CN112750043B

CN112750043B - Service data pushing method, device and server

Info

Publication number: CN112750043B
Application number: CN202110047200.2A
Authority: CN
Inventors: 陈珊珊; 郭宏
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2021-01-14
Filing date: 2021-01-14
Publication date: 2024-02-02
Anticipated expiration: 2041-01-14
Also published as: CN112750043A

Abstract

The specification provides a service data pushing method, a service data pushing device and a service data pushing server. Based on the method, in the artificial intelligence field, when the server pushes service data to a user object, firstly determining the guest group type of the user object, and then calling a prediction sub-model matched with the guest group type in a preset target user prediction model to determine whether the user object is a potential target user receiving the service data; the preset target user prediction model comprises a first prediction sub-model which is built based on transverse federal learning according to first sample data of a first data party and second sample data of a second data party, and a second prediction sub-model which is built based on federal migration learning according to the first sample data, the second sample data and the first prediction sub-model; under the condition that the user object is determined to be the target user, the proper target service data is pushed to the user object, so that a better pushing effect can be obtained, and the pushing success rate is improved.

Description

Service data pushing method, device and server

Technical Field

The specification belongs to the technical field of artificial intelligence, and particularly relates to a service data pushing method, device and server.

Background

In many push scenarios of business data (e.g., recommendation scenarios of financial business, etc.), it is often difficult to accurately pre-determine whether a user object to push business data is a potential target user that receives the business data, subject to the model accuracy of the predictive model used. The technical problems that the pushing is inaccurate, the service data pushing effect is poor and the like often exist when the service data is pushed to the user object based on the existing method.

In view of the above problems, no effective solution has been proposed at present.

Disclosure of Invention

The specification provides a service data pushing method, device and server, so that whether a user object is a target user or not can be accurately determined by using a preset target user prediction model established based on transverse federal learning and federal migration learning; and under the condition that the user object is determined to be the target user, corresponding target service data is pushed to the user object, so that a better pushing effect can be obtained, and the pushing success rate is improved.

The service data pushing method provided by the specification comprises the following steps:

acquiring identification information of a user object and characteristic data of the user object; wherein the feature data at least comprises non-business type feature data;

Determining the guest group type of the user object according to the identification information of the user object;

invoking a prediction sub-model matched with the guest group type of the user object in a preset target user prediction model to process the characteristic data of the user object so as to determine whether the user object is a target user; the preset target user prediction model at least comprises a first prediction sub-model and a second prediction sub-model; the first predictor model is established based on transverse federal learning in advance according to first sample data in a first data party and second sample data in a second data party; the second predictor model is established based on federation migration learning by combining data knowledge of the first predictor model according to the first sample data and the second sample data in advance;

and pushing proper target business data to the user object under the condition that the user object is determined to be a target user.

In one embodiment, the guest group type includes: a first, well-characterized guest type and a second, well-characterized guest type.

In one embodiment, the first predictor model matches a first guest type and the second predictor model matches a second guest type.

In one embodiment, the preset target user prediction model is established as follows:

the first server responds to a first training request about a preset target user prediction model, and performs sample data fusion with the second server according to a protocol rule based on transverse federal learning so as to determine a first positive sample user and a first negative sample user in sample users of a first guest group type, and a second positive sample user and a second negative sample user in sample users of a second guest group type; the first server is a server deployed at one side of a first data party, and the held first sample data at least comprises non-business characteristic data of a total sample user, wherein the total sample user comprises sample users of a first customer group type and sample users of a second customer group type; the second server is a server deployed at one side of a second data party, and the held second sample data at least comprises service class characteristic data of a total number of positive sample users;

acquiring non-business type characteristic data of a first positive sample user and non-business type characteristic data of a first negative sample user from the first sample data as first grouping sample data; acquiring non-business type characteristic data of a second positive sample user and non-business type characteristic data of a second negative sample user from the first sample data as second packet sample data;

Training to obtain a first predictor model according to the first grouping sample data;

obtaining a second predictor model through federal transfer learning training according to the second packet sample data and the first predictor model;

and combining the first predictor model and the second predictor model to obtain a preset target user prediction model.

In one embodiment, training to obtain a first predictor model based on the first packet sample data includes:

extracting a plurality of data from non-business type characteristic data of a first positive sample user in the first grouping sample data, and marking the data as first positive sample training data; extracting a plurality of data from non-business type characteristic data of a first negative sample user in the first grouping sample data, and marking the data as first negative sample training data;

combining the first positive sample training data and the first negative sample training data to obtain first class training data;

and training to obtain a first predictor model by using the first type of training data.

In one embodiment, obtaining the second predictor model through federal transfer learning training according to the second packet sample data and the first predictor model comprises:

Extracting a plurality of data from non-business type characteristic data of a second positive sample user in the second packet sample data, and marking the data as second positive sample training data; extracting a plurality of data from non-business type characteristic data of a second negative sample user in the second packet sample data, and marking the data as second negative sample training data;

combining the second positive sample training data and the second negative sample training data to obtain second class training data;

invoking a preset first predictor model to process the second class training data to obtain a corresponding predicted value;

combining the second-class training data with the corresponding predicted value to obtain combined second-class training data;

and training to obtain a second predictor model by using the combined second type training data.

In one embodiment, the first guest type includes a banking payroll guest group and the second guest type includes a non-banking payroll guest group.

In one embodiment, in case the user object is determined to be a target user, the method further comprises:

acquiring and determining portrait labels of the user objects according to the service class feature data of the user objects;

Generating a target pushing rule matched with the user object according to the portrait tag of the user object;

and pushing proper target business data to the user object according to the target pushing rule.

In one embodiment, determining the portrait tag of the user object based on the business class feature data of the user object includes:

invoking a preset user portrait prediction model to process the business class feature data of the user object to obtain a corresponding processing result; the preset user portrait prediction model is established based on longitudinal federation learning according to first sample data and second sample data in advance;

and determining the portrait tag of the user object according to the processing result.

In one embodiment, the target business data includes insurance business products and/or insurance business services that are tailored to the user object; correspondingly, the target pushing rule comprises an insurance business marketing scheme.

In one embodiment, the non-business class feature data includes at least one of: monthly revenue, monthly consumption amount, deposit balance; the service class characteristic data comprises at least one of the following: gift identification, premium, and premium.

The specification also provides a training method of the model, which is applied to a first server deployed at one side of a first data party, and comprises the following steps:

responding to a first training request about a preset target user prediction model, and carrying out sample data fusion with a second server according to a protocol rule based on transverse federal learning to determine a first positive sample user and a first negative sample user in sample users of a first guest group type, and a second positive sample user and a second negative sample user in sample users of a second guest group type; the first server is a server deployed at one side of a first data party, and the held first sample data at least comprises non-business characteristic data of a total sample user, wherein the total sample user comprises sample users of a first customer group type and sample users of a second customer group type; the second server is a server deployed at one side of a second data party, and the held second sample data at least comprises service class characteristic data of a total number of positive sample users;

The present specification also provides a service data pushing device, including:

the acquisition module is used for acquiring the identification information of the user object and the characteristic data of the user object; wherein the feature data at least comprises non-business type feature data;

the determining module is used for determining the guest group type of the user object according to the identification information of the user object;

the calling module is used for calling a predictor model matched with the guest group type of the user object in a preset target user prediction model to process the characteristic data of the user object so as to determine whether the user object is a target user or not; the preset target user prediction model at least comprises a first prediction sub-model and a second prediction sub-model; the first predictor model is established based on transverse federal learning in advance according to first sample data in a first data party and second sample data in a second data party; the second predictor model is established based on federation migration learning by combining data knowledge of the first predictor model according to the first sample data and the second sample data in advance;

And the pushing module is used for pushing the proper target service data to the user object under the condition that the user object is determined to be the target user.

The present disclosure also provides a training device of a model, applied to a first server deployed on a side of a first data side, including:

the fusion module is used for responding to a first training request about a preset target user prediction model, and carrying out sample data fusion with a second server according to a protocol rule based on transverse federal learning so as to determine a first positive sample user and a first negative sample user in sample users of a first guest group type, and a second positive sample user and a second negative sample user in sample users of a second guest group type; the first server is a server deployed at one side of a first data party, and the held first sample data at least comprises non-business characteristic data of a total sample user, wherein the total sample user comprises sample users of a first customer group type and sample users of a second customer group type; the second server is a server deployed at one side of a second data party, and the held second sample data at least comprises service class characteristic data of a total number of positive sample users;

The acquisition module is used for acquiring non-business type characteristic data of a first positive sample user and non-business type characteristic data of a first negative sample user from the first sample data to serve as first grouping sample data; acquiring non-business type characteristic data of a second positive sample user and non-business type characteristic data of a second negative sample user from the first sample data as second packet sample data;

the first training module is used for training to obtain a first predictor model according to the first grouping sample data;

the second training module is used for obtaining a second predictor model through federal transfer learning training according to the second packet sample data and the first predictor model;

and the combination module is used for combining the first predictor model and the second predictor model to obtain a preset target user prediction model.

The present disclosure also provides a server comprising a processor and a memory for storing processor-executable instructions, the processor implementing the steps associated with the push method of business data when executing the instructions.

According to the service data pushing method, device and server provided by the specification, when service data is pushed to a user object, the server can firstly determine the guest group type of the user object, and then can call a prediction sub-model matched with the guest group type in a preset target user prediction model to determine whether the user object is a potential target user which can accept the pushed service data according to the characteristic data of the user object; the preset target user prediction model comprises a first prediction sub-model which is built based on transverse federal learning according to first sample data of a first data party and second sample data of a second data party, and a second prediction sub-model which is built based on federal migration learning according to the first sample data, the second sample data and the first prediction sub-model; under the condition that the user object is determined to be the target user, the proper target service data is pushed to the user object, so that a better pushing effect can be obtained, and the pushing success rate is improved.

Drawings

In order to more clearly illustrate the embodiments of the present disclosure, the drawings that are required for the embodiments will be briefly described below, in which the drawings are only some of the embodiments described in the present disclosure, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of one embodiment of a structural composition of a data processing system to which the method for pushing business data provided in the embodiments of the present specification is applied;

fig. 2 is a flow chart of a method for pushing service data according to an embodiment of the present disclosure;

fig. 3 is a schematic diagram of an embodiment of a pushing method of service data provided by the embodiment of the present disclosure in one scenario example;

FIG. 4 is a flow diagram of a model training method provided by one embodiment of the present disclosure;

FIG. 5 is a schematic diagram of the structural composition of a server according to one embodiment of the present disclosure;

fig. 6 is a schematic structural diagram of a service data pushing device according to an embodiment of the present disclosure;

FIG. 7 is a schematic diagram of the structural composition of a model training apparatus according to one embodiment of the present disclosure;

Fig. 8 is a schematic diagram of an embodiment of a pushing method of service data provided by the embodiment of the present disclosure in one scenario example;

fig. 9 is a schematic diagram of an embodiment of a pushing method of service data provided by the embodiment of the present disclosure in one scenario example;

fig. 10 is a schematic diagram of an embodiment of a pushing method of service data provided by the embodiment of the present disclosure in one scenario example.

Detailed Description

In order to make the technical solutions in the present specification better understood by those skilled in the art, the technical solutions in the embodiments of the present specification will be clearly and completely described below with reference to the drawings in the embodiments of the present specification, and it is obvious that the described embodiments are only some embodiments of the present specification, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are intended to be within the scope of the present disclosure.

Considering the existing service data pushing method, under the condition that data isolation exists, a prediction model for predicting a target user potentially receiving pushed service data is often trained by using sample data in a single data party, then a user object to be pushed with the service data is predicted by using the prediction model, so as to determine whether the user object is a potential target user, and service data pushing is performed on the user object according to a prediction result.

First, since the sample data used in the method for training the prediction model is often only sample data in a single data party, the sample data used is relatively limited, which results in relatively poor model accuracy of the training prediction model.

Second, there may be a large difference in data quality of feature data of different sample users among held sample data. For example, for a part of sample users, the collected characteristics are complete, the sample data contains 30 different characteristics of the sample users, and the data quality of the characteristic data is higher. For another part of sample users, the acquired characteristics are seriously lost, the sample data only comprises 5 different characteristics of the sample users, and the data quality of the characteristic data is lower. At this time, if the feature data of the sample user with serious feature loss and poor data quality is directly used to participate in model training, the model accuracy is obviously affected.

For the root cause of the above problem, the present specification first considers that the first sample data in the first data party and the second sample data in the second data party can be comprehensively utilized by the protocol rule based on the lateral federal learning to expand the sample data for model training.

When the first sample data in the first data party and the second sample data in the second data party are specifically utilized, it is further found that: different data parties have different data channels and data isolation, so that the marks of the different data parties on sample data often have larger differences, and even the problem of data label confusion exists; also, limited by data isolation, it is difficult to directly collaborate on using sample data held by different data parties.

For example, a head office, while not specifically responsible for pushing business data, may hold a full sample of users' non-business class feature data. Whereas a subsidiary specially responsible for service data pushing typically performs service data pushing for a sample user existing in the main company. When pushing business data specifically, the head office often pushes business data of the subsidiary only through a common channel shared with the subsidiary. On the side of the subsidiary, besides using the shared channel, the business data can be pushed through the own channel which the main company does not have. For example, on the side of the head office, record data based on a common channel is labeled as a negative-sample user for sample user a. On the side of the subsidiary, the same sample user a is marked as a positive sample based on the recorded data of the own channel.

In some cases, the sub-company only records the service class feature data of the user (corresponding to the positive sample user) who has successfully pushed, and does not record the related data of the user (corresponding to the negative sample user) who has failed to push.

Thus, a head office tends to hold non-traffic class feature data (which may be denoted as first sample data) for a full number of sample users, while a subsidiary tends to hold traffic class feature data (which may be denoted as second sample data) for a full number of positive sample users (i.e., users for whom traffic data push was successful).

In addition, due to the restrictions of relevant regulations, the head office and the subsidiary are generally unable to directly transfer the data held by each other, and it is difficult to train a model by directly combining the data held by both parties.

In view of the above, the present disclosure further combines specific features in a data processing scenario and a correlation between the first sample data and the second sample data, and before implementation, the first server disposed on the first data side and holding the first sample data and the second server disposed on the second data side and holding the second sample data can distinguish the sample users of the first customer group type with complete features and the sample users of the second customer group type with complete features in the sample users first, and then perform data fusion through cooperation according to protocol rules based on transverse federal learning to obtain first type training data for training the first predictor model corresponding to the first customer group type and second type training data for training the second predictor model corresponding to the second customer group type. Further, first-class training data with higher data quality can be utilized to train to obtain a first predictor model with higher precision; and then, the second type of training data is utilized, and the first predictor model obtained by the previous training is combined, and the data knowledge of the first predictor model is introduced through federal migration learning, so that the influence of incomplete characteristics of the second type of training data on model training is weakened, and the second predictor model with relatively high precision is obtained by training, so that the preset target user prediction model with relatively high precision and good effect on the two guest group types can be obtained.

In this way, in the implementation, after the server accesses the feature data of the user object of the service data to be pushed, the client group type of the user object can be determined first, and then the prediction sub-model matched with the client group type of the user object in the preset target user prediction model can be called to process the feature data of the user object, so as to accurately identify whether the user object belongs to the potential target user which can accept the pushed service data. And under the condition that the user object is determined to be a target user, pushing the proper business data to the user object. Therefore, a better pushing effect can be obtained, the pushing success rate is improved, and the accurate pushing of the business data of the user object is realized.

The embodiment of the specification provides a service data pushing method, which can be particularly applied to a data processing system comprising a first server and a second server. Wherein reference may be made to fig. 1. The first server and the second server can be connected in a wired or wireless manner to perform corresponding data interaction.

In this embodiment, the first server may be specifically understood as a server disposed on a side of the first data party (or referred to as a first data domain, for example, a head office). The first server may hold first sample data in a first data party. The first sample data may be understood as sample data obtained from a first data side. Each piece of the first sample data may specifically include two pieces of data including identification information of the user object (for example, an ID of the user first, a mobile phone number of the user first, a name of the user first, etc.) and non-business type feature data (for example, occupation of the user first, monthly income of the user first, number of violations of the user first, etc.) which is not directly related to the business data and corresponds to the identification information. The non-business class feature data may further include a plurality of different non-business class features.

In this embodiment, the second server may be specifically understood as a server disposed on the second data side (or referred to as a second data domain, for example, a subsidiary). The second server may hold second sample data in a second party. The second sample data may be understood as sample data obtained from a second data side. Each piece of the second sample data may specifically include two parts of data, that is, identification information of the user object and service class feature data directly related to the service data corresponding to the identification information. The traffic class feature data may further comprise a plurality of different traffic class features.

Wherein the identification information of the user object in the second sample data includes the identification information of the user object in the first sample data.

In addition, since there is also data isolation between the first data party and the second data party, the first server cannot directly transfer the held first sample data to the second server. Accordingly, the second server cannot directly transmit the held second sample data to the first server.

In this embodiment, before implementation (modeling stage), as shown in fig. 1, the first server and the second server may cooperate to perform data fusion according to a protocol rule based on horizontal federal learning, and determine a total positive sample user from a total sample user according to first sample data and second sample data respectively held; and separating the first training data (corresponding to the first positive sample user and the first negative sample user in the first guest group type) with complete characteristics and high data quality and the second training data (corresponding to the second positive sample user and the second negative sample user in the second guest group type) with incomplete characteristics and low data quality by distinguishing the first guest group type from the second guest group type in the full sample user.

Then, the first type of training data can be utilized to train to obtain a first predictor model corresponding to the first customer group type with higher precision. And then, utilizing the second type of training data, combining the first predictor model obtained by the previous training, and introducing data knowledge of the first type of training data to reduce the influence of the feature deletion of the second type of training data on model training by federal migration learning to obtain a second predictor model with relatively high precision. And finally, combining the first predictor model and the second predictor model to obtain a preset target user prediction model. Therefore, a preset target user prediction model with high precision for both the first guest group type and the second guest group type can be obtained.

According to the mode, the first server and the second server can obtain the preset target user prediction model with higher precision and better effect through cooperative training.

The first server and/or the second server may hold and call the trained predictive model, or may provide the predictive model to a third server of a third party for use according to a collaboration protocol.

In the specific implementation (pushing stage), a server (which may be a first server, a second server, or a third server) accessing a data object of service data to be pushed may determine a guest group type of a user object according to identification information of the user object; and then, a prediction sub-model matched with the guest group type of the user object in a preset target user prediction model can be called to process the characteristic data of the user object so as to accurately identify whether the user object belongs to a potential target user which can accept the pushed service data; and pushing the proper target service data to the user object under the condition that the user object is determined to be the target user.

Therefore, the user object is more willing to accept the pushed business data, a better pushing effect is obtained, the pushing success rate is improved, and the accurate pushing of the business data of the user object is realized.

In this embodiment, the first server and the second server may specifically include a background server applied to a side of the service data processing platform and capable of implementing functions such as data transmission and data processing. Specifically, the first server and the second server may be, for example, an electronic device having a data operation function, a storage function, and a network interaction function. Alternatively, the first server and the second server may be software programs running in the electronic device to support data processing, storage, and network interaction. In the present embodiment, the number of servers included in the first server and the second server is not particularly limited. The first server and the second server may be one server, or may be several servers, or may be a server cluster formed by several servers.

Referring to fig. 2, an embodiment of the present disclosure provides a method for pushing service data. The method, when embodied, may include the following.

S201: acquiring identification information of a user object and characteristic data of the user object; wherein the feature data at least comprises non-business type feature data;

s202: determining the guest group type of the user object according to the identification information of the user object;

s203: invoking a prediction sub-model matched with the guest group type of the user object in a preset target user prediction model to process the characteristic data of the user object so as to determine whether the user object is a target user; the preset target user prediction model at least comprises a first prediction sub-model and a second prediction sub-model; the first predictor model is established based on transverse federal learning in advance according to first sample data in a first data party and second sample data in a second data party; the second predictor model is established based on federation migration learning by combining data knowledge of the first predictor model according to the first sample data and the second sample data in advance;

s204: and pushing proper target business data to the user object under the condition that the user object is determined to be a target user.

Through the embodiment, the server can accurately identify and determine whether the user object of the service data to be pushed is a potential target user receiving the service data by utilizing a preset target user prediction model established based on the combination of the transverse federal learning and the federal transfer learning according to the sample data respectively held by multiple parties in advance; and pushing proper target service data to the user object under the condition that the user object is determined to be the target user. Therefore, the user object is willing to accept the pushed business data, a better pushing effect is obtained, and the pushing success rate is improved; meanwhile, the user object can obtain better pushing experience.

In one embodiment, the method may be applied specifically to the server side. The server may be a first server disposed on the first data side, a second server disposed on the second data side, or a third server disposed on a third party different from the first data side and the second data side.

The first data party may specifically be a data party holding first sample data. The first sample data may specifically include non-service characteristic data of a total number of sample users. The second data party may specifically be a data party holding second sample data. The first sample data may specifically include service class feature data of a total number of positive sample users. The full positive sample users are a subset of the full sample users. The third party may be specifically distinguished from the first party and the second party, and may not be a third party holding sample data.

In one embodiment, the service class feature data may be specifically understood as feature data directly associated with the service data to be pushed, and the non-service class feature data may be specifically understood as feature data indirectly associated with the service data to be pushed. The service data may specifically include: insurance business products, insurance business services, financial business products, financial business services, and the like. Of course, the above listed business data is only one illustrative. In specific implementation, the service data may further include other types of service data according to specific application scenarios and processing requirements. The present specification is not limited to this.

In one embodiment, the service class feature data and the non-service class feature data may specifically include different types of features according to the service data, the first data party, and the second data party.

The following specifically describes an example in which service data is taken as an insurance service product, a first data party is a bank headquarter, and a second data party is an insurance service subsidiary subordinate to the bank headquarter.

In this embodiment, the non-service feature data may specifically include one or more of the following features: month income, month consumption, repository amount, occupation, etc. Accordingly, the service class feature data may specifically include one or more of the following features: a gift certificate, a historical premium, a risk of historical purchase, and so forth.

In this embodiment, the first server disposed on the first data side may hold the non-service class feature data of the full-volume sample user as the first sample data. Since the first data party is not mainly responsible for pushing service data, the second data party is assisted to push service data through a common channel (e.g. banking, etc.) shared with the second data party. Thus, the positive sample users marked out of the total number of sample users in the first sample data held by the first data side (i.e., users who accept the pushed service data) tend to be incomplete.

The second server deployed at one side of the second data party is mainly responsible for pushing service data. Moreover, due to the association relationship between the second data party and the first data party, the second data party is usually used for pushing related service data mainly for the total number of sample users held by the first data party, and records the sample users (marked as positive sample users) which are successfully pushed and the service class characteristic data of the positive sample users which are successfully pushed.

In addition, the second data party can push the service data by using the shared channel shared with the first data party, push the service data by using the own channels (such as the self-sales channels of insurance companies, etc.) which are not possessed by other first data parties, and record the service class characteristic data of the positive sample user and the positive sample user. For such positive sample users, the first data party is not known nor marked on the first data party side due to the data isolation that exists between the first data party and the second data party.

Thus, the second server can hold the traffic class feature data of the full number of positive sample users as the second sample data. And the full positive sample users described above tend to be also a subset of the full sample users.

In one embodiment, the server may first obtain the identification information of the user object and the feature data of the user object when preparing to push the service data of the user object.

The identification information may specifically include identification information such as a name, an identity ID, a mobile phone number, an account name, etc. for indicating the user object. The feature data of the user object may specifically be non-business class feature data of the user object. Of course, the feature data of the user object may be a combination of non-service class feature data and service class feature data of the user object.

In one embodiment, the server may determine, according to the identification information of the user object, a group type to which the user object belongs by querying data, for example, querying a preset group list.

In one embodiment, the group types may specifically include: a first, well-characterized guest type and a second, well-characterized guest type.

In this embodiment, whether the user object is a target user is determined accurately by dividing the user object based on whether the user object is complete or not and according to the first guest group type with complete characteristics and the second guest group type with incomplete characteristics, so that a matched prediction sub-model in a preset target user prediction model can be called later.

In one embodiment, the first group type may be specifically understood as a group type having more complete and complete characteristics held by the first party. The features may specifically refer to features in non-service feature data. Specifically, for example, the first customer base type may be a banking proxy payroll customer base. For this type of guest group, the non-business type characteristic data (including asset characteristics, transaction characteristics, etc.) that the first party (i.e., the bank) often holds is complete and complete, including, for example, the user's deposit balance, the user's credit card running water, the user's monthly revenue, etc.

In one embodiment, the second group type may be specifically understood as a group type having incomplete features and missing features held by the first data party. The features may specifically refer to features in non-service feature data. Specifically, for example, the second guest group type may be a non-banking-generation payroll guest group. For this type of guest group, the non-business class feature data that the first party (i.e., the bank) would hold is feature missing, incomplete.

Accordingly, the total sample users in the first sample data held by the first server may also include: sample users of a first group of well-characterized, and sample users of a second group of well-characterized. In the first sample data, the non-business characteristic data of the sample user of the first customer group type contains complete characteristics, and the data quality is relatively high; the non-business type characteristic data of the sample user of the second group type contains incomplete characteristics and has the defects, and the data quality is relatively poor.

In one embodiment, when the server is implemented, a predictor model matched with a guest group type can be determined from predictor models contained in a preset target user prediction model according to the guest group type of the user object; and then the matched prediction sub-model can be called to process the characteristic data of the user object, and whether the user object is a target user is determined according to the processing result output by the model.

In one embodiment, the target user may be specifically understood as a potential group of users having a relatively high probability of accepting the pushed traffic data.

In one embodiment, the preset target user prediction model may be specifically understood as a prediction model obtained by the first server and the second server in advance through cooperation with the first sample data and the second sample data held by the first server and the second server respectively and based on transverse federal learning and federal migration learning training, and capable of predicting a probability value of the user object belonging to the target user based on feature data of the user object. The training method of the preset target user prediction model will be described later.

In one embodiment, the specific group types include: in the case of the first guest group type and the second guest group type, the preset target user prediction model may specifically include two different prediction sub-models, i.e., a first prediction sub-model and a second prediction sub-model. Wherein the first predictor model matches a first guest type and the second predictor model matches a second guest type.

In this embodiment, the first predictive sub-model may be specifically obtained by training in advance with non-service feature data (denoted as first class training data) of a sample user of the first client group type, and it may be determined, for the first client group type, whether the user object is a predictive sub-model of the target user. The second predictor model may specifically be a predictor model that is obtained by training in advance using non-service feature data (denoted as second class training data) of a sample user of the second guest group type and in combination with the first predictor model, and is capable of accurately determining whether the user object is a target user for the second guest group type.

By establishing and utilizing the preset target user prediction model comprising the first prediction sub-model corresponding to the first guest group type and the second prediction sub-model corresponding to the second guest group type, whether the user object is a potential target user receiving service data can be determined more precisely.

In one embodiment, in the case that the guest group type of the user object is determined to be the first guest group type, a first predictor model in the preset target user prediction model may be determined to be a matched predictor model; the first predictor model may then be invoked to process the feature data of the user object to determine whether the user object is a target user.

Under the condition that the guest group type of the user object is determined to be the tired heart of the second guest group, a second predictor model in the preset target user prediction model can be determined to be a matched predictor model; in turn, a second predictor model may be invoked to process the feature data of the user object to determine whether the user object is a target user.

In one embodiment, in the event that a user object is determined to be a target user, appropriate target business data may be pushed to the user object.

In one embodiment, in the case that the user object is determined to be the target user, service class feature data of the user object or other associated background data can be further acquired; according to the personalized requirements and the personal preferences of the user object, the user object can be pushed with proper target service data more accurately, the pushing success rate is further improved, and the user object can obtain better pushing experience.

In one embodiment, when the user object is determined to be the target user, the method may further include the following steps when implemented:

s1: acquiring and determining portrait labels of the user objects according to the service class feature data of the user objects;

S2: generating a target pushing rule matched with the user object according to the portrait tag of the user object;

s3: and pushing proper target business data to the user object according to the target pushing rule.

Through the embodiment, the server can more precisely and accurately determine and utilize the portrait tag of the user object to analyze the demand trend and the personal preference of the user object, and further more effectively push the proper target business data to the user object.

In one embodiment, the portrait tag may be specifically understood as tag data for reflecting the requirement or preference of a user object for a certain attribute dimension in the service data that is prone to be accepted.

In one embodiment, in the case that the service data is an insurance service product or an insurance service, and the application scenario is an insurance service marketing scenario, the portrait tag may specifically include one or more of a plurality of portrait tags listed below: a payment term trend label, a premium trend label, and the like. Of course, the image labels listed above are only illustrative.

In specific implementation, the portrait tag may further include other types of portrait tags according to specific service data and specific application scenarios. For example, in the case where the service data is a financial product and the application scenario is a marketing scenario of the financial product, the portrait tag may be an annual rate of return trend tag, a fixed holding time trend tag, a financial risk level trend tag, or the like.

In one embodiment, the determining the portrait tag of the user object according to the service class feature data of the user object may include the following when implemented: invoking a preset user portrait prediction model to process the business class feature data of the user object to obtain a corresponding processing result; the preset user portrait prediction model is established based on longitudinal federation learning according to first sample data and second sample data in advance; and determining the portrait tag of the user object according to the processing result.

Through the embodiment, the portrait tag of the user object can be accurately and efficiently determined, and further, the proper business data can be more accurately pushed to the user object.

In one embodiment, the preset user portrait prediction model may be specifically understood as a prediction model that is obtained by the first server and the second server by pre-cooperation using the first sample data and the second sample data held by the first server and the second server respectively and based on longitudinal federal learning training, and can predict portrait labels about business data matched with the user object based on feature data of the user object.

In one embodiment, the predetermined user portrait prediction model may specifically include a plurality of portrait prediction sub-models. Wherein each of the plurality of image predictor models corresponds to one image tag.

Correspondingly, the server can call a plurality of portrait forecast sub-models contained in a preset user portrait forecast model to respectively process the characteristic data of the user object to obtain a plurality of different portrait labels corresponding to the user object, and further, the personalized demand trend or preference of the user object can be more finely and comprehensively depicted by utilizing the plurality of different portrait labels.

In one embodiment, matching target pushing rules can be customized for the user object in a targeted manner according to the portrait tag of the user object; and then, according to the target pushing rule, the user object can be pushed with proper target service data.

In one embodiment, the target pushing rule may specifically be a construction mode of service data matched with a user object and applicable to the user object, a marketing scheme of the service data, or a recommendation policy of the service data. The target pushing rule may also be rule data of other contents corresponding to different service data and application scenarios. The present specification is not limited to this.

In one embodiment, the server may first determine, according to the portrait tag of the user object, a target pushing rule suitable for the user object in a targeted manner; furthermore, according to the target pushing rule, the target business data with higher matching degree with the user object and higher user object acceptance probability can be configured; and then according to the target pushing rule, selecting a pushing mode which is preferred by the user object and can be timely touched to push the target business data to the user object.

When receiving the target service data pushed according to the target pushing rule, the user object has relatively high probability of being willing to accept the target service data. Therefore, the success rate of pushing can be effectively improved, and a good pushing effect is obtained.

From the above, based on the service data pushing method provided in the embodiment of the present disclosure, when the server pushes service data to the user object, the client group type of the user object may be determined first, and then a prediction sub-model matched with the client group type in the preset target user prediction model may be invoked to determine whether the user object is a target user potentially receiving the pushed service data according to the feature data of the user object; the preset target user prediction model comprises a first prediction sub-model which is built based on transverse federal learning according to first sample data of a first data party and second sample data of a second data party, and a second prediction sub-model which is built based on federal migration learning according to the first sample data, the second sample data and the first prediction sub-model; under the condition that the user object is determined to be the target user, the proper target service data is pushed to the user object, so that a better pushing effect can be obtained, and the pushing success rate is improved.

In one embodiment, the preset target user prediction model (may be denoted as a model a) invoked by the server may be specifically obtained by the first server and the second server through establishment based on lateral federal learning and federal migration according to first sample data in a first data party and second sample data in a second data party respectively held by the first server and the second server in advance.

In an embodiment, in a specific implementation, for the first server side, the preset target user prediction model may be specifically trained in the following manner.

S1: the first server responds to a first training request about a preset target user prediction model, and performs sample data fusion with the second server according to a protocol rule based on transverse federal learning so as to determine a first positive sample user and a first negative sample user in sample users of a first guest group type, and a second positive sample user and a second negative sample user in sample users of a second guest group type; the first server is a server deployed at one side of a first data party, and the held first sample data at least comprises non-business characteristic data of a total sample user, wherein the total sample user comprises sample users of a first customer group type and sample users of a second customer group type; the second server is a server deployed at one side of a second data party, and the held second sample data at least comprises service class characteristic data of a total number of positive sample users;

S2: acquiring non-business type characteristic data of a first positive sample user and non-business type characteristic data of a first negative sample user from the first sample data as first grouping sample data; acquiring non-business type characteristic data of a second positive sample user and non-business type characteristic data of a second negative sample user from the first sample data as second packet sample data;

s3: training to obtain a first predictor model according to the first grouping sample data;

s4: obtaining a second predictor model (which can be marked as a model A2) through federation migration learning training according to the second packet sample data and the first predictor model (which can be marked as a model A1);

s5: and combining the first predictor model and the second predictor model to obtain a preset target user prediction model.

Through the embodiment, the first server can firstly cooperate with the second server through the transverse federal learning framework according to the protocol rule based on the transverse federal learning, and on the premise that the respective held sample data do not go out of the domain, the full positive sample users are determined and marked in the full sample users held at one side of the first server; and classifying the total number of sample users based on the guest group types into sample users of a first guest group type that is well-characterized and sample users of a second guest group type that is not well-characterized. Furthermore, non-business characteristic data of the sample users with higher data quality of the first customer group type can be firstly obtained from the first sample data to be used as first type training data; and training to obtain a first predictor model with higher precision corresponding to the first client group type by using the first type of training data with higher data quality. Meanwhile, non-business type characteristic data of a second guest group type sample user with poor data quality can be obtained from the first sample data to serve as second type training data; and the second type of training data with poor data quality is utilized, and meanwhile, a first predictor model with higher precision obtained by previous training is combined to introduce data knowledge of the first type of training data, so that the influence of incomplete characteristics on model training in the second type of training data is reduced, and a second predictor model with higher precision is obtained. Therefore, the prediction model with higher precision and better effect on the user objects of the first guest group type and the second guest group type can be trained.

In one embodiment, the first training request may be specifically initiated by the first server or may be initiated by the second server. The present specification is not limited to this.

In one embodiment, considering that in some cases there may be data isolation between the first server and the second server, both parties cannot directly interact with each holding the first sample data and the second sample data, respectively. Therefore, the transverse federal learning framework can be utilized to perform safe sample data fusion according to the protocol rule cooperation based on the transverse federal learning so as to break the limitation of data isolation, and meanwhile, the sample data held by the own party cannot be directly leaked to the other party.

Through the above data fusion, the first server may determine and mark a full amount of positive sample users among the full amount of sample users in the first sample data held by the own party (for example, set preset mark information for indicating the positive sample users in the positive sample users); meanwhile, the first server may divide the total sample data into two groups of sample users of the first guest group type and sample users of the second guest group type.

Further, for the sample users of the first group type, the first server may first determine the positive sample user from the sample users of the first group type by retrieving the tag information of the sample users of the first group type, and record the positive sample user as the first positive sample user. And determining the rest sample users except the first positive sample user in the sample users of the first customer group type as negative sample users, and recording the negative sample users as first negative sample users.

For sample users of the second group type, the first server may first determine a positive sample user from the sample users of the second group type by retrieving the tag information of the sample users of the second group type, and record the positive sample user as a second positive sample user. And determining the rest sample users except the second positive sample user in the sample users of the second group type as negative sample users, and recording the negative sample users as second negative sample users.

With the above embodiment, the first server and the second server may determine and identify the first positive sample user and the first negative sample user among the sample users of the first client group type among the total number of sample users held by the first server by cooperating based on the protocol rules of the lateral federation learning; meanwhile, a second positive sample user and a second negative sample user are determined and identified in the sample users of the second group type.

In one embodiment, for a sample user of a first client group type, the first server may obtain, from the held first sample data, non-traffic class feature data of a first positive sample user, non-traffic class feature data of a first negative sample user, as first packet sample data. Therefore, sample data with complete characteristics and higher data quality can be screened out.

Meanwhile, for the sample users of the second group type, the first server can acquire non-business type characteristic data of the second positive sample user and non-business type characteristic data of the second negative sample user from the held first sample data as second packet sample data. Thus, sample data with missing characteristics and poor data quality can be stripped.

In this way, the first server can accurately separate the first component sample data and the second component sample data which are used for training different predictor models aiming at different guest group types from the held first sample data.

In an embodiment, the training to obtain the first predictor model according to the first packet sample data may include the following when implemented.

S1: extracting a plurality of data from non-business type characteristic data of a first positive sample user in the first grouping sample data, and marking the data as first positive sample training data; extracting a plurality of data from non-business type characteristic data of a first negative sample user in the first grouping sample data, and marking the data as first negative sample training data;

s2: combining the first positive sample training data and the first negative sample training data to obtain first class training data;

S3: and training to obtain a first predictor model by using the first type of training data.

Through the embodiment, the first server can independently obtain the first predictor model with higher precision by splitting the first training data with complete characteristics and higher data quality from the first sample data.

In one embodiment, when the first server is specifically trained, an initial model based on logistic regression may be constructed first; and continuously training and adjusting the initial model by using the first type of training data to obtain a first predictor model meeting the requirements.

In one embodiment, referring to fig. 3, the second predictor model is obtained by performing federal migration learning training according to the second packet sample data and the first predictor model, and when implemented, the method may include the following.

S1: extracting a plurality of data from non-business type characteristic data of a second positive sample user in the second packet sample data, and marking the data as second positive sample training data; extracting a plurality of data from non-business type characteristic data of a second negative sample user in the second packet sample data, and marking the data as second negative sample training data;

S2: combining the second positive sample training data and the second negative sample training data to obtain second class training data;

s3: invoking a preset first predictor model to process the second class training data to obtain a corresponding predicted value;

s4: combining the second-class training data with the corresponding predicted value to obtain combined second-class training data;

s5: and training to obtain a second predictor model by using the combined second type training data.

Through the embodiment, when the first server trains the second predictive sub-model, the first predictive sub-model with higher accuracy obtained by previous training is utilized to process second-class training data with incomplete characteristics and poor data quality, and a corresponding predictive value is obtained; combining the second-class training data with the corresponding predicted value to introduce data knowledge of the first-class training data with higher data quality, so as to obtain combined second-class training data with better training effect; and then the combined second-class training data can be used for replacing the original second-class training data with poor data quality to perform model training, so that the influence of model training by independently using the second-class training data with incomplete characteristics and poor data quality on model accuracy can be effectively reduced, and a second predictor model with relatively high accuracy is obtained.

In one embodiment, specifically, referring to fig. 3, the second type of training data may be input into the trained first predictor model, and the first predictor model is operated to output a corresponding predicted value.

In one embodiment, since the features originally included in the second-class training data are missing and incomplete, and the first predictor model is obtained by training using the training data with complete features, the first predictor model cannot directly process the second-class training data. Before invoking the preset first predictor model to process the second class of training data, the method further comprises: preprocessing the second class of training data.

In specific implementation, the first type training data and the second type training data can be subjected to feature comparison, and missing features of the second type training data relative to the first type training data are determined; and adding data bits with missing features into the second class of training data, and setting the numerical values of the data bits with missing features to zero to obtain the preprocessed second class of training data. And further, a preset first predictive sub-model can be called to process the preprocessed second class training data, so that a corresponding predictive value is obtained.

In one embodiment, in the specific combination, the obtained predicted value can be used as a new feature to be combined with the original feature in the second-class training data, so as to expand the original feature space with missing and incomplete second-class training data, and obtain the combined second-class training data with relatively richer features and relatively better training effect.

Specifically, for example, taking a certain piece of training data Xi in the second class of training data as an example, the training data originally only includes 3 features, which are respectively marked as a first feature, a second feature and a fourth feature, and corresponding data values are respectively x1, x2 and x4. Accordingly, xi may be represented as [ x1, x2, x4]. And comparing the second training data with the first training data with complete characteristics, and determining that the missing characteristics in the second training data are the third characteristics and the fifth characteristics. In this case, the training data Xi may be preprocessed to obtain preprocessed training data, and the preprocessed training data is denoted as Xi' as [ x1, x2,0, x4,0]. And inputting the preprocessed training data Xi' into a first predictor model, calling the first predictor model to process the data, and obtaining a corresponding predicted value which is recorded as Px. And combining the predicted value with the original second-class training data Xi to obtain combined second-class training data, wherein the combined second-class training data is marked as Xi' and expressed as [ x1, x2, x4, and Px ]. Therefore, the data knowledge of the first type of training data is introduced by using the first predictor model, and the combined second type of training data with better training effect is obtained.

In one embodiment, when the second class of training data is combined with the corresponding predicted value, the association weight about the predicted value may be further determined according to the degree of feature missing of the second class of training data and the degree of influence of the original features in the second class of training data on the user receiving the service data. And combining the original characteristics in the second-class training data with the product of the associated weights and the predicted values, so that the second-class training data with better training effect can be obtained.

When the association weight of the predicted value is specifically determined, if the degree of feature deletion of the second class of training data is more serious and the degree of influence of the original features in the second class of training data on the service data accepted by the user is smaller, the value of the association weight of the predicted value can be set to be relatively larger. Conversely, if the degree of feature missing of the second class of training data is smaller and the degree of influence of the original features in the second class of training data on the user's acceptance of the service data is larger, the value of the association weight of the predicted value may be set to be relatively smaller.

In one embodiment, the preset user portrait prediction model (may be denoted as model B) called by the server may be specifically obtained by the first server and the second server through establishment based on longitudinal federal learning according to first sample data in the first data party and second sample data in the second data party respectively held by the first server and the second server in advance.

In one embodiment, in the implementation, for the first server side, the preset user portrait prediction model may be specifically trained in the following manner.

S1: the first server responds to a second training request about a preset user portrait prediction model, and cooperates with the second server to perform sample data fusion according to a protocol rule based on longitudinal federal learning to obtain third training data carrying a preset portrait label; wherein the third training data comprises non-business class feature data of the positive sample user; the preset portrait tag is determined according to the service class characteristic data of the positive sample user;

s2: and training to obtain a preset user portrait prediction model by using the third training data carrying the preset portrait tag.

Through the embodiment, the first server can cooperate with the second server by utilizing the longitudinal federal learning framework according to the protocol rule based on longitudinal federal learning, the second server determines the preset portrait tag of the positive sample user by utilizing the held business type characteristic data, and then the preset portrait tag is fused with the non-business type characteristic data of the positive sample user held by the first server to obtain the third training data carrying the preset portrait tag containing the data knowledge of the business type characteristic data; and further, the third training data can be utilized to train to obtain a preset user portrait prediction model which can accurately predict and determine portrait labels of users about service data.

In one embodiment, the second training request may be specifically initiated by the first server or may be initiated by the second server. The present specification is not limited to this.

In one embodiment, considering that there may be data isolation between the first server and the second server in some scenarios, both parties cannot directly interact with each holding the first sample data and the second sample data, respectively. Therefore, the longitudinal federation learning framework can be utilized to perform safe sample data fusion according to protocol rule cooperation based on longitudinal federation learning so as to break the limitation of data isolation, and meanwhile, the data held by own can not be directly leaked to the opposite side.

Through the data fusion, the second server can determine the preset portrait label (such as a payment term trend label, a premium trend label and the like) of the sample user by using the business type characteristic data in the second sample data held by the second server; meanwhile, the first server can acquire non-business type characteristic data of the positive sample user from the held first sample data; and the first server and the second server can combine the preset portrait tag and the non-business feature data corresponding to the same positive sample user through cooperation to obtain the third training data carrying the preset portrait tag.

In one embodiment, during specific fusion, the first server and the second server can also combine the preset portrait tag corresponding to the same positive sample user, and the non-service characteristic data and the service characteristic data through cooperation to obtain relatively more complete and rich third training data carrying the preset portrait tag, so that a more accurate preset user portrait prediction model can be trained.

In one embodiment, when the preset user portrait prediction model is specifically trained, the third training data carrying the preset portrait tag may be used to learn and train the initial model, so as to obtain the preset user portrait prediction model.

In one embodiment, the positive sample user training the pre-set user portrayal prediction model may specifically be a sample user pre-determined using the pre-set target user prediction model and marked as a target user. Therefore, the preset target user prediction model obtained through training and the preset user portrait prediction model have stronger combination.

In one embodiment, the preset user portrait prediction model may specifically include one portrait prediction sub-model, or may include a plurality of portrait prediction sub-models, where different portrait prediction sub-models are used for predicting different portrait labels.

Through the embodiment, a plurality of different portrait labels of the user object can be predicted simultaneously by using a preset user portrait prediction model; furthermore, the demand trend and the personal preference of the user object can be more finely marked according to the plurality of different portrait tags, so that a target pushing rule with higher matching degree and better effect can be generated for the user object.

In one embodiment, in the case that the preset user portrait prediction model includes a plurality of portrait prediction sub-models, the second server may generate a plurality of different preset portrait labels according to different service class feature data of the held positive sample user in the process of data fusion with the first server according to a protocol rule based on longitudinal federal learning. Therefore, the third training data finally obtained through data fusion can contain a plurality of groups of third training data. The third training data of the same group carry the same preset portrait tag.

Correspondingly, during training, a plurality of initial models can be respectively trained by utilizing the plurality of groups of third training data, so as to obtain a plurality of portrait forecast sub-models respectively corresponding to different portrait labels; and combining the plurality of portrait predication sub-models to obtain a preset user portrait predication model.

In one embodiment, taking an insurance business marketing scenario as an example, the target business data may specifically include an insurance business product and/or an insurance business service suitable for a user object; accordingly, the target pushing rule may specifically include an insurance business marketing scheme.

In the insurance business marketing scene, a preset target user prediction model can be utilized to determine whether the current user object to be marketed is a target user which potentially can accept the marketed insurance business; under the condition that the current user object is determined to belong to a target user, a preset user portrait prediction model is utilized to predict portrait labels related to insurance business of the user object, and then the requirement trend and personal preference of the current user object for the insurance business can be determined according to the portrait labels; further, the personalized demand trend and personal preference can be combined to generate an insurance business marketing scheme which is matched with the current user object and has higher acceptance of the current user object; and then, according to the insurance business marketing scheme, the marketing processing of the appropriate insurance business products and/or insurance business services can be accurately carried out to the current user object. Therefore, the marketing success rate can be effectively improved, meanwhile, the user can be accurately recommended to the appropriate insurance service according to the personalized demand trend and preference of the user, and the service experience of the user is improved.

In one embodiment, taking an insurance business marketing scenario as an example, the non-business type feature data may specifically include at least one of the following: monthly income, monthly expense amount, deposit balance, etc.; the service class characteristic data may specifically include at least one of the following: gift identification, premium, etc. Of course, the above listed non-service class feature data, service class feature data are only one illustrative type. In specific implementation, the non-service feature data and the service feature data may further include other types of data according to specific application scenarios and processing requirements. The present specification is not limited to this.

By acquiring and utilizing the diversified business feature data and non-business feature data, a preset target user prediction model and a preset user portrait prediction model which are higher in accuracy and better in effect aiming at insurance business marketing scenes can be obtained through training.

In one embodiment, in a insurance marketing scenario, the first party may specifically be a bank or a bank headquarters. Accordingly, the second party may be in particular an associated insurance sub-company.

In one embodiment, the first guest type may include a banking payroll guest and the second guest type may include a non-banking payroll guest.

According to the embodiment, the guest groups are distinguished according to whether the guest groups are the bank generation wages or not, so that sample users of the first guest group type with complete characteristics and sample users of the second guest group type with incomplete characteristics can be accurately split from a bank or a whole number of sample users held by the bank.

In one embodiment, taking an insurance business marketing scenario as an example, the portrait tag may specifically include at least one of the following: a payment term tendency label, a premium tendency label, and the like. Of course, the above-listed image labels are only illustrative. In specific implementation, the portrait tag may further include other types of tag data according to specific application scenarios and processing requirements. The present specification is not limited to this.

By utilizing the diversified portrait labels, the personalized demand trend and personal preference of the user object about insurance business can be more finely and comprehensively described in the insurance business marketing scene, so that the user object can be more accurately subjected to proper insurance business marketing.

As can be seen from the foregoing, in the method for pushing service data provided in the embodiments of the present disclosure, when service data is pushed to a user object, a guest group type of the user object may be determined first, and then a prediction sub-model matched with the guest group type in a preset target user prediction model may be invoked to determine whether the user object is a potential target user that may accept the pushed service data according to feature data of the user object; the preset target user prediction model comprises a first prediction sub-model which is built based on transverse federal learning according to first sample data of a first data party and second sample data of a second data party, and a second prediction sub-model which is built based on federal migration learning according to the first sample data, the second sample data and the first prediction sub-model; under the condition that the user object is determined to be the target user, the proper target service data is pushed to the user object, so that a better pushing effect can be obtained, and the pushing success rate is improved. When a preset target user prediction model is trained, distinguishing a first customer group type with complete characteristics from a second customer group type with incomplete characteristics, so that a first server holding first sample data in a first data party and a second server holding second sample data in a second data party can perform data fusion through cooperation according to protocol rules based on transverse federal learning to separate and obtain first training data with complete characteristics and second training data with incomplete characteristics; further, the first type of training data with complete characteristics can be utilized to train to obtain a first predictor model with higher precision; and according to the second class training data, combining the first predictor model, reducing the influence of incomplete characteristics on model training in the second class training data through federation transfer learning to obtain a second predictor model with relatively high precision, thereby training to obtain a predictor model with relatively high precision and relatively good effect on user objects of the first guest group type and the second guest group type. Further acquiring and determining the portrait tag of the user object according to the service class characteristic data of the user object under the condition that the user object is determined to be a target user receiving the pushed service data; analyzing personalized demand trend and personal preference of the user object according to the portrait tag of the user object, and generating a target pushing rule suitable for the user object; and according to the target pushing rule, the proper target service data is accurately pushed to the user object, so that the pushing success rate can be further improved.

Referring to fig. 4, the embodiment of the present disclosure further provides a training method of the model, so as to train to obtain a preset target user prediction model meeting the requirements. The method can be particularly applied to a first server deployed on one side of a first data party. The specific embodiments may include the following.

S401: responding to a first training request about a preset target user prediction model, and carrying out sample data fusion with a second server according to a protocol rule based on transverse federal learning to determine a first positive sample user and a first negative sample user in sample users of a first guest group type, and a second positive sample user and a second negative sample user in sample users of a second guest group type; the first server is a server deployed at one side of a first data party, and the held first sample data at least comprises non-business characteristic data of a total sample user, wherein the total sample user comprises sample users of a first customer group type and sample users of a second customer group type; the second server is a server deployed at one side of a second data party, and the held second sample data at least comprises service class characteristic data of a total number of positive sample users;

S402: acquiring non-business type characteristic data of a first positive sample user and non-business type characteristic data of a first negative sample user from the first sample data as first grouping sample data; acquiring non-business type characteristic data of a second positive sample user and non-business type characteristic data of a second negative sample user from the first sample data as second packet sample data;

s403: training to obtain a first predictor model according to the first grouping sample data;

s404: obtaining a second predictor model through federal transfer learning training according to the second packet sample data and the first predictor model;

s405: and combining the first predictor model and the second predictor model to obtain a preset target user prediction model.

In this embodiment, when a preset target user prediction model is trained, distinguishing a first guest group type with complete characteristics from a second guest group type with incomplete characteristics, so that a first server holding first sample data in a first data party and a second server holding second sample data in a second data party perform data fusion through cooperation according to protocol rules based on transverse federal learning, and separate and obtain first training data with complete characteristics and second training data with incomplete characteristics; further, the first type of training data with complete characteristics can be utilized to train to obtain a first predictor model with higher precision; and according to the second class training data, combining the first predictor model, reducing the influence of incomplete characteristics on model training in the second class training data through federation transfer learning to obtain a second predictor model with relatively high precision, thereby training to obtain a predictor model with relatively high precision and relatively good effect on user objects of the first guest group type and the second guest group type.

The embodiment of the specification also provides a server, which comprises a processor and a memory for storing instructions executable by the processor, wherein the processor can execute the following steps according to the instructions when being implemented: acquiring identification information of a user object and characteristic data of the user object; wherein the feature data at least comprises non-business type feature data; determining the guest group type of the user object according to the identification information of the user object; invoking a prediction sub-model matched with the guest group type of the user object in a preset target user prediction model to process the characteristic data of the user object so as to determine whether the user object is a target user; the preset target user prediction model at least comprises a first prediction sub-model and a second prediction sub-model; the first predictor model is established based on transverse federal learning in advance according to first sample data in a first data party and second sample data in a second data party; the second predictor model is established based on federation migration learning by combining data knowledge of the first predictor model according to the first sample data and the second sample data in advance; and pushing proper target business data to the user object under the condition that the user object is determined to be a target user.

In order to more accurately complete the above instructions, referring to fig. 5, another specific server is provided in this embodiment of the present disclosure, where the server includes a network communication port 501, a processor 502, and a memory 503, and the above structures are connected by an internal cable, so that each structure may perform specific data interaction.

The network communication port 501 may be specifically configured to obtain identification information of a user object and feature data of the user object; wherein the feature data at least comprises non-business type feature data.

The processor 502 may be specifically configured to determine a guest group type of the user object according to the identification information of the user object; invoking a prediction sub-model matched with the guest group type of the user object in a preset target user prediction model to process the characteristic data of the user object so as to determine whether the user object is a target user; the preset target user prediction model at least comprises a first prediction sub-model and a second prediction sub-model; the first predictor model is established based on transverse federal learning in advance according to first sample data in a first data party and second sample data in a second data party; the second predictor model is established based on federation migration learning by combining data knowledge of the first predictor model according to the first sample data and the second sample data in advance; and pushing proper target business data to the user object under the condition that the user object is determined to be a target user.

The memory 503 may be used to store a corresponding program of instructions.

In this embodiment, the network communication port 501 may be a virtual port that binds with different communication protocols, so that different data may be sent or received. For example, the network communication port may be a port responsible for performing web data communication, a port responsible for performing FTP data communication, or a port responsible for performing mail data communication. The network communication port may also be an entity's communication interface or a communication chip. For example, it may be a wireless mobile network communication chip, such as GSM, CDMA, etc.; it may also be a Wifi chip; it may also be a bluetooth chip.

In this embodiment, the processor 502 may be implemented in any suitable manner. For example, the processor may take the form of, for example, a microprocessor or processor, and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a programmable logic controller, and an embedded microcontroller, among others. The description is not intended to be limiting.

In this embodiment, the memory 503 may include a plurality of layers, and in a digital system, the memory may be any memory as long as it can hold binary data; in an integrated circuit, a circuit with a memory function without a physical form is also called a memory, such as a RAM, a FIFO, etc.; in the system, the storage device in physical form is also called a memory, such as a memory bank, a TF card, and the like.

The embodiment of the specification also provides a server, which comprises a processor and a memory for storing instructions executable by the processor, wherein the processor can execute the following steps according to the instructions when being implemented: responding to a first training request about a preset target user prediction model, and carrying out sample data fusion with a second server according to a protocol rule based on transverse federal learning to determine a first positive sample user and a first negative sample user in sample users of a first guest group type, and a second positive sample user and a second negative sample user in sample users of a second guest group type; the first server is a server deployed at one side of a first data party, and the held first sample data at least comprises non-business characteristic data of a total sample user, wherein the total sample user comprises sample users of a first customer group type and sample users of a second customer group type; the second server is a server deployed at one side of a second data party, and the held second sample data at least comprises service class characteristic data of a total number of positive sample users; acquiring non-business type characteristic data of a first positive sample user and non-business type characteristic data of a first negative sample user from the first sample data as first grouping sample data; acquiring non-business type characteristic data of a second positive sample user and non-business type characteristic data of a second negative sample user from the first sample data as second packet sample data; training to obtain a first predictor model according to the first grouping sample data; obtaining a second predictor model through federal transfer learning training according to the second packet sample data and the first predictor model; and combining the first predictor model and the second predictor model to obtain a preset target user prediction model.

The embodiment of the specification also provides a computer storage medium of the pushing method based on the service data, wherein the computer storage medium stores computer program instructions, and the computer program instructions are realized when executed: acquiring identification information of a user object and characteristic data of the user object; wherein the feature data at least comprises non-business type feature data; determining the guest group type of the user object according to the identification information of the user object; invoking a prediction sub-model matched with the guest group type of the user object in a preset target user prediction model to process the characteristic data of the user object so as to determine whether the user object is a target user; the preset target user prediction model at least comprises a first prediction sub-model and a second prediction sub-model; the first predictor model is established based on transverse federal learning in advance according to first sample data in a first data party and second sample data in a second data party; the second predictor model is established based on federation migration learning by combining data knowledge of the first predictor model according to the first sample data and the second sample data in advance; and pushing proper target business data to the user object under the condition that the user object is determined to be a target user.

In the present embodiment, the storage medium includes, but is not limited to, a random access Memory (Random Access Memory, RAM), a Read-Only Memory (ROM), a Cache (Cache), a Hard Disk (HDD), or a Memory Card (Memory Card). The memory may be used to store computer program instructions. The network communication unit may be an interface for performing network connection communication, which is set in accordance with a standard prescribed by a communication protocol.

The embodiments of the present specification also provide a computer storage medium storing computer program instructions for implementing the training method based on the above model, when the computer program instructions are executed: responding to a first training request about a preset target user prediction model, and carrying out sample data fusion with a second server according to a protocol rule based on transverse federal learning to determine a first positive sample user and a first negative sample user in sample users of a first guest group type, and a second positive sample user and a second negative sample user in sample users of a second guest group type; the first server is a server deployed at one side of a first data party, and the held first sample data at least comprises non-business characteristic data of a total sample user, wherein the total sample user comprises sample users of a first customer group type and sample users of a second customer group type; the second server is a server deployed at one side of a second data party, and the held second sample data at least comprises service class characteristic data of a total number of positive sample users; acquiring non-business type characteristic data of a first positive sample user and non-business type characteristic data of a first negative sample user from the first sample data as first grouping sample data; acquiring non-business type characteristic data of a second positive sample user and non-business type characteristic data of a second negative sample user from the first sample data as second packet sample data; training to obtain a first predictor model according to the first grouping sample data; obtaining a second predictor model through federal transfer learning training according to the second packet sample data and the first predictor model; and combining the first predictor model and the second predictor model to obtain a preset target user prediction model.

In this embodiment, the functions and effects of the program instructions stored in the computer storage medium may be explained in comparison with other embodiments, and are not described herein.

Referring to fig. 6, on a software level, the embodiment of the present disclosure further provides a service data pushing device, where the device may specifically include the following structural modules.

The acquiring module 601 may be specifically configured to acquire identification information of a user object and feature data of the user object; wherein the feature data at least comprises non-business type feature data;

the determining module 602 may be specifically configured to determine a group type of the user object according to the identification information of the user object;

the invoking module 603 may be specifically configured to invoke a predictor model in a preset target user prediction model, where the predictor model is matched with a guest group type of the user object, to process feature data of the user object, so as to determine whether the user object is a target user; the preset target user prediction model at least comprises a first prediction sub-model and a second prediction sub-model; the first predictor model is established based on transverse federal learning in advance according to first sample data in a first data party and second sample data in a second data party; the second predictor model is established based on federation migration learning by combining data knowledge of the first predictor model according to the first sample data and the second sample data in advance;

The pushing module 604 may be specifically configured to push, when the user object is determined to be the target user, appropriate target service data to the user object.

It should be noted that, the units, devices, or modules described in the above embodiments may be implemented by a computer chip or entity, or may be implemented by a product having a certain function. For convenience of description, the above devices are described as being functionally divided into various modules, respectively. Of course, when the present description is implemented, the functions of each module may be implemented in the same piece or pieces of software and/or hardware, or a module that implements the same function may be implemented by a plurality of sub-modules or a combination of sub-units, or the like. The above-described apparatus embodiments are merely illustrative, for example, the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.

From the above, based on the service data pushing device provided in the embodiment of the present disclosure, whether the user object is a target user can be accurately determined, and in the case that the user object is determined to be the target user, appropriate target service data is pushed to the user object, so that a better pushing effect can be obtained, and the pushing success rate is improved.

Referring to fig. 7, on a software level, the embodiment of the present disclosure further provides a training device for a model, where the device may specifically include the following structural modules.

The fusion module 701 may be specifically configured to perform sample data fusion with a second server according to a protocol rule based on lateral federal learning in response to a first training request about a preset target user prediction model, so as to determine a first positive sample user and a first negative sample user among sample users of a first guest group type, and a second positive sample user and a second negative sample user among sample users of a second guest group type; the first server is a server deployed at one side of a first data party, and the held first sample data at least comprises non-business characteristic data of a total sample user, wherein the total sample user comprises sample users of a first customer group type and sample users of a second customer group type; the second server is a server deployed at one side of a second data party, and the held second sample data at least comprises service class characteristic data of a total number of positive sample users;

The obtaining module 702 may be specifically configured to obtain, from the first sample data, non-service feature data of a first positive sample user and non-service feature data of a first negative sample user, as first packet sample data; acquiring non-business type characteristic data of a second positive sample user and non-business type characteristic data of a second negative sample user from the first sample data as second packet sample data;

the first training module 703 may be specifically configured to train to obtain a first predictor model according to the first packet sample data;

the second training module 704 may be specifically configured to obtain a second predictor model through federal transfer learning training according to the second packet sample data and the first predictor model;

the combination module 705 may be specifically configured to combine the first predictor model and the second predictor model to obtain a preset target user prediction model.

From the above, the training device based on the model provided in the embodiment of the present disclosure can train to obtain a prediction model with higher accuracy and better effect for both the user objects of the first guest group type and the second guest group type.

In a specific scenario, the method provided in the present disclosure may be applied to assist an insurance sub-company subordinate to a bank in performing accurate marketing of insurance service. For specific implementation, reference may be made to the following.

In this scenario example, it is noted that insurance business marketing belongs to an important part of banking marketing business. The insurance types are various, and can comprise risk transfer insurance and financial value-added insurance. Further, the risk transfer insurance may also include life insurance, health insurance, accident insurance, etc. And each type of branch may in turn comprise a plurality of sub-branches. Therefore, the marketing difficulty of insurance is high, and an insurance agent is assisted to accurately mine insurance potential purchasing customers by means of an intelligent model.

Typically, many banks will have associated insurance sub-companies with insurance marketing data that are mostly problematic as follows: 1. data isolation. Due to regulatory policy restrictions, data of insurance sub-companies and banks are not circulated, so that insurance marketing modeling samples are separated and can only be independently modeled based on data of all parties. 2. The non-generation payroll group (corresponding to the second group type) has imperfect portraits and large proportion of total groups, and has poor effect on the marketing model of the groups. Wherein, the non-generation payroll group has a difference in feature distribution compared with the generation payroll group (corresponding to the first guest group type), and the feature missing data proportion is high. In general, the bank has perfect images of the generation payroll guest group, and has good marketing model effect on the guest group.

In this scenario example, before implementation, the training method of the model provided in the present specification may be applied, and the data held by each of the bank and the insurance sub-company is utilized to perform joint modeling through federal learning (including lateral federal learning and federal migration learning).

Wherein, federal learning specifically can refer to that each participant can perform joint modeling by means of other party data in the process of performing machine learning. And all the parties do not need to share the original data, namely, under the condition that the data does not go out of the local area, the data joint training is carried out, and a shared machine learning model is established.

Federal migration learning may specifically refer to joint modeling techniques that combine federal learning and migration learning techniques with little overlap of users and features.

The transverse federation learning specifically refers to that when the sample features of the data set overlap more and the sample IDs overlap less, we align the data according to the feature dimensions by adopting the transverse federation, and take out the part of data with the same sample features and the incomplete user features for training.

In the present scenario example, the above-mentioned problems with secure accurate marketing modeling in bank+secure mode are solved based on federal transfer learning techniques. The bank and insurance sub-company which limit data isolation due to the supervision policy realize the safe fusion of data, and the bank generation wage customer group model with perfect insurance portrait is utilized to optimize the effect of the whole marketing model. In addition, the insurance sample size and feature dimension are also extended. The customer mining accuracy of the insurance marketing model of the bank non-generation guest group with huge volume is obviously improved.

In the specific implementation, the fact that the insurance marketing data of both the bank and the insurance sub-company are not circulated is considered, and data isolation exists. The primary premise of machine learning model training is that the sample data is sufficient and obeys independent co-distribution. The more core features, the more complete the representation, the better the modeling is typically required for sample data. The modeling effect is good, the proportion of the generation payroll guest group to the total guest group of the bank is small, but the image features are comprehensive, and the feature of purchasing and protecting the core features of the personal images such as payroll, income, academic and professional information (for example, the feature data of the complete non-business type) is included, so that high-value users can be effectively mined. The characteristic missing proportion of the non-generation payroll group is extremely high, and particularly, the key characteristic missing proportion is larger in the above purchase protection.

The specific data case can be seen in fig. 8. Thus, mixing two classes of guest modeling, the sample distribution shifts, and the model effect based on such data modeling is greatly affected.

In this scenario example, during specific modeling, the data of the bank and the insurance sub-company may be first fused safely by using the federal learning technology, so as to train the insurance client mining model (corresponding to the first predictor model) of the generation payroll guest group, and obtain the insurance periscope list of the generation payroll guest group by using the model.

And then training an insurance client mining model (corresponding to a second predictor model) of the non-generation payroll client group by using the non-generation payroll client group based on the federal migration technology, and obtaining an insurance potential client list of the client group by using the model, so that knowledge of the non-generation payroll client group can be migrated to the non-generation payroll client group.

A specific training process may be seen in fig. 9. The training process comprises the steps of 1, respectively carrying out data preparation work by banks (or headquarters) and insurance sub-companies; 2. based on the federal learning technology, implementing safe fusion on sample data of banks and insurance sub-companies, and training a federal model of a generation-sending payroll group, namely an insurance client mining model A1; 3. training a federal model of a non-generation payroll group based on federal transfer learning, namely an insurance client mining model A2; 4. and generating a list of potential customers. When the specific model is applied, the insurance agent can distinguish the guest group types, separately use the model, use A1 for the instead of sending payroll guest group, and predict A2 for the non-instead of sending payroll guest group. Specifically, the following steps may be included.

Step 1: the worker and insurance sub-company perform data preparation work, respectively.

Step 2: based on federal learning technology, the sample data of the headquarter and the insurance sub-company are subjected to safe fusion and a federal model of the generation and transmission payroll group, namely an insurance client mining model A1, is trained, and a generation and transmission payroll group insurance periscope list is predicted and generated by using the model.

In the step, according to the data classification condition, the process is based on the traditional transverse federal technology, based on bank characteristics, positive samples of both parties institutions are superimposed, and sample modeling of a wage customer group is replaced.

Step 3: based on the model of the second step, the federal model of the non-generation payroll group, namely an insurance client mining model A2, is trained by utilizing federal migration learning, and an insurance potential customer list is predicted and generated by utilizing the model.

Referring to fig. 8, after data preparation, the data distribution difference between the bank and its insurance sub-company is still relatively large. The total amount of customers held by the bank is divided into a generation payroll customer group and a non-generation payroll customer group. Besides, the method has the following characteristics: insurance purchase channels are divided into self-sales (e.g., owned channels) and escrow (e.g., shared channels). The insurance sub-company holds a full-volume insurance purchase positive sample, including a self-sales positive sample and a silver-insurance positive sample. The customers of the insurance sub-company are typically customers of the bank, i.e. the customers of the insurance sub-company are typically all using the bank account. But under regulatory policy constraints, banks have no positive sample of the insurance sub-company's self-sales channels. Referring specifically to fig. 8, a group (1) represents a customer purchasing insurance of a subsidiary in a headquarter, and the group (1) corresponds to a group (2) of the insurance subsidiary. The guest group (1) and the guest group (2) are both positive samples. The customer group (3) represents the insurance purchasing customer of the insurance sub-company from the sales channel sub-company, and the insurance sub-company side is a positive sample. The group (3) corresponds to the group (4) at the headquarter, and the bank does not know the purchase condition of the group (4), so the group (4) is regarded as a negative sample at the bank side. The bank also contains other mass users, i.e. guest groups (6). The guest group (4) is contained within the guest group (6).

When the non-generation guest group insurance client mining model A2 is specifically constructed, referring to fig. 10, the method includes the following steps:

step S01: acquiring non-generation wage sample data;

step S02: obtaining a substitute wage group insurance marketing model A1;

step S03: estimating the non-agent passenger group by using the model A1 to obtain a potential passenger probability Y (for example, a predicted value);

step S04: adding the above potential customer probability Y as a new feature into a non-generation wage model feature space to obtain a combined feature (for example, combined second-class training data), wherein the data processing logic is the same as that of the generation wage model A1, and the federal migration learning technology is applied and realized;

step S05: based on the features combined as above, the non-generation model A2 is trained using non-generation guest group samples.

Through the scene example, precise marketing modeling of insurance in a bank and insurance mode based on federal transfer learning is realized. The federal migration learning is realized in the insurance marketing field, so that the joint modeling of banks and insurance sub-companies which cannot be realized due to the limitation of data supervision authorities becomes possible; and the high-value portrait information of the bank generation wage guest group is transferred to the large-base bank guest group, so that the accuracy of the overall marketing of the bank insurance is effectively improved, the marketing cost of the insurance agency is saved, and a marketing model with good effect is obtained.

Although the present description provides method operational steps as described in the examples or flowcharts, more or fewer operational steps may be included based on conventional or non-inventive means. The order of steps recited in the embodiments is merely one way of performing the order of steps and does not represent a unique order of execution. When implemented by an apparatus or client product in practice, the methods illustrated in the embodiments or figures may be performed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment, or even in a distributed data processing environment). The terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, it is not excluded that additional identical or equivalent elements may be present in a process, method, article, or apparatus that comprises a described element. The terms first, second, etc. are used to denote a name, but not any particular order.

Those skilled in the art will also appreciate that, in addition to implementing the controller in a pure computer readable program code, it is well possible to implement the same functionality by logically programming the method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. Such a controller can be regarded as a hardware component, and means for implementing various functions included therein can also be regarded as a structure within the hardware component. Or even means for achieving the various functions may be regarded as either software modules implementing the methods or structures within hardware components.

The description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, classes, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

From the above description of embodiments, it will be apparent to those skilled in the art that the present description may be implemented in software plus a necessary general hardware platform. Based on such understanding, the technical solutions of the present specification may be embodied essentially in the form of a software product, which may be stored in a storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., and include several instructions to cause a computer device (which may be a personal computer, a mobile terminal, a server, or a network device, etc.) to perform the methods described in the various embodiments or portions of the embodiments of the present specification.

Various embodiments in this specification are described in a progressive manner, and identical or similar parts are all provided for each embodiment, each embodiment focusing on differences from other embodiments. The specification is operational with numerous general purpose or special purpose computer system environments or configurations. For example: personal computers, server computers, hand-held or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments that include any of the above systems or devices, and the like.

Although the present specification has been described by way of example, it will be appreciated by those skilled in the art that there are many variations and modifications to the specification without departing from the spirit of the specification, and it is intended that the appended claims encompass such variations and modifications as do not depart from the spirit of the specification.

Claims

1. The service data pushing method is characterized by comprising the following steps:

2. The method of claim 1, wherein the group of people type comprises: a first, well-characterized guest type and a second, well-characterized guest type.

3. The method of claim 2, wherein the first predictor model matches a first guest type and the second predictor model matches a second guest type.

4. A method according to claim 3, wherein the pre-set target user prediction model is built in the following way:

5. The method of claim 4, wherein training a first predictor model based on the first packet sample data comprises:

6. The method of claim 4, wherein obtaining a second predictor model from the second packet sample data and the first predictor model by federal transfer learning training comprises:

7. The method of claim 2, wherein the first guest type comprises a banking payroll guest group and the second guest type comprises a non-banking payroll guest group.

8. The method according to claim 1, wherein in case the user object is determined to be a target user, the method further comprises:

9. The method of claim 8, wherein determining the portrait tag of the user object based on business class feature data of the user object comprises:

10. The method of claim 8, wherein the target business data comprises insurance business products and/or insurance business services appropriate for the user object; correspondingly, the target pushing rule comprises an insurance business marketing scheme.

11. The method of claim 8, wherein the non-business class feature data comprises at least one of: monthly revenue, monthly consumption amount, deposit balance; the service class characteristic data comprises at least one of the following: gift identification, premium, and premium.

12. A method for training a model, applied to a first server disposed on a side of a first data party, comprising:

13. A traffic data pushing apparatus, comprising:

14. A training device for a model, applied to a first server disposed on a side of a first data party, comprising:

15. A server comprising a processor and a memory for storing processor-executable instructions, which when executed by the processor implement the steps of the method of any one of claims 1 to 11.