advance, the embodiment of the invention can also improve the generation efficiency of the response rate and reduce the manual calculation cost.

Drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:

FIG. 1 is a flow chart of a method for generating a response rate according to an embodiment of the present invention;

FIG. 2 is another flow chart of a response rate training method according to an embodiment of the present invention;

FIG. 3 is a flowchart illustrating an exemplary method for generating a response rate according to an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a response rate generating apparatus according to an embodiment of the present invention;

FIG. 5 is another schematic structural diagram of a response rate generator according to an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of a response rate training apparatus according to an embodiment of the present invention;

fig. 7 is another structural diagram of a training apparatus for response rate according to an embodiment of the present invention.

Detailed Description

Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

The terms first, second and the like in the description and in the claims of the present application are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are, for example, capable of operation in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus. In the description and in the claims "and/or" means at least one of the connected objects.

The following description provides examples and does not limit the scope, applicability, or configuration set forth in the claims. Changes may be made in the function and arrangement of elements discussed without departing from the spirit and scope of the disclosure. Various examples may omit, substitute, or add various procedures or components as appropriate. For example, the described methods may be performed in an order different than described, and various steps may be added, omitted, or combined. In addition, features described with reference to certain examples may be combined in other examples.

As described in the background art, the response rate generation method in the prior art generally has the problems of low generation efficiency and low accuracy, and in order to solve at least one of the above problems, embodiments of the present invention provide a response rate generation method, which can improve the generation efficiency of the response rate and improve the accuracy of the response rate result.

As shown in fig. 1, a method for generating a response rate according to an embodiment of the present invention includes:

and 11, extracting behavior records of the target user and the preset service in the service database.

Here, at least one of an occurrence time, an occurrence frequency, a duration, a place where the behavior occurs, and a content of the behavior of the target user related to the preset service may be extracted from a service database. The behavior record is a record of the behavior of the target user associated with the preset service. The preset service may be set according to different application scenarios, for example, for a service database of a shopping website, the preset service may be a commodity purchase service; for another example, for a financial loan transaction database, the predetermined transaction may be a loan transaction or the like. The behavior record may include the occurrence time, occurrence frequency, duration, place where the behavior occurs, and the specific content of the behavior, such as using/purchasing/reserving preset services, the occurrence time and the cost (such as the amount of money spent) of the above-mentioned using/purchasing/reserving behavior, and the like.

For example, in a scenario where the embodiment of the present invention is applied to the field of financial loan, the preset transaction may be a loan transaction, and the behavior record may be at least one of a usage time, a usage amount, and a usage frequency of a credit line provided by the loan transaction.

And step 12, determining the target user classification to which the target user belongs according to the behavior record of the target user.

Here, the embodiment of the present invention may preset a user classification manner based on the behavior record, and then classify the target user into a corresponding user classification (for convenience of description, the user classification to which the target user belongs is referred to as a target user classification) according to the user classification manner. Specifically, the target user may be classified into corresponding user classifications according to at least one parameter of occurrence time, occurrence frequency, duration, place where the behavior occurs, and content of the behavior in the behavior record of the target user, where different user classifications correspond to different values of the at least one parameter.

For example, a plurality of user classifications corresponding to different levels of activity may be pre-partitioned by the time and/or frequency of occurrence of the activity in the activity record, e.g., a first user classification corresponding to a first level of activity, a second user classification corresponding to a second level of activity, etc. And then, determining the target user classification to which the target user belongs according to the time and/or frequency of behavior occurrence in the behavior record of the target user. Here, the activity level is characterized by the time and/or frequency of occurrence of the behavior.

And step 13, selecting a target response rate model corresponding to the target user classification from response rate models corresponding to different pre-established user classifications, and generating a response probability value of the target user according to the target response rate model, wherein the response probability value is used for expressing the response probability of the target user to a preset interaction behavior.

Here, in the embodiment of the present invention, corresponding response rate models are obtained by training in advance for different user classifications, so that in step 13, the embodiment of the present invention may generate a response probability value of the target user according to the response rate model corresponding to the target user classification to which the target user belongs (for convenience of description, the response rate model is referred to as a target response rate model), so as to determine the response probability of the target user to the preset interaction behavior, that is, determine the response rate of the target user to the preset interaction behavior.

Through the steps, the embodiment of the invention respectively establishes the corresponding response rate models for the users of different user classifications, thereby generating the targeted response rate probability based on the target user classification of the target user and improving the accuracy of the response rate result.

Before the step 11, the embodiment of the present invention may further establish a response rate model corresponding to different user classifications by the following steps:

1) extracting behavior records of a plurality of users in a service database, which are related to a preset service, and performing user classification on the plurality of users according to the behavior records to obtain users under each user classification;

2) according to the response result of the plurality of users to the preset interaction behavior, labeling the plurality of users, wherein the label is used for indicating whether the users respond to the preset interaction behavior;

3) and aiming at each user classification, generating training data comprising the user attribute characteristics of each user under the user classification and the corresponding label, and training the response rate model by using the training data of the user classification to obtain the response rate model corresponding to the user classification.

Here, the user attribute feature of the user may be constructed based on the user attribute information stored in the service database. The user attribute information may include user basic attribute information, and the user basic attribute information may specifically include at least one of gender (male, female, or null value), age, native place, living condition, marital status (married, not married, divorced, or unknown), academic calendar (student, college subject, college, high school, beginning and middle school, etc.), work information (such as work unit name, work place, industry category, work age, and position, etc.). The user attribute information may further include a user service attribute associated with the preset service. According to different application scenes, the user service attributes have different contents. For example, in the case of financial loan, the user service attribute may further include at least one of the following information:

a) pre-loan user information, such as pre-loan user asset information and the like;

b) the loan behavior information comprises one or more of the type, the frequency, the credit line and the application time of applying for loan products by the user;

c) the credit-in-line behavior information comprises at least one of limit information, repayment information and withdrawal information, wherein the limit information comprises at least one of change condition, use condition and remaining limit of credit limit, and the repayment information comprises at least one of time of last repayment, repayment amount of last m months and repayment times; the cash withdrawal information comprises one or more of time from last successful cash withdrawal, last n months cash withdrawal amount and cash withdrawal times, and m and n are preset positive numbers.

The user attribute information includes a plurality of category variables and numerical variables, and the category variables may be digitized by one-hot (for example, for gender, male is 1, and female is 0, so as to construct a continuous feature.

Here, after the feature construction is completed, the response rate model is constructed. The embodiment of the invention can adopt various regression prediction algorithms, decision tree regression, xgboost models, neural networks and other algorithms to construct the response rate model.

According to at least one embodiment of the present invention, after step 13, the embodiment of the present invention may further set the trigger policy of the target user according to the response probability value of the target user. The trigger policy may provide a trigger mode or a trigger means for the target user, and the like.

For example, as an implementation manner, when setting the trigger policy of the target user, a target trigger policy library corresponding to a target user category to which the target user belongs may be selected from pre-established trigger policy libraries corresponding to different user categories; and determining a first value interval where the response probability value of the target user is located, and selecting a first trigger strategy corresponding to the first value interval from the target trigger strategy library as the trigger strategy of the target user.

For another example, as another implementation manner, when the trigger policy of the target user is set, a preset attribute feature corresponding to the target user category may be selected; according to the preset attribute characteristics of the target user, performing attribute classification on the target user, determining the target attribute classification to which the target user belongs, and obtaining a target classification combination comprising the target user classification and the target attribute classification; selecting a target trigger strategy library corresponding to the target classification combination from pre-established trigger strategy libraries corresponding to different classification combinations; and determining a first value interval where the response probability value of the target user is located, and selecting a first trigger strategy corresponding to the first value interval from the target trigger strategy library as the trigger strategy of the target user.

The method for generating the response rate in the embodiment of the invention comprises the following steps: extracting behavior records of a target user related to a preset service in a service database; determining a target user classification to which the target user belongs according to the behavior record of the target user; selecting a target response rate model corresponding to the target user classification from response rate models which are established in advance and correspond to different user classifications; and generating a response probability value of the target user according to the target response rate model, wherein the response probability value is used for representing the response probability of the target user to a preset interaction behavior. The embodiment of the invention can improve the generation efficiency of the response rate and improve the accuracy of the response rate result.

Another embodiment of the present invention further provides a method for training a response rate model, as shown in fig. 2, including the following steps:

and step 21, according to the collected behavior records of the plurality of users related to the preset service, performing user classification on the plurality of users to obtain users under each user classification.

Here, the behavior record may include at least one of an occurrence time, an occurrence frequency, a duration, a place where the behavior occurs, and a content of the behavior related to the preset service. When the users are classified, the users can be classified into corresponding user classifications according to at least one parameter of the occurrence time, the occurrence frequency, the duration, the place where the behavior occurs and the content of the behavior in the behavior record of the users, wherein different user classifications correspond to different values of the at least one parameter.

And step 22, respectively marking labels on the plurality of users according to the response results of the plurality of users to the preset interactive behaviors, wherein the labels are used for indicating whether the users respond to the preset interactive behaviors or not.

Here, the response result of the user to the preset interaction behavior may be acquired manually, and a corresponding tag is marked to the user according to whether the user responds to the preset interaction behavior, for example, a tag value "1" indicates that the user responds to the preset interaction behavior, and a tag value "0" indicates that the user does not respond to the preset interaction behavior. In addition, the preset interaction behavior can be set according to different service scenes and application requirements.

And step 23, respectively generating training data including the user attribute characteristics and the corresponding labels of the users under the user classification for each user classification, and training a response rate model by using the training data of the user classification to obtain the response rate model corresponding to the user classification.

Here, the embodiment of the present invention may collect the user attribute information of the multiple users, construct the user attribute features based on the user attribute information, then use the user attribute features of each user under each user classification as the prediction variables of the model, use the labels of each user as the target variables of the model, and generate the training data under the user classification. And carrying out supervised training on the response rate model corresponding to each user classification by utilizing the training data of each user classification to obtain the response rate model corresponding to the user classification.

The method may adopt the method for generating the response rate in the above embodiment to obtain the response rate probability value of the target client, then determine a first value interval in which the response probability value of the target user is located, and select a first trigger policy corresponding to the first value interval from a target trigger policy library as the trigger policy of the target user. The trigger policy may be used to provide a trigger mode or a trigger means for the target user, and the like.

Similarly, as an implementation manner, when determining the trigger policy of the target user, the embodiment may select, from pre-established trigger policy bases corresponding to different user categories, a target trigger policy base corresponding to a target user category to which the target user belongs; and determining a first value interval where the response probability value of the target user is located, and selecting a first trigger strategy corresponding to the first value interval from the target trigger strategy library as the trigger strategy of the target user.

As another implementation manner, in the embodiment, when the trigger policy of the target user is set, a preset attribute feature corresponding to the target user classification may be selected; according to the preset attribute characteristics of the target user, performing attribute classification on the target user, determining the target attribute classification to which the target user belongs, and obtaining a target classification combination comprising the target user classification and the target attribute classification; selecting a target trigger strategy library corresponding to the target classification combination from pre-established trigger strategy libraries corresponding to different classification combinations; and determining a first value interval where the response probability value of the target user is located, and selecting a first trigger strategy corresponding to the first value interval from the target trigger strategy library as the trigger strategy of the target user.

The above scheme of the embodiment of the present invention will be described in detail by taking the marketing response rate of the financial loan field to the user as an example through a plurality of examples. The marketing of the user or the marketing of the user is a specific application scene of the response rate, and after the method provided by the embodiment of the invention is applied, the response rate of the marketing can be improved, and the marketing cost is reduced. In addition, the user is also sometimes referred to herein as a customer.

With the development of big data technology and the financial field, the demand of marketing to obtain customers is rising, and the cost of obtaining new customers in the financial industry is higher and higher. At present, enterprises obtain a large number of users through various promotion modes and high cost, but many clients become diving clients never using the credit line after credit granting, or the credit line is used for a short time and is not used any more subsequently. Therefore, the credit management of the users becomes a very important link in the whole life cycle of the users, and how to more accurately and effectively screen users who are continuously silent or have lost for upgrading to be active by using the lowest cost becomes more and more important and also receives more and more attention from enterprises.

Currently, many companies perform customer promotion activities for their existing stock customers, mainly based on the current collected basic attributes and behavior data in credits of the customers to make manual decisions. The basic attributes commonly used include age, gender, region, etc.; the commonly used behavior data comprises the availability ratio of the quota, the repayment condition, the days from the last borrowing time to the present and the like. By analyzing the relationship between the customer response rate and the customer response rate, after a large amount of combined data analysis is carried out, a corresponding rule or a rule combination mode is formulated to determine a marketing scheme for marketing, for example, marketing is carried out on users who are 30 to 40 years old, have credit line of more than or equal to 5000 yuan and have the latest borrowing time of 90 to 180 days from now.

This whole link needs the analyst that experience is very abundant to accomplish, screens the method that the target customer confirms marketing scheme through the rule moreover, because needs manual handling, leads to the treatment effeciency low, and utilizes the rule to screen and have certain limitation, and the marketing is accurate inadequately, leads to marketing effect unstable, if the rule changes, marketing effect just can not guarantee yet.

To aid in understanding the following examples, a brief description of concepts that may be involved follows.

Credit line granting: the credit line is a credit line which can be used by a client by a bank loan institution or a non-bank loan institution according to the credit condition, the qualification condition and the like of the client.

Managing in loan: and managing the client who obtains the credit line to prompt the client to use the credit line and repay the credit line according to the time.

In the following example, the service database may specifically be a loan service database, the preset service may specifically be a loan service, and the behavior record related to the preset service may specifically be a usage record of a credit line provided by the loan service. When the user is classified, the classification is specifically performed according to the time of using the credit line. The triggering policy described above may specifically be a marketing policy. The trigger policy library described above may specifically be a marketing policy table. The preset interactive behavior described above may specifically be a marketing campaign or a marketing test campaign.

Referring to fig. 3, a method for generating a response rate according to another embodiment of the present invention includes:

step 31, extracting at least one of the service time, the service amount and the service frequency of the credit line of the target user recorded in the loan service database; and determining the target user classification to which the target user belongs according to at least one of the use time, the use amount and the use frequency of the credit line of the target user and a preset activity degree classification mode of the use of the credit line.

Here, the user types of this example include a first user type and a second user type, where the user of the first user type is a user who has not used the credit line, and the user of the second user type is a user who has used the credit line last time for a time exceeding a preset time threshold.

Here, the service database stores usage records of credit lines of each user, and the usage records may generally include data such as time of usage lines, usage amount of lines, and whether debt is settled. The embodiment of the invention mainly reflects the activity degree of the user through the using time of the quota. This example classifies users according to how active they are using credit lines. The activity degree may be divided by the usage of the credit line by the user, specifically, the activity degree may be divided into a first user type that the credit line has not been used, and a second user type that the time of using the credit line last time exceeds a preset time threshold (e.g., 3 months). Users of the first user type are also sometimes referred to herein as silent users, and users of the second user type are referred to herein as active-to-silent users.

Step 32, selecting a target marketing response rate model corresponding to the target user classification from response rate models which are established in advance and correspond to different user classifications; and generating a response probability value of the target user according to the target marketing response rate model.

Here, the marketing response rate model of the present example includes a first model corresponding to the first user type and a second model corresponding to the second user type. In this example, a marketing response rate model for different user types of customer groups is obtained by training in advance through a machine learning algorithm, and the method specifically includes: the model comprises a first model corresponding to the first user type and a second model corresponding to the second user type.

When the response rate probability value of the target user is generated, the user attribute information of the target user recorded in a loan service database can be extracted, and the user attribute feature of the target user is generated according to the user attribute information of the target user; then, the user attribute characteristics of the target user are input into the target response rate model, and the response probability value of the target user output by the target response rate model is obtained.

The response probability value of the target user can be obtained by using the model, the response probability value can be represented by the model score output by the model, and the relationship between the model and the response probability value is related to a specific model and the training mode thereof, for example, the higher the model score is, the higher the response probability value is, or the higher the model score is, the lower the response probability value is.

Through the above steps, the present example can obtain a response probability value of the target user.

Still further, the present example may also set the marketing strategy of the target user through the following step 33.

And step 33, setting a marketing strategy of the target user according to the type of the target user to which the target user belongs and the response probability value, and outputting the marketing strategy to a service execution system.

After the response probability value of the target user is obtained, the marketing strategy of the target user is set according to the user type of the target user and the response probability value. There are many different implementations, and several specific implementations are provided below:

one implementation of setting the marketing strategy of the target user may be:

s1) selecting a target marketing policy table corresponding to a user type of the target user from a first marketing policy table corresponding to the first user type and a second marketing policy table corresponding to a second user type.

Here, the first and second marketing strategy tables record different value intervals of the response probability value and the corresponding marketing strategies.

S2), according to the first value interval where the response probability value of the target user is located, selecting a first marketing strategy corresponding to the first value interval from the target marketing strategy table as the marketing strategy of the target user.

Assuming that the response probability value is 0-1, the value intervals can be divided for the guest groups of different user types. For example, for a first user type, 3 value intervals of [0,0.3), [0.3,0.7) and [0.7,1] are divided; and dividing 2 numerical value intervals of [0,0.6 ] and [0.6,1] for the second user type. And respectively setting corresponding marketing strategies aiming at each numerical value interval of the customer groups of different user types. Assuming that the target user is of the first user type and the response probability value is 0.9, a marketing strategy with a guest group corresponding to the first user type and a response probability value interval of [0.7,1] can be selected as the marketing strategy of the target user, and output to the service execution system, so that the service execution system can execute corresponding marketing operation based on the marketing strategy of the target user.

Another implementation of setting the marketing strategy of the target user may be:

s1) based on the target user type of the target user, selecting preset attribute characteristics corresponding to the target user classification.

S2) according to the preset attribute characteristics of the target user, performing attribute classification on the target user, determining the target attribute classification to which the target user belongs, and obtaining a target classification combination comprising the target user classification and the target attribute classification.

Here, the preset attribute feature is an attribute according to which the attribute classification is performed on the user, and may be selected according to needs and specific application scenarios.

For example, for a user of the first user type, the credit line may be selected as an attribute of the attribute classification, specifically, a plurality of thresholds of the credit line may be set, the credit line may be divided into different line intervals according to the thresholds, and users in the same line interval may be classified as the same user. For example, a user whose credit line is 1 ten thousand yuan or more is regarded as a high-rated user, a user whose credit line is 5000 to less than 1 ten thousand is regarded as a medium-rated user, and a user whose credit line is less than 5000 is regarded as a low-rated user. For another example, for a user of the second user type, the user arrears may be selected as the attribute for attribute classification, and specifically, the user may be classified into a settled user who settles arrears and an unsettled user who unsettles arrears according to whether the user arrears are greater than 0.

S3) selecting a target trigger policy library corresponding to the target classification combination from pre-established trigger policy libraries corresponding to different classification combinations, where different classification combinations are pre-set, each classification combination corresponding to one combination of a user classification and an attribute classification, and a corresponding marketing policy table is set for the classification combinations, and different value intervals of response probability values and corresponding marketing policies thereof are recorded in the marketing policy table.

S4) determining a first value interval where the response probability value of the target user is located, and selecting a first trigger strategy corresponding to the first value interval from the target trigger strategy library as the trigger strategy of the target user.

Here, assuming that the response probability value is 0 to 1, the user classification and the attribute classification may be combined, and the value intervals are respectively divided for different combinations of guest groups. For example, for a guest group of a first user type and high-rated users, 3 numerical value intervals of [0,0.3), [0.3,0.7) and [0.7,1] are divided; and for the second user type and the passenger group of the clear users, dividing 2 numerical value intervals of [0,0.6 ] and [0.6,1 ]. And respectively setting corresponding marketing strategies aiming at each numerical value interval of the passenger groups with different combinations. Assuming that the target user is a first user type and a high-rate user, and the response probability value is 0.9, a marketing strategy corresponding to the first user type + high-rate guest group and the response probability value interval [0.7,1] may be selected as the marketing strategy of the target user.

Through the steps, the marketing response rate model constructed in advance is used for calculating the user response probability value, and compared with a processing mode of manually carrying out rule matching in the prior art, the marketing response rate model based on the user selection method can screen the target users through the model, does not need heavy analysis work before each marketing activity, can guarantee marketing results, improves marketing processing efficiency and reduces labor cost. In addition, the embodiment of the invention establishes corresponding models aiming at the guest groups of different user types, can also divide each type of guest group into users without responding to the probability value interval according to the output result of the models, and uses different marketing strategies for the users with different responding probability value intervals, thereby realizing targeted awakening activities, reducing the cost of managing the activities and improving the accuracy rate and the success rate of the marketing activities.

Before the step 31, the embodiment of the present invention may further establish a marketing response rate model according to the above steps:

step A, according to the use record of the credit line granted by the stock user in the service database, classifying the type of the stock user to obtain a first stock user of a first user type and a second stock user of a second user type.

Here, according to at least one embodiment of the present invention, in the step a, the embodiment of the present invention extracts usage records of credit line granted by the stock users from the service database, and performs cleaning and denoising processing on the extracted data, thereby obtaining a usage record set including credit line granted by a plurality of stock users.

Here, the service database may be one or more, and the embodiment of the present invention may be executed by a service analysis module independent of the service database. When there are a plurality of service databases, a distributed storage system may be used to store the usage records among the plurality of service databases, so as to ensure the storage security of data and improve the data access efficiency. And a safe data transmission channel is established between the business analysis module and the business database, and the use record of the stock user is extracted from the business database through the data transmission channel.

In addition, after the use records of the inventory users are extracted, the embodiment of the invention can carry out processing such as cleaning and noise removal on the extracted data so as to improve the training efficiency of subsequent models. Data cleansing is the re-examination and verification of data, including deleting duplicate information, correcting existing errors (e.g., completing the data), and providing data consistency, obtaining data with the same format. The noise removal processing is to remove data with errors or anomalies (deviation from expected values), and may specifically use a distance-based detection algorithm (such as a k-nearest neighbor method), a detection algorithm based on a statistical method, and a detection algorithm based on density clustering.

And B, according to the response result of the marketing test activities performed on the first inventory user and the second inventory user, marking labels on the first inventory user and the second inventory user, wherein the labels are used for indicating whether the marketing test activities are responded or not.

And step C, generating first training data according to the user attribute characteristics of the first stock user and the corresponding labels, and generating second training data according to the user attribute characteristics of the second stock user and the corresponding labels.

Here, the inventory user may be a user in a previously collected loan or a user in an existing loan of a financial company. To train the model, embodiments of the present invention may classify the inventory users into user types and then perform marketing test activities on the inventory users to obtain a label of whether the inventory users successfully respond. For example, if a target action related to marketing is executed (e.g., using a credit line) within a preset time period after the user is marketed, the response may be considered successful, otherwise the response may be considered failed. Through marketing test activities, whether the users respond or not can be labeled, so that supervised learning is carried out.

During model training, user attribute features also need to be generated, so that corresponding training data is generated for different user types of customer groups, and the training data comprises the user attribute features and labels. Specifically, the user attribute features that may be employed by the embodiments of the present invention may be constructed based on one or more of the following information:

1) user basic information including at least one of age, gender, marital status, administrative region, occupation, position, and year of employment of the user;

2) pre-loan user information, such as pre-loan user asset information and the like;

3) the loan behavior information comprises at least one of the type, the frequency, the credit line and the application time of applying for loan products by the user;

4) the loan behavior information can specifically comprise at least one of amount information and repayment information withdrawal information, wherein the amount information comprises at least one of change condition, use condition and remaining amount of a credit line, and the repayment information comprises at least one of time of the latest repayment, repayment amount of the latest m months and repayment times; the cash withdrawal information comprises at least one of time from last successful cash withdrawal, last n months cash withdrawal amount and cash withdrawal times. Here, m and n are preset positive numbers.

Step D, training a marketing response rate model by using the first sample data to obtain the first model corresponding to the first user type; and training a marketing response rate model by using the second sample data to obtain the second model corresponding to the second user type.

In addition, various regression prediction algorithms, decision tree regression, xgboost models, neural networks and other algorithms can be adopted to construct the marketing response rate model. And training the model corresponding to the user type by respectively utilizing the training data of the guest groups of different user types.

The embodiment of the invention provides a method for training a response rate model, and marketing activities are carried out by utilizing the trained response rate model, so that marketing resources of a certain financial company are saved, the labor cost is reduced, the accurate marketing is realized, the activity efficiency is improved, the response condition of activities of users in stock is analyzed, the marketing response rate model is established, the users are divided into three grades with high, medium and low response rates through the model, and awakening activities with different strategies are carried out.

In this example, the user classes include two classes, namely "silent users" and "active to silent users", respectively. The basis of the user classification is still the usage record of the credit line. In addition, aiming at the silent user, the attribute is the credit line according to the attribute when the attribute classification is carried out; aiming at 'active to silent users', whether the loan is finished or not is judged according to the attribute when the attribute classification is carried out, namely, finished users and unfinished users are judged.

Target customer base for marketing response rate model: and (4) storing the user.

Marketing responsiveness model usage scenarios: the method is mainly applied to the inventory user management activities.

This example mainly includes:

1. model training

(1) Sample collection

And carrying out a marketing test activity for a period of time on the inventory users meeting the conditions, and taking the users obtained by the activity as modeling samples. And then preprocessing the modeling sample according to the user label and the activity execution time. For example, a test marketing campaign may be conducted for inventory users with credit time above 3 months, within the contract's validity period, and not currently overdue. According to the test, the label of the user management activity and the execution time of the specific activity can be obtained, and the part of the users are modeling samples.

According to the rule of screening users according to marketing activities, one user can be marketed for many times, so that the modeling is used for carrying out duplicate removal processing on the sample. If the rules of the test campaign allow multiple marketing campaigns for one user, the modeling samples need to be preprocessed before modeling. For example, the deduplication processing rule may be as follows:

for a user, obtaining the latest marketing data which is not responded by a plurality of marketing activities;

the earliest marketing data is taken for a plurality of marketing activities of one user;

and taking data of an earlier time in response to multiple marketing activities of a user, wherein the responses are not responded.

(2) Sample analysis

After the modeling sample is obtained, the modeling sample data needs to be analyzed, and meaningful information for modeling is mined. The inventory users can be divided into the following categories: a) silent users, b) active to silent users and c) active users. User engagement in credit is primarily for silent users and active to silent users, and thus the user categories considered in this example include only a and b as described above. And dividing the modeling samples into two types of silent user samples and active-to-silent user samples, and analyzing the response rate conditions and behavior data of the two types of samples.

The response rate analysis of the two passenger groups can know the difference between the passenger groups, and the passenger group with higher response rate is easier to activate. And the model can be contrastively analyzed after being constructed, whether the promotion degree of the model meets the requirement or not is judged, and whether the activity response rate of the user is far larger than that of a normal marketing activity or not is screened by using the model.

For silent users, the relationship between the user attribute information and data of application behaviors, such as credit line, days of application time and current, and response rate, can be analyzed. For the active-to-silent user, besides the basic information and loan behavior information of the user, the more important loan behavior information (such as repayment behavior and withdrawal behavior of the user) is also included, and the relationship between the credit line utilization rate of the user, the days from the last repayment/withdrawal and the current day and the like and the response rate can be analyzed. By analyzing the behavior data, the information which has a strong relation with the response rate can be known, and a direction is provided for variable design.

In order to further understand sample data, variables with strong business are analyzed, and the influence of the variables such as the last borrowing date from the start days of the activity, whether the principal balance is 0, the last repayment date from the start days of the activity and the like on the response rate can be found through analysis, so that the method has guiding significance on the customer group subdivision during modeling and later model use.

(3) Layered passenger group and marking by user

The guest groups are hierarchically classified as users. Here, the customer group of the modeling sample is divided into silent users and active-to-silent users for modeling respectively according to whether the user has a borrowing behavior and the number of days since the last borrowing. On one hand, the data dimensionality of the two classes of passenger groups is different, the dimensionality of the actively-silent users is richer, behavior data in credit of the users exist, on the other hand, the response rates of the two classes of passenger groups are different, and the models obtained by adopting the same model are not reasonable. According to the response conditions of the silent user and the active silent user found by analyzing the sample, the behavior records are greatly different. Thus, the final determination is to model separately for both classes of guest groups.

Here, a silent user may refer to a user who has not borrowed for several months after the credit has been completed; an actively silent user may refer to a user who has a historical debit record but has not yet been debited for the last few months. For example, a silent user is a user who is credited with completing three months without borrowing and/or consuming; b. the active-to-silent user refers to a user who has a historical borrowing/consumption record but does not have borrowing/consumption behavior in the last three months.

Marking by a user: and marking whether the user responds according to the marketing activity period. And if the user borrows in the activity period, judging the user to respond, otherwise, judging the user not to respond. The active period can be defined according to actual conditions. For example, each marketing campaign is typically 7 days, and if the user borrows during the campaign (specifically, according to the start time and the end time of the campaign in the campaign system table), the user is determined to be responsive, otherwise, the user is not responsive.

(4) Variable design

Before model training, training data under each user classification needs to be generated, and the training data comprises user attribute features and corresponding labels of various users. In the variable design, the user attribute information of each user under the user classification recorded in the loan service database is extracted, and corresponding user attribute characteristics are generated according to the user attribute information of each user. The user attribute information of the user is fully utilized, the characteristics of the user are interpreted in all directions to construct variables, and the user attribute characteristics are generated.

For the user classification of "silent users", the user attribute features can be constructed based on the information of the users before loan, the basic information of the users and the information of loan behaviors.

For the user classification of 'active-to-silent users', user attribute characteristics can be constructed based on pre-loan user information, user basic information, loan behavior information and loan behavior information.

The pre-loan user data may be pre-loan user assets, etc. Specific examples of the basic information of the user, the loan behavior information, and the loan behavior information are shown in table 1:

TABLE 1

(5) Model training

In this example, the marketing response rate model training uses the xgboost algorithm, which is a supervised learning, that is, a learning with a label in the training data. In step 1, a part of modeling samples with labels are available, the label of the user response is 1, and the label of the non-response is 0. The modeling characteristics are designed according to the 4 th step variable, and the marketing response rate model needs to judge the probability of user response according to the learned characteristics from the modeling samples.

The task of the marketing response rate model training is to find the best parameter group to obtain the better model effect. Dividing samples into a training set and an off-time sample set (OOT), training a model on the training set, and performing model verification on the off-time sample set. Model training is carried out on a training set by adopting an xgboost algorithm, model parameters are continuously adjusted to obtain the optimal model effect, the effect is evaluated by adopting a cross validation method, whether the model effect is relatively stable or not and whether the model effect has better discrimination or not are judged, and overfitting of the model is prevented. And performing model verification on the cross-time sample set, and performing model evaluation according to the distribution of model scores on the cross-time sample set.

(6) Model evaluation

And (4) observing the differentiation degree of the model by observing indexes such as AUC, KS and the like output by the model. Machine Learning (ML) is a field of specialized research for computers to simulate humans or to master learning abilities to acquire new knowledge or skills. Wherein auc (area Under curve) is defined as the area Under the ROC curve, and is a performance index for measuring the quality of the learner. The closer the AUC is to 1.0, the better the classifier works. And the KS index is a performance index for measuring the discrimination of the model, and the larger the numerical value is, the better the effect is.

It is assumed here that the higher the model score, the greater the probability of response. Sequencing and grouping the models from high to low, and counting response rate related information of each group (Quantile), wherein the response rate related information specifically comprises the following information:

the Quantile: grouping, taking values of 5%, 10%, 15%, …, 100%, for example, 95% represents the first 5% of users of the model, i.e., the users with the highest response probability of 5%.

Dividing a model into: the probability value of the model output, here assuming that the higher the model score, the greater the probability of the response.

Response number: the number of response users per interval;

the interval recall rate: the proportion of the response number of each interval to the total response number;

cumulative recall: the threshold value is cut off to the proportion of the response number of the interval to the total response number;

interval response rate: sample response rate for each interval;

cumulative response rate: the threshold is cut off to the sample response rate for this interval.

According to Quantille grouping information, the response rate of the first 20% of guest groups and the proportion of the number of users responding in the first 20% of guest groups to the total number of responding users can be obtained, and the average response rate of the first 20% of users can be obtained by calculation and is several times higher than that of the average response rate. One possible response rate related information is shown in table 2:

TABLE 2

(7) Model online

After the model is built, the model needs to be online, and after the model is online, the model result and the customer group grouping condition can be automatically pushed to a marketing system.

2. Model application

According to the verification result, the difference of the overall response rate of the unclosed user and the cleared user is large, and the distribution of the model is different. The people who have different credit limits also have different credit limits. Therefore, the example carries out more detailed passenger group layering to carry out accurate marketing, and is mainly divided into high and low credit lines, settlement and unsettling. And for different passenger groups, the result is divided into high, medium and low third-grade response rates according to the model. The specific marketing strategy may be: the customer group with high response rate carries out short message reminding marketing, the customer group with medium response rate sends coupons or coupons to stimulate marketing, the customer group with low response rate can select whether to carry out marketing according to the actual conditions of a company, according to statistics, after the customer group is subdivided by using the model of the example to carry out marketing, the activity response rate of 10% of the users in the front of the model is 3-4 times that of the users in the front of the model, therefore, the targeted awakening activity is carried out, the cost of management activity can be reduced, and the activity response rate is improved.

3. Activity execution

According to the list plan and the screening strategy, the screening task is established in the marketing system and the corresponding parameters are configured, after the auditing task passes, the task is activated to be automatically executed, so that the labor cost is greatly reduced, the marketing can be carried out after the marketing system configures the corresponding parameters, and the method is friendly to non-experienced personnel.

Before each marketing activity, marketing colleagues configure a determined model subthreshold value in a marketing system according to a list plan, screen out target users, set different marketing strategies according to high, medium and low three-gear response rates, submit the marketing task, and activate the task to execute automatically after the auditing task passes.

The method mainly aims at the conditions that the marketing success rate of the existing credit inventory users is low and the effect is unstable, the information (including basic information, behavior information in credit and pre-credit information) of the users is fully utilized, a marketing response rate model is built by using a machine learning algorithm, the users are subdivided according to model results, then the users with high, medium and low three-grade response rates are divided, the awakening activities are pertinently carried out, the marketing strategies of the users with different grades are different, the cost of managing the activities is reduced, and the activity response rate is improved. From the above, in the example, the marketing response rate model is constructed by using the machine learning algorithm, so that the response probability of the user is obtained more accurately. Has higher accuracy than the previous method. In addition, the customer group is subdivided according to the model result, so that accurate marketing of the user is realized, and marketing resources are saved.

Various methods of embodiments of the present invention have been described above. An apparatus for carrying out the above method is further provided below.

Referring to fig. 4, an embodiment of the present invention provides a response rate generating apparatus 30, including:

a first record extracting module 41, configured to extract a behavior record of a target user in a service database, where the behavior record is related to a preset service;

a first classification determining module 42, configured to determine, according to the behavior record of the target user, a target user classification to which the target user belongs;

a response probability obtaining module 43, configured to select a target response rate model corresponding to the target user category from response rate models corresponding to different user categories, and generate a response probability value of the target user according to the target response rate model, where the response probability value is used to indicate a response probability of the target user to a preset interaction behavior.

Optionally, the generating device further includes:

the model training module is used for extracting behavior records of a plurality of users in a service database, which are related to a preset service, and performing user classification on the plurality of users according to the behavior records to obtain users under each user classification; according to the response result of the plurality of users to the preset interaction behavior, labeling the plurality of users, wherein the label is used for indicating whether the users respond to the preset interaction behavior; and aiming at each user classification, generating training data comprising the user attribute characteristics of each user under the user classification and the corresponding label, and training the response rate model by using the training data of the user classification to obtain the response rate model corresponding to the user classification.

Optionally, the generating device further includes:

and the strategy setting module is used for setting the trigger strategy of the target user according to the response probability value of the target user.

According to at least one embodiment of the present invention, the policy setting module is further configured to select a target trigger policy library corresponding to the target user category from pre-established trigger policy libraries corresponding to different user categories; and determining a first value interval where the response probability value of the target user is located, and selecting a first trigger strategy corresponding to the first value interval from the target trigger strategy library as the trigger strategy of the target user.

According to at least one embodiment of the present invention, the policy setting module is further configured to select a preset attribute feature corresponding to the target user classification; according to the preset attribute characteristics of the target user, performing attribute classification on the target user, determining the target attribute classification to which the target user belongs, and obtaining a target classification combination comprising the target user classification and the target attribute classification; selecting a target trigger strategy library corresponding to the target classification combination from pre-established trigger strategy libraries corresponding to different classification combinations; and determining a first value interval where the response probability value of the target user is located, and selecting a first trigger strategy corresponding to the first value interval from the target trigger strategy library as the trigger strategy of the target user.

According to at least one embodiment of the present invention, the first record extracting module 41 is further configured to extract at least one of an occurrence time, an occurrence frequency, a duration, a place where an action occurs, and a content of the action of the target user related to the preset service from a service database;

the first classification determining module 42 is further configured to classify the target user into corresponding user classifications according to at least one parameter of occurrence time, occurrence frequency, duration, place where the behavior occurs, and content of the behavior in the behavior record of the target user, where different user classifications correspond to different values of the at least one parameter.

According to at least one embodiment of the present invention, in a case where the service database is a loan service database and the preset service is a loan service:

the model training module is also used for extracting the use record of the user on the credit line in the loan service database, wherein the use record comprises at least one of the use time, the use line and the use frequency of the credit line; determining user classifications to which the users belong according to at least one of the using time, the using amount and the using frequency of each user on the credit line and a preset active degree classification mode for using the credit line, wherein different user classifications correspond to different active degrees for using the credit line; extracting user attribute information of each user under the user classification recorded in a loan service database, and generating corresponding user attribute characteristics according to the user attribute information of each user, wherein the user attribute information comprises at least one of the following information:

user basic information including one or more of age, gender, marital status, administrative region, occupation, position, and year of employment of the user;

pre-loan user information, wherein the pre-loan user information comprises pre-loan user asset information;

the loan behavior information comprises one or more of the type, the frequency, the credit line and the application time of applying for loan products by the user;

the credit-in-line behavior information comprises at least one of limit information, repayment information and withdrawal information, wherein the limit information comprises at least one of change condition, use condition and remaining limit of credit limit, and the repayment information comprises at least one of time of last repayment, repayment amount of last m months and repayment times; the cash withdrawal information comprises one or more of time from last successful cash withdrawal, last n months cash withdrawal amount and cash withdrawal times, and m and n are preset positive numbers.

According to at least one embodiment of the present invention, the first record extracting module 41 is further configured to extract at least one of the usage time, usage amount, and usage frequency of the credit line granted by the target user, which are recorded in the loan service database;

the first classification determining module 42 is further configured to determine a target user classification to which the target user belongs according to at least one of a usage time, a usage amount, and a usage frequency of the target user for the credit line, and a preset activity classification manner for using the credit line;

the response probability obtaining module 43 is further configured to extract the user attribute information of the target user recorded in the loan service database, and generate the user attribute feature of the target user according to the user attribute information of the target user; and inputting the user attribute characteristics of the target user into the target response rate model to obtain the response probability value of the target user output by the target response rate model.

Referring to fig. 5, a schematic structural diagram of a response rate generating device according to an embodiment of the present invention is shown, where the training device 500 includes: a processor 501, a transceiver 502, a memory 503, a user interface 504, and a bus interface.

In the embodiment of the present invention, the marketing device 500 further includes: a program stored 503 in memory and executable on the processor 501.

The processor 501, when executing the program, implements the following steps:

It can be understood that, in the embodiment of the present invention, when being executed by the processor 501, the computer program can implement each process of the embodiment of the response rate generating method described in fig. 1, and can achieve the same technical effect, and in order to avoid repetition, details are not repeated here.

In fig. 5, the bus architecture may include any number of interconnected buses and bridges, with one or more processors represented by processor 501 and various circuits of memory represented by memory 503 being linked together. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, power management circuits, and the like, which are well known in the art, and therefore, will not be described any further herein. The bus interface provides an interface. The transceiver 502 may be a number of elements including a transmitter and a receiver that provide a means for communicating with various other apparatus over a transmission medium. For different user devices, the user interface 504 may also be an interface capable of interfacing with a desired device externally, including but not limited to a keypad, display, speaker, microphone, joystick, etc.

The processor 501 is responsible for managing the bus architecture and general processing, and the memory 503 may store data used by the processor 501 in performing operations.

In some embodiments of the invention, there is also provided a computer readable storage medium having a program stored thereon, which when executed by a processor, performs the steps of:

When being executed by a processor, the program can realize all the implementation modes in the method for generating the response rate, and can achieve the same technical effect, and the details are not repeated here to avoid repetition.

An embodiment of the present invention further provides a device for training a response rate model, as shown in fig. 6, including:

the second record extraction module 61 is configured to perform user classification on the multiple users according to the collected behavior records of the multiple users related to the preset service, so as to obtain a user under each user classification;

a second classification determining module 62, configured to respectively mark a label on each of the multiple users according to a response result of the multiple users to a preset interaction behavior, where the label is used to indicate whether the user responds to the preset interaction behavior;

and a model training module 63, configured to generate, for each user classification, training data including the user attribute features and corresponding labels of each user under the user classification, and train a response rate model using the training data of the user classification to obtain a response rate model corresponding to the user classification.

Optionally, when the service database is a loan service database, and the preset service is a loan service:

the model training module is also used for extracting the user attribute information of each user under the user classification recorded in the loan service database, generating corresponding user attribute characteristics according to the user attribute information of each user, and combining the user attribute characteristics with the label of the user to generate training data under the user classification; wherein the user attribute information includes at least one of the following information:

Referring to fig. 7, a schematic structural diagram of a training apparatus for a response rate model according to an embodiment of the present invention is shown, where the training apparatus 700 includes: a processor 701, a transceiver 702, a memory 703, a user interface 704 and a bus interface.

In the embodiment of the present invention, the marketing device 700 further includes: programs stored on the memory 703 and executable on the processor 701.

The processor 701 implements the following steps when executing the program:

It can be understood that, in the embodiment of the present invention, when being executed by the processor 701, the computer program can implement the processes of the embodiment of the response rate training method described in fig. 2, and can achieve the same technical effect, and in order to avoid repetition, details are not repeated here.

In fig. 7, the bus architecture may include any number of interconnected buses and bridges, with one or more processors, represented by processor 701, and various circuits, represented by memory 703, being linked together. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, power management circuits, and the like, which are well known in the art, and therefore, will not be described any further herein. The bus interface provides an interface. The transceiver 702 may be a number of elements including a transmitter and a receiver that provide a means for communicating with various other apparatus over a transmission medium. The user interface 704 may also be an interface capable of interfacing with a desired device for different user devices, including but not limited to a keypad, display, speaker, microphone, joystick, etc.

The processor 701 is responsible for managing the bus architecture and general processing, and the memory 703 may store data used by the processor 701 in performing operations.

When being executed by a processor, the program can realize all the implementation modes in the training method for the response rate, and can achieve the same technical effect, and the details are not repeated here to avoid repetition.

Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment of the present invention.

In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: various media capable of storing program codes, such as a U disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and all the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for generating a response rate, comprising:

2. The method of claim 1, wherein prior to the step of extracting the record of the behavior of the target user with respect to the preset service from the service database, the method further comprises:

extracting behavior records of a plurality of users in a service database, which are related to a preset service, and performing user classification on the plurality of users according to the behavior records to obtain users under each user classification;

according to the response result of the plurality of users to the preset interaction behavior, labeling the plurality of users, wherein the label is used for indicating whether the users respond to the preset interaction behavior;

and aiming at each user classification, generating training data comprising the user attribute characteristics of each user under the user classification and the corresponding label, and training the response rate model by using the training data of the user classification to obtain the response rate model corresponding to the user classification.

3. The method of claim 1, further comprising:

and setting a trigger strategy of the target user according to the response probability value of the target user.

4. The method of claim 3, wherein the step of setting the trigger policy of the target user according to the response probability value of the target user comprises:

selecting a target trigger strategy library corresponding to the target user classification from pre-established trigger strategy libraries corresponding to different user classifications;

and determining a first value interval where the response probability value of the target user is located, and selecting a first trigger strategy corresponding to the first value interval from the target trigger strategy library as the trigger strategy of the target user.

5. The method of claim 3, wherein the step of setting the trigger policy of the target user according to the response probability value of the target user comprises:

selecting preset attribute features corresponding to the target user classification;

according to the preset attribute characteristics of the target user, performing attribute classification on the target user, determining the target attribute classification to which the target user belongs, and obtaining a target classification combination comprising the target user classification and the target attribute classification;

selecting a target trigger strategy library corresponding to the target classification combination from pre-established trigger strategy libraries corresponding to different classification combinations;

6. The method of claim 2,

the step of extracting the behavior record of the target user related to the preset service in the service database comprises the following steps: extracting at least one of occurrence time, occurrence frequency, duration, behavior occurrence place and behavior content of the target user related to the preset service from a service database;

the step of determining the target user classification to which the target user belongs according to the behavior record of the target user comprises the following steps: and classifying the target user into corresponding user classifications according to at least one parameter of the occurrence time, the occurrence frequency, the duration, the occurrence place of the behavior and the content of the behavior in the behavior record of the target user, wherein different user classifications correspond to different values of the at least one parameter.

7. The method of claim 2, wherein in the case where the service database is a loan service database and the predetermined service is a loan service:

the step of extracting behavior records of a plurality of users related to the preset service in the service database comprises the following steps: extracting the use record of the user on the credit line in the loan service database, wherein the use record comprises at least one of the use time, the use line and the use frequency of the credit line;

the step of classifying the plurality of users according to the behavior record includes: determining user classifications to which the users belong according to at least one of the using time, the using amount and the using frequency of each user on the credit line and a preset active degree classification mode for using the credit line, wherein different user classifications correspond to different active degrees for using the credit line;

the step of generating the user attribute features of the users under the user classification comprises the following steps: extracting user attribute information of each user under the user classification recorded in a loan service database, and generating corresponding user attribute characteristics according to the user attribute information of each user, wherein the user attribute information comprises at least one of the following information:

8. The method of claim 7,

the step of extracting the behavior record of the target user related to the preset service in the service database comprises the following steps: extracting at least one of the service time, the service amount and the service frequency of the target user to the credit line recorded in the loan service database;

the step of determining the target user classification to which the target user belongs according to the behavior record of the target user comprises the following steps: determining a target user classification to which the target user belongs according to at least one of the use time, the use amount and the use frequency of the credit line of the target user and a preset active degree classification mode of the use of the credit line;

the step of generating the response probability value of the target user according to the target response rate model includes: extracting the user attribute information of the target user recorded in a loan service database, and generating the user attribute characteristics of the target user according to the user attribute information of the target user; and inputting the user attribute characteristics of the target user into the target response rate model to obtain the response probability value of the target user output by the target response rate model.

9. A method for marketing to a lending customer, comprising: the method of any one of claims 1 to 8 is utilized to obtain a response rate probability value of a target client, determine a first value interval in which the response probability value of the target user is located, and select a first trigger policy corresponding to the first value interval from a target trigger policy library as the trigger policy of the target user.

10. A method for training a response rate model, comprising:

11. An apparatus for generating a response rate, comprising:

12. An apparatus for training a response rate model, comprising:

13. A processing device, comprising: a processor, a memory, and a program stored on the memory and executable on the processor,

the processor, for reading a program in a memory, implementing the steps of the method for generating a response rate comprising any one of claims 1 to 8, or implementing the steps of the method for training a response rate model according to claim 10.

14. A computer-readable storage medium, characterized in that a computer program is stored thereon, which, when being executed by a processor, carries out the steps of a method for generating a response rate comprising any one of claims 1 to 8, or a method for training a response rate model according to claim 10.