CN112632137A

CN112632137A - Business data pushing method and device and server

Info

Publication number: CN112632137A
Application number: CN202011501811.1A
Authority: CN
Inventors: 李瑾瑜; 宋虎; 饶彭彦; 彭岗
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2020-12-18
Filing date: 2020-12-18
Publication date: 2021-04-09

Abstract

The specification provides a service data pushing method, a service data pushing device and a service data pushing server. Based on the method, an artificial intelligence technology is combined, after the push success rate of the data objects is predicted by using a preset push result prediction model, a first push list which is correspondingly ordered is generated based on the push success rate, the difference degree of the push effect between different data object type groups and the unbiased estimation quantity of the relevance between the push rule and the data objects are further determined, the data is used for adjusting the ordering in the first push list, so that the error caused by the fact that the fairness between the different data object type groups and the relevance between the data objects and the push rule are not considered in the preset push result prediction model is eliminated, and a second push list which is ordered relatively more accurately is obtained; and then, target service data can be pushed to the plurality of data objects according to the second push list, so that a better pushing effect can be obtained.

Description

Business data pushing method and device and server

Technical Field

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

Background

When business data is pushed to a client, a prediction model obtained by training based on an artificial intelligence technology (for example, a supervised learning algorithm) is often used to predict the pushing success rate of the client; sequencing the clients based on the predicted push success rate to generate a push list of the clients; and further pushing corresponding service data to the client according to the push list.

However, due to the limitations of the above prediction models, which are limited by the training mechanism and the factors considered, the models themselves have bias in the process of learning and training. The push lists generated by subsequently sorting the predicted push success rate based on the prediction model also have errors, and the subsequent push effect is influenced.

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

Disclosure of Invention

The specification provides a method, a device and a server for pushing service data, so as to generate a second push list which is more accurate and reasonable in sequencing, and more accurately push target service data to a data object based on the second push list, thereby obtaining a better pushing effect.

The present specification provides a method for pushing service data, including:

acquiring attribute data of a plurality of data objects and historical push records; the historical pushing record records historical service data of a data object pushed based on a historical pushing rule and a corresponding historical pushing result;

calling a preset pushing result prediction model, and determining the pushing success rate of pushing target service data to each data object in the plurality of data objects according to the attribute data of the plurality of data objects;

according to the push success rate, sequencing the identity identifications of the data objects to obtain a first push list;

determining the difference degree of the pushing effect between different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object according to the attribute data of the data objects and the historical pushing record;

adjusting the sequence of the identity identifiers of the data objects in the first push list according to the difference degree of the push effect and the unbiased estimation quantity of the relevance to obtain a second push list;

and pushing target business data to a plurality of data objects according to the second pushing list and a target pushing rule.

In one embodiment, the attribute data of the data object includes at least one of: a scholarly calendar of the data object, a monthly income of the data object, a profession of the data object.

In one embodiment, the historical push results include one or more of the following: the pushing is carried out, and the pushing is determined to be successful; the push is already carried out, and the push failure is determined; and not pushed.

In one embodiment, determining the degree of difference in pushing effectiveness between different data object type groups according to the attribute data of a plurality of data objects and the historical pushing records includes:

determining a data object type group to which each data object in the plurality of data objects belongs according to the attribute data of the data object;

respectively calculating fairness value parameters of each data object type group according to historical pushing records of data objects in each data object type group; the fairness value parameter is used for representing the biased degree of a preset pushing result prediction model to a data object type group;

respectively calculating the average value of the pushing effect of the data objects in each data object type group aiming at the historical pushing rule according to the historical pushing records of the data objects in each data object type group;

and determining the difference degree of the pushing effect between different data object type groups according to the fairness value parameter of each data object type group and the average value of the pushing effect of the data objects in each data object type group aiming at the historical pushing rule.

In one embodiment, determining a data object type group to which each data object of the plurality of data objects belongs according to attribute data of the data object includes:

and according to the attribute data of the data objects, clustering the data objects to determine a data object type group to which each data object in the data objects belongs.

In one embodiment, determining the difference degree of the push effectiveness between different data object type groups according to the fairness value parameter of each data object type group and the average value of the push effectiveness of the data objects in each data object type group for the historical push rule includes:

the degree of difference in push effectiveness between the data object type group numbered i and the data object type group numbered j is calculated according to the following equation:

wherein the content of the first and second substances,

info, the degree of difference in push success between the data object type group numbered i and the data object type group numbered j_τ(G_i) Is the average value of the push results of the data objects in the data object type group with the number i against the history push rule, Info_τ(G_j) Is the average Value, Value (G), of the push results of the data objects in the data object type group numbered j against the history push rule_i) A fair Value parameter, Value (G), for the set of data object types numbered i_j) A fairness value parameter for the set of data object types numbered j.

In one embodiment, determining an unbiased estimated quantity of association of a push rule with a data object based on attribute data of a plurality of data objects and a historical push record includes:

constructing a corresponding unbiased estimator solving loss function according to the attribute data of the data object and a historical pushing rule;

determining an explicit correlation parameter of the data object relative to a historical pushing rule according to a historical pushing result;

calculating the global unbiased estimation quantity of the data object sequencing according to the pushing success rate and the explicit correlation parameter of the data object;

and solving a loss function according to the global unbiased estimate of the data object and the unbiased estimate, and calculating the unbiased estimate of the relevance between the corresponding push rule and the data object.

In one embodiment, adjusting the ordering of the identifiers of the data objects in the first push list according to the difference degree of the push results and the unbiased estimation amount of the relevance to obtain a second push list includes:

determining the sorting position of each data object in the second push list according to the following formula:

wherein σ_τIs the sorting position of the data object in the second push list, D is the data object set formed by a plurality of data objects, tau is time, V is the history push rule set formed by a plurality of history push rules, lambda is the adjusting parameter,

is an unbiased estimate of the relevance of the push rule to the data object.

In one embodiment, after obtaining the second push list, the method further comprises:

determining the diversity distribution probability of the data objects according to the attribute data of the data objects and the historical push records;

adjusting the sequence of the identity identifiers of the data objects in the second push list according to the diversity distribution probability of the data objects to obtain a third push list;

accordingly, the method can be used for solving the problems that,

and pushing target business data to a plurality of data objects according to the third pushing list and a target pushing rule.

In one embodiment, adjusting the ordering of the identifiers of the data objects in the second push list according to the diversity distribution probability of the data objects to obtain a third push list includes:

training a processing model based on an LSTM model according to the attribute data of the data object and the diversity distribution probability of the data object, and extracting an abstract feature characterization vector through a hidden layer in the processing model;

calculating an average diversity index according to the diversity distribution probability of the data object and the abstract feature characterization vector;

and adjusting the sequence of the identity identifiers of the data objects in the second push list according to the average diversity index to obtain a third push list.

In one embodiment, the data object comprises a client object, the target pushing rule comprises a currently adopted marketing scheme, and the target business data comprises a business product or business service to be promoted currently.

This specification also provides a service data pushing device, including:

the acquisition module is used for acquiring attribute data of a plurality of data objects and historical push records; the historical pushing record records historical service data of a data object pushed based on a historical pushing rule and a corresponding historical pushing result;

the first determining module is used for calling a preset pushing result prediction model and determining the pushing success rate of pushing the target service data to each data object in the plurality of data objects according to the attribute data of the plurality of data objects;

the first sequencing module is used for sequencing the identity identifications of the plurality of data objects according to the pushing success rate to obtain a first pushing list;

the second determining module is used for determining the difference degree of the pushing effect among different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object according to the attribute data of the data objects and the historical pushing record;

a second sorting module, configured to adjust, according to the difference degree of the push results and the unbiased estimation amount of the relevance, a sorting of the identity identifiers of the data objects in the first push list, so as to obtain a second push list;

and the pushing module is used for pushing the target service data to the plurality of data objects according to the second pushing list and the target pushing rule.

In one embodiment, the apparatus further comprises a third determining module and a third ordering module, wherein,

the third determining module is used for determining the diversity distribution probability of the data objects according to the attribute data of the data objects and the historical pushing records;

the third sorting module is configured to adjust sorting of the identity identifiers of the data objects in the second push list according to the diversity distribution probability of the data objects, so as to obtain a third push list;

accordingly, the method can be used for solving the problems that,

and the pushing module is used for pushing the target service data to the plurality of data objects according to the third pushing list and the target pushing rule.

The present specification provides a server comprising a processor and a memory for storing processor-executable instructions, the processor implementing obtaining attribute data for a plurality of data objects when executing the instructions, and a historical push record; the historical pushing record records historical service data of a data object pushed based on a historical pushing rule and a corresponding historical pushing result; calling a preset pushing result prediction model, and determining the pushing success rate of pushing target service data to each data object in the plurality of data objects according to the attribute data of the plurality of data objects; according to the push success rate, sequencing the identity identifications of the data objects to obtain a first push list; determining the difference degree of the pushing effect between different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object according to the attribute data of the data objects and the historical pushing record; adjusting the sequence of the identity identifiers of the data objects in the first push list according to the difference degree of the push effect and the unbiased estimation quantity of the relevance to obtain a second push list; and pushing target business data to a plurality of data objects according to the second pushing list and a target pushing rule.

The present specification also provides a computer readable storage medium having stored thereon computer instructions that, when executed, enable obtaining attribute data for a plurality of data objects, and a historical push record; the historical pushing record records historical service data of a data object pushed based on a historical pushing rule and a corresponding historical pushing result; calling a preset pushing result prediction model, and determining the pushing success rate of pushing target service data to each data object in the plurality of data objects according to the attribute data of the plurality of data objects; according to the push success rate, sequencing the identity identifications of the data objects to obtain a first push list; determining the difference degree of the pushing effect between different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object according to the attribute data of the data objects and the historical pushing record; adjusting the sequence of the identity identifiers of the data objects in the first push list according to the difference degree of the push effect and the unbiased estimation quantity of the relevance to obtain a second push list; and pushing target business data to a plurality of data objects according to the second pushing list and a target pushing rule.

According to the pushing method, the pushing device and the pushing server for the business data, before target business data are pushed to a plurality of data objects, attribute data of the data objects and historical pushing records can be obtained; a preset pushing result prediction model is called to predict the pushing success rate of the data objects according to the attributes of the data objects, and the identity identifications of the data objects are sequenced based on the pushing success rate to obtain a corresponding first pushing list; furthermore, the difference degree of the pushing effect between different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object can be determined according to the attribute data of the data object and the historical pushing record, and the data is utilized to correspondingly adjust the sequence in the first pushing list, so that the error of a preset pushing result prediction model caused by the fact that the fairness in the prediction process between the different data object type groups and the relevance between the data object and the pushing rule are not considered is eliminated, and a second pushing list which is relatively accurate and reasonable in sequence is obtained; and then, according to the second push list, target service data can be pushed to a plurality of data objects more accurately and more pertinently, so that a better pushing effect can be obtained, the pushing effect is improved, and the technical problems that the ordering of the data objects in the generated push list is inaccurate and unreasonable and the pushing effect is poor when the service data is pushed to the data objects based on the push list in the existing method are solved.

Drawings

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

Fig. 1 is a schematic diagram of an embodiment of a structural composition of a system to which a pushing method of service data provided by an embodiment of the present specification is applied;

fig. 2 is a flowchart illustrating a pushing method of service data according to an embodiment of the present disclosure;

FIG. 3 is a schematic diagram of a server according to an embodiment of the present disclosure;

fig. 4 is a schematic structural composition diagram of a service data pushing device provided in an embodiment of the present specification;

fig. 5 is a schematic diagram illustrating an embodiment of a pushing method for service data provided by an embodiment of the present specification, in a scenario example;

fig. 6 is a schematic diagram of an embodiment of a pushing method for service data provided by an embodiment of the present specification, in a scenario example;

fig. 7 is a schematic diagram illustrating an embodiment of a pushing method for service data provided by an embodiment of the present specification, in a scenario example;

fig. 8 is a schematic diagram of an embodiment of a pushing method for service data provided by an embodiment of the present specification, in a scenario example;

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

fig. 10 is a schematic diagram of an embodiment of a pushing method for service data provided by an embodiment of the present specification, in an example scenario;

fig. 11 is a schematic diagram of an embodiment of a pushing method for service data provided by an embodiment of the present specification, in an example scenario.

Detailed Description

In order to make those skilled in the art better understand the technical solutions in the present specification, 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 a part of the embodiments of the present specification, and not all of the embodiments. All other embodiments obtained by a person skilled in the art based on the embodiments in the present specification without any inventive step should fall within the scope of protection of the present specification.

Considering that when the existing method pushes target service data to data objects (such as clients and the like), the pushing success rate of each data object is usually predicted by using a trained prediction model; sorting the data objects according to the pushing success rate of each data object to obtain a corresponding pushing list; and finally, pushing the target service data to the data object according to the pushing list.

However, by the inventive analysis it was found that: the prediction model used is limited by the training mechanism and the limitation of the considered factors, and the deviation exists in the process of learning and training.

Specifically, firstly, the prediction model is usually trained by using positive sample data (e.g., client data that has been successfully pushed) and negative sample data (e.g., client data that has been failed to be pushed), so that data objects belonging to the same type group and similar to the positive sample data are continuously predicted with higher probability; while the prediction probability obtained for data objects similar to negative sample data will continue to be low. Therefore, the trained prediction model does not consider the fairness of the prediction result, and tends to favor over data objects similar to the positive sample data when generating the prediction result, so that the prediction result has bias and errors, and a phenomenon such as "the richer people are, the poorer people are" occurs.

Secondly, the prediction model does not consider pushing based on a specific pushing rule when pushing the target business data in the training and learning process. For example, when marketing a business product to a customer, the customer is marketed based on a particular marketing plan. Therefore, the prediction model often ignores the relevance, the engagement degree and other factors of the data object and the push rule in the interactive emergency, and the generated prediction result has a deviation.

For the above two reasons, it is not reasonable to order the data objects directly based on the prediction result of the prediction model, and there is an error.

As a result of creative efforts, the present specification considers that after a push success rate of a data object is predicted by using a preset push result prediction model and a first push list is generated based on the push success rate, a difference degree of push effects between different data object type groups and an unbiased estimation amount of association between a push rule and the data object can be further determined, and the first push list is adjusted by using the data, so as to eliminate an error caused by the fact that fairness between different data object type groups and the association between the data object and the push rule are not considered in the preset push result prediction model, and obtain a second push list with relatively more accurate and reasonable ordering; and then target business data can be pushed to the plurality of data objects more accurately according to the second pushing list, so that a better pushing effect can be obtained, and the pushing effect is improved.

The embodiment of the specification provides a method for pushing service data, which can be particularly applied to a system comprising a first server, a second server and a terminal device.

Specifically, as shown in fig. 1, the first server, the second server and the terminal device may be connected in a wired or wireless manner to perform data interaction. The terminal equipment is responsible for collecting relevant data of the data object. The first server is responsible for ordering the data objects and generating a corresponding push list. The second server is responsible for pushing the target business data to the data object.

In specific implementation, the terminal device may collect attribute data of a plurality of data objects and a history push record. The historical pushing record records historical service data of a data object pushed based on a historical pushing rule and corresponding historical pushing results. And sending the attribute data of the plurality of data objects and the history push record to the first server.

Correspondingly, the first server receives the attribute data of the plurality of data objects sent by the terminal equipment, and the history push records are sent to the first server.

The first server firstly calls a preset pushing result prediction model, and determines the pushing success rate of pushing the target service data to each data object in the data objects according to the attribute data of the data objects; and according to the push success rate, sequencing the identity identifications of the plurality of data objects to obtain a first push list.

Further, the first server can determine the difference degree of the pushing effect among different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object according to the attribute data of the data objects and the historical pushing record; and adjusting the sequence of the identity identifiers of the data objects in the first push list according to the difference degree of the push effect and the unbiased estimation quantity of the relevance to obtain a second push list. And sending the second push list to a second server.

Correspondingly, the second server receives and pushes the target service data to the plurality of data objects according to the second push list and the corresponding target push rule.

In this embodiment, the first server and the second server may specifically include a background server that is applied to a service data processing platform side and is 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 data operation, storage functions and network interaction functions. Alternatively, the first server and the second server may also be software programs running in the electronic device and providing support for data processing, storage and network interaction. In this embodiment, the number of the servers included in the first server and the second server is not specifically limited. The first server and the second server may be specifically one server, or several servers, or a server cluster formed by several servers.

In this embodiment, the terminal device may specifically include a front-end device that is applied to a user side and can implement functions such as data acquisition and data transmission. Specifically, the terminal device may be, for example, a desktop computer, a tablet computer, a data interface of a system, and the like.

Referring to fig. 2, an embodiment of the present disclosure provides a method for pushing service data. The method is particularly applied to the server side. In particular implementations, the method may include the following.

S21: acquiring attribute data of a plurality of data objects and historical push records; the historical pushing record records historical service data of a data object pushed based on a historical pushing rule and corresponding historical pushing results.

S22: and calling a preset pushing result prediction model, and determining the pushing success rate of pushing the target service data to each data object in the plurality of data objects according to the attribute data of the plurality of data objects.

S23: and sequencing the identity identifications of the plurality of data objects according to the pushing success rate to obtain a first pushing list.

S24: and determining the difference degree of the pushing effect between different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object according to the attribute data of the plurality of data objects and the historical pushing record.

S25: and adjusting the sequence of the identity identifiers of the data objects in the first push list according to the difference degree of the push effect and the unbiased estimation quantity of the relevance to obtain a second push list.

S26: and pushing target business data to a plurality of data objects according to the second pushing list and a target pushing rule.

Through the embodiment, the difference degree of the pushing effect among different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object can be determined and utilized to eliminate the error of the preset pushing result prediction model caused by not considering the fairness among different data object type groups and the relevance between the data object and the pushing rule, so that a second pushing list which is relatively more accurate and reasonable in sequencing is obtained, further, the target service data can be more accurately pushed to a plurality of data objects according to the second pushing list, a better pushing effect is obtained, and the pushing effect is improved.

In this embodiment, the data object may specifically include a user object. In specific implementation, the user objects may be different types of user objects according to specific application scenarios.

Specifically, for example, in a financial product promotion scenario of a bank, the user object may be a customer object of the bank. For example, in a product push scenario of an e-commerce, the user object may be a shopper at a shopping site. Of course, the above listed data objects are only illustrative. In specific implementation, the data object may also include other types of user objects in combination with specific application scenarios. The present specification is not limited to these.

In this embodiment, the attribute data of the data object may be specifically understood as parameter data capable of reflecting some attribute features of the data object.

In one embodiment, the attribute data of the data object may specifically include at least one of: a scholarly calendar of the data object, a monthly income of the data object, a profession of the data object, etc. Of course, the above listed attribute data is only an illustrative illustration. Other types of attribute data may also be introduced in conjunction with specific application scenarios and processing requirements.

Through the embodiment, the server can acquire and utilize more types and diversified attribute data so as to more accurately and comprehensively sequence a plurality of data objects.

In this embodiment, the history push record may specifically record information such as history service data that the data object is pushed based on the history push rule, and a corresponding history push result.

In one embodiment, the historical push results include one or more of the following: the pushing is carried out, and the pushing is determined to be successful; the push is already carried out, and the push failure is determined; not pushed, etc. Of course, the above listed history push results are only illustrative. During specific implementation, further refinement can be performed according to specific application scenarios and specific processing conditions, so as to obtain a more refined pushing result.

In this embodiment, by using the above manner, it is avoided that, as in the conventional manner, only two historical pushing results, namely a successful pushing corresponding to positive sample data and a failed pushing corresponding to negative sample data, are simply considered, so that detailed information in the process of pushing historical service data to a data object according to a historical pushing rule is omitted, and richer and more detailed historical pushing results can be recorded and obtained, so that the relevance between the data object and the pushing rule and the pushed service data can be further analyzed according to the historical pushing results, and the deficiency of a preset pushing result prediction model can be made up.

In this embodiment, it should be noted that, when pushing the service data, in general, in order to improve the success rate of pushing so that the data object is more willing to accept and participate in the service activity related to the service data, the service data is often pushed to the data object based on some specific pushing rules.

Specifically, for example, when a financial product or a financial service (which may be understood as a kind of business data) pushed by a bank is pushed to a bank client, a specific pushing marketing may be performed to the bank client based on a specific marketing scheme or marketing strategy (which may be understood as a kind of pushing rule) so as to prompt the bank client to be willing to purchase the financial product or the financial service, thereby increasing the billing rate.

In this embodiment, the history pushing record may further record information related to history pushing experience of other data objects, such as history pushing time of the data object, an initiator of the pushed history service data, and a history pushing rule, so that the acceptance degree and the pushing effect of the data object in being pushed with the target pushing rule based on the target pushing rule may be more comprehensively and accurately analyzed by using the history pushing record in the following.

In an embodiment, the preset pushing result prediction model may be specifically understood as a model obtained by training with positive sample data and negative sample data, and capable of predicting a pushing success rate of pushing the service data to the data object according to the attribute data of the data object.

In an embodiment, in specific implementation, the attribute data of each data object may be processed by using a preset push result prediction model, so as to predict the push success rate of each data object. And then, according to the push success rate of each data object, sorting the identification (such as the name, account name, object number and the like) of the data object according to the push success rate in descending order to obtain a corresponding first push list.

In an embodiment, the preset push result prediction model may be obtained by training in the following manner: acquiring a plurality of sample data; wherein, each sample data at least comprises the attribute data of the corresponding sample data object and the pushing result; according to the pushing result of the sample data, marking the sample data with the pushing result of successful pushing as positive sample data, and marking the sample data with the pushing result of failure as negative sample data to obtain marked sample data; and performing model training by using the labeled sample data to obtain a preset pushing result prediction model.

In this embodiment, it should be noted that, because the preset push result prediction model obtained through the training in the above manner does not take fairness of prediction results into consideration, when performing prediction using the model, it is more likely to favor over data objects belonging to the same data object type group and similar to the positive sample data, so as to cause prediction deviation and affect precision of the model.

In an embodiment, after the first push list is obtained, the fairness between different data object type groups and the error caused by the association between the data object and the push rule are also considered, and the ranking of the identity of the data object in the first push list is further determined and adjusted in a targeted manner according to the difference degree of the push effect between the different data object type groups and the unbiased estimation quantity of the association between the push rule and the data object, so as to obtain a second push list with more reasonable and accurate ranking.

In this embodiment, the plurality of data objects included in each data object type group have a common characteristic in at least one attribute data.

In this embodiment, the above-mentioned difference degree of the pushing effect between different data object type groups may be specifically understood as a kind of parameter data capable of quantitatively reflecting the unfairness degree of the preset pushing result prediction model to different data object type groups at the time of prediction.

In this embodiment, the difference degree of the push effect between the different data object type groups can be determined and utilized to compensate for the deviation caused by independently using the preset push result prediction model without considering the fairness of the prediction result, so as to obtain a push list with more reasonable and accurate ordering.

In an embodiment, the determining, according to the attribute data of the plurality of data objects and the historical pushing record, a difference degree of pushing effectiveness between different data object type groups may be implemented as follows.

S1: determining a data object type group to which each data object in the plurality of data objects belongs according to the attribute data of the data object;

s2: respectively calculating fairness value parameters of each data object type group according to historical pushing records of data objects in each data object type group; the fairness value parameter is used for representing the biased degree of a preset pushing result prediction model to a data object type group;

s3: respectively calculating the average value of the pushing effect of the data objects in each data object type group aiming at the historical pushing rule according to the historical pushing records of the data objects in each data object type group;

s4: and determining the difference degree of the pushing effect between different data object type groups according to the fairness value parameter of each data object type group and the average value of the pushing effect of the data objects in each data object type group aiming at the historical pushing rule.

Through the embodiment, the difference degree of the pushing effect between different data object type groups can be accurately determined in a quantitative mode.

In an embodiment, the determining, according to the attribute data of the data object, a data object type group to which each data object of the plurality of data objects belongs may include, in specific implementation, the following: and according to the attribute data of the data objects, clustering the data objects to determine a data object type group to which each data object in the data objects belongs.

By the above-described embodiments, a plurality of data objects can be efficiently divided into corresponding data object type groups.

Of course, the above-listed manner of determining the data object type group to which the data object belongs is only an illustrative example. In specific implementation, the division rules of different data object type groups can be set according to specific application scenes; and then, according to the division rule and the attribute data of the data objects, determining the data object type group to which each data object belongs.

In an embodiment, the determining, according to the fairness value parameter of each data object type group and an average value of the pushing effectiveness of the data objects in each data object type group for the historical pushing rule, a difference degree of the pushing effectiveness between different data object type groups may include the following steps:

wherein the content of the first and second substances,

specifically, it can be expressed as the difference degree of the push effect between the data object type group with the number i and the data object type group with the number j, Info_τ(G_i) It can be specifically expressed as the average value, Info, of the push effect of the data object in the data object type group with the number i against the history push rule_τ(G_j) It can be specifically expressed as an average Value, Value (G), of the push effect of the data object in the data object type group with the number j against the history push rule_i) The fairness Value parameter, Value (G), which may be specifically expressed as a set of data object types numbered i_j) And may specifically be expressed as a fairness value parameter for the set of data object types numbered j.

In this embodiment, the unbiased estimation amount of the association between the push rule and the data object may be specifically understood as a parameter data capable of quantitatively reflecting the influence of the fitness between the data object and the push rule, which is not considered in the prediction of the preset push result prediction model, on the push effect.

In this embodiment, the bias introduced by using the preset push result prediction model without considering the relevance between the data object and the push rule alone can be compensated by determining and utilizing the unbiased estimation amount of the relevance between the push rule and the data object, so as to obtain a push list with further more reasonable and accurate ranking.

In an embodiment, the determining an unbiased estimation amount of the association between the push rule and the data object according to the attribute data of the plurality of data objects and the historical push record may be implemented as follows.

S1: constructing a corresponding unbiased estimator solving loss function according to the attribute data of the data object and a historical pushing rule;

s2: determining an explicit correlation parameter of the data object relative to a historical pushing rule according to a historical pushing result;

s3: calculating the global unbiased estimation quantity of the data object sequencing according to the pushing success rate and the explicit correlation parameter of the data object;

s4: and solving a loss function according to the global unbiased estimate of the data object and the unbiased estimate, and calculating the unbiased estimate of the relevance between the corresponding push rule and the data object.

By the embodiment, the unbiased estimation quantity of the relevance between the pushing rule and the data object can be accurately determined in a quantitative mode.

In one embodiment, a targeted adjustment may be performed based on the ranking position of the identifiers of the data objects in the first push list according to the degree of difference in the push results and the unbiased estimation amount of the relevance; and then based on the adjusted sorting of the identity marks of the data objects, a second push list with more reasonable and accurate corresponding sorting is obtained.

In an embodiment, the aforementioned adjusting, according to the difference degree of the push results and the unbiased estimation amount of the relevance, the ordering of the identifiers of the data objects in the first push list to obtain a second push list may include the following contents in specific implementation:

determining the sorting position of any one data object d in each data object in the second push list according to the following formula:

wherein σ_τSpecifically, the rank position of the data object in the second push list may be represented, D may be represented as the data object specifically, D may be represented as a data object set formed by a plurality of data objects specifically, τ may be represented as time specifically, V may be represented as a history push rule set formed by a plurality of history push rules specifically, λ may be represented as an adjustment parameter specifically,

may be embodied as a push gaugeAn unbiased estimate of the relevance of the data object.

In the present embodiment, t represents a specific time point,

attribute data representing the data object d at the point in time t,

the rule feature of the historical pushing rule v used for pushing the historical business data to the data object d at the time point t is represented, and the number of combinations between different data object type groups is m.

By the embodiment, the ranking position of the identity of the data object in the first push list can be adjusted in a targeted manner by using the difference degree of the push effect and the unbiased estimation quantity of the relevance, so that the corresponding second push list is obtained. Therefore, the obtained sequence of the identity identifiers of the data objects in the second push list is more reasonable and accurate compared with the sequence of the identity identifiers of the data objects in the first push list, and the sequence has higher reference value.

In this embodiment, the data object with the top-ranked id in the second push list has a relatively higher push success rate when being pushed with the target service data. For example, customers with top-ranked ids in the second push list are relatively more likely to be marketed successfully to purchase promoted financial products.

In one embodiment, the plurality of data objects may be divided into a plurality of different push hierarchies according to the second push list; and distinguishing data objects of different pushing levels, and pushing target business data according to target pushing rules.

Specifically, for example, a batch of data objects with top identities (e.g., located at the top 30%) in the second push list may be divided into a first push hierarchy, and the remaining data objects in the second push list may be divided into a second push hierarchy.

Because the success rate of pushing the data object in the first pushing level is relatively high, the data object in the first pushing level can be pushed once by a combination mode based on two pushing modes, namely a pushing mode 1 (for example, short message pushing) and a pushing mode 2 (for example, mail pushing), according to the target pushing rule.

Because the push success rate of the data object of the second push level is relatively low, the data object in the second push level is continuously pushed for three times by a combined mode based on three push modes, namely a push mode 1, a push mode 2 and a push mode 3 (for example, telephone push), so that a better push effect can be obtained.

In this embodiment, before pushing target service data to a plurality of data objects, attribute data of the data objects and a history pushing record may be obtained; then, a preset pushing result prediction model is called to predict the pushing success rate of the data objects according to the attributes of the data objects, and the identity identifications of the data objects are sequenced based on the pushing success rate to obtain a first pushing list; furthermore, the difference degree of the pushing effect between different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object can be determined according to the attribute data of the data object and the historical pushing record, and the data is utilized to correspondingly adjust the first pushing list so as to eliminate the error of a preset pushing result prediction model caused by not considering the fairness between the different data object type groups and the relevance between the data object and the pushing rule, and obtain a second pushing list which is relatively more accurate and reasonable in order; and then, target service data can be pushed to the plurality of data objects more accurately according to the second push list, so that a better pushing effect can be obtained, the pushing effect is improved, and the technical problem that the pushing effect is poor when the service data is pushed to the data objects based on the push list because the sequencing of the data objects in the generated push list is inaccurate and unreasonable in the existing method is solved.

In an embodiment, further, when a preset push result processing model is used for learning and training, the influence of the diversity among data object individuals on the prediction result is not considered, so that the model is also deviated in prediction. In order to obtain a push list with relatively more accurate and reasonable sequencing on the basis of the second push list, the sequencing in the second push list can be adjusted again by further determining and utilizing the diversity distribution probability of the data objects so as to obtain a third push list with relatively more accurate and reasonable sequencing.

In an embodiment, after obtaining the second push list, when the method is implemented, the following may be further included: determining the diversity distribution probability of the data objects according to the attribute data of the data objects and the historical push records; adjusting the sequence of the identity identifiers of the data objects in the second push list according to the diversity distribution probability of the data objects to obtain a third push list; correspondingly, target business data are pushed to the plurality of data objects according to the third push list and the target push rule.

Through the embodiment, the third push list which is relatively more accurate and reasonable in sequencing can be obtained, and the third push list can be used for replacing the second push list to push the target service data of the data object. Therefore, the pushing effect of the target service data can be further improved.

In an embodiment, the foregoing adjusts the ordering of the identifiers of the data objects in the second push list according to the diversity distribution probability of the data objects to obtain a third push list, and the specific implementation may include the following contents.

S1: training a processing model based on an LSTM model according to the attribute data of the data object and the diversity distribution probability of the data object, and extracting an abstract feature characterization vector through a hidden layer in the processing model;

s2: calculating an average diversity index according to the diversity distribution probability of the data object and the abstract feature characterization vector;

s3: and adjusting the sequence of the identity identifiers of the data objects in the second push list according to the average diversity index to obtain a third push list.

In this embodiment, in a specific implementation, an abstract feature characterization vector with a relatively good effect may be extracted through the last hidden layer in the processing model, so as to be used later.

In this embodiment, the average diversity index may be specifically understood as a parameter data capable of quantitatively reflecting that the diversity factor of the data object that is not considered by the preset push result prediction model during prediction affects the push result.

Through the embodiment, the average diversity index can be accurately determined in a quantitative mode, and then the diversity index can be utilized to adjust the diversity factor based on the data object through the sorting in the second push list, so that the third push list which is relatively more accurate and reasonable in sorting and better in pushing effect is obtained.

In an embodiment, the method can be applied to a plurality of different application scenarios to effectively push related service data in the plurality of different application scenarios, so as to obtain a better pushing effect.

In one embodiment, the method can be applied to marketing scenes of products. In a product marketing scenario, the data objects may specifically include customer objects, such as financial customers of a certain bank. The target business data comprises business products or business services to be promoted currently, such as financial services newly released by a certain bank. The target pushing rule may specifically include a currently adopted marketing scheme, for example, a marketing strategy currently designed by a certain bank for the financial service, and the like.

Furthermore, the method provided by the specification can be used for determining the third push list of the client objects in the marketing scene, distinguishing the client objects according to the third push list, and promoting the business products by using the corresponding marketing scheme, so that a better marketing effect can be obtained, and the unit yield can be improved.

Of course, the marketing scenario listed above is merely illustrative. In particular, the method can be applied to other types of application scenarios. Correspondingly, the data object, the push rule and the target service data may also include related data in other types of application scenarios. The present specification is not limited to these.

In an embodiment, in a specific implementation, the third push list may be compared with the ranking in the first push list to obtain a comparison result; and according to the comparison result, the preset push result prediction model is adjusted in a targeted manner, and the preset push result prediction model is updated. Therefore, the error of the preset pushing result prediction model can be reduced, the model precision is improved, and the updated preset pushing result prediction model can be subsequently utilized to obtain a relatively more accurate and reasonable pushing success rate.

As can be seen from the above, the method for pushing service data provided in this specification may obtain attribute data of a data object and a history pushing record before pushing target service data to a plurality of data objects; then, a preset pushing result prediction model is called to predict the pushing success rate of the data objects according to the attributes of the data objects, and the identity identifications of the data objects are sequenced based on the pushing success rate to obtain a first pushing list; furthermore, the difference degree of the pushing effect between different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object can be determined according to the attribute data of the data object and the historical pushing record, and the data is utilized to correspondingly adjust the first pushing list so as to eliminate the error of a preset pushing result prediction model caused by not considering the fairness between the different data object type groups and the relevance between the data object and the pushing rule, and obtain a second pushing list which is relatively more accurate and reasonable in order; and then, target service data can be pushed to the plurality of data objects more accurately according to the second push list, so that a better pushing effect can be obtained, the pushing effect is improved, and the technical problem that the pushing effect is poor when the service data is pushed to the data objects based on the push list because the sequencing of the data objects in the generated push list is inaccurate and unreasonable in the existing method is solved. After the second push list is determined, consideration of diversity factors of the data objects is introduced, and an average diversity index is determined according to the diversity distribution probability of the data objects; the average diversity index is utilized again to adjust the sorting of the data objects in the second push list again, so that a third push list which is relatively more accurate and reasonable in sorting and better in pushing effect is obtained; and then, the third push list can be utilized to push the target service data to the data object, so that the push effect can be further improved.

Embodiments of the present specification further provide a server, including a processor and a memory for storing processor-executable instructions, where the processor, when implemented, may perform the following steps according to the instructions: acquiring attribute data of a plurality of data objects and historical push records; the historical pushing record records historical service data of a data object pushed based on a historical pushing rule and a corresponding historical pushing result; calling a preset pushing result prediction model, and determining the pushing success rate of pushing target service data to each data object in the plurality of data objects according to the attribute data of the plurality of data objects; according to the push success rate, sequencing the identity identifications of the data objects to obtain a first push list; determining the difference degree of the pushing effect between different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object according to the attribute data of the data objects and the historical pushing record; adjusting the sequence of the identity identifiers of the data objects in the first push list according to the difference degree of the push effect and the unbiased estimation quantity of the relevance to obtain a second push list; and pushing target business data to a plurality of data objects according to the second pushing list and a target pushing rule.

In order to more accurately complete the above instructions, referring to fig. 3, another specific server is provided in the embodiments of the present specification, wherein the server includes a network communication port 31, a processor 32, and a memory 33, and the above structures are connected by an internal cable, so that the structures can perform specific data interaction.

The network communication port 31 may be specifically configured to obtain attribute data of a plurality of data objects and a history push record; the historical pushing record records historical service data of a data object pushed based on a historical pushing rule and corresponding historical pushing results.

The processor 32 may be specifically configured to invoke a preset pushing result prediction model, and determine, according to the attribute data of the plurality of data objects, a pushing success rate for pushing the target service data to each data object of the plurality of data objects; according to the push success rate, sequencing the identity identifications of the data objects to obtain a first push list; determining the difference degree of the pushing effect between different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object according to the attribute data of the data objects and the historical pushing record; adjusting the sequence of the identity identifiers of the data objects in the first push list according to the difference degree of the push effect and the unbiased estimation quantity of the relevance to obtain a second push list; and pushing target business data to a plurality of data objects according to the second pushing list and a target pushing rule.

The memory 33 may be specifically configured to store a corresponding instruction program.

In this embodiment, the network communication port 31 may be a virtual port bound with different communication protocols, so that different data can be sent or received. For example, the network communication port may be a port responsible for web data communication, a port responsible for FTP data communication, or a port responsible for mail data communication. In addition, the network communication port can also be a communication interface or a communication chip of an entity. For example, it may be a wireless mobile network communication chip, such as GSM, CDMA, etc.; it can also be a Wifi chip; it may also be a bluetooth chip.

In the present embodiment, the processor 32 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 that stores computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, an embedded microcontroller, and so forth. The description is not intended to be limiting.

In this embodiment, the memory 33 may include multiple layers, and in a digital system, the memory may be any memory as long as it can store binary data; in an integrated circuit, a circuit without a physical form and with a storage function is also called a memory, such as a RAM, a FIFO and the like; 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.

An embodiment of the present specification further provides a computer storage medium based on the above service data pushing method, where the computer storage medium stores computer program instructions, and when the computer program instructions are executed, the computer storage medium implements: acquiring attribute data of a plurality of data objects and historical push records; the historical pushing record records historical service data of a data object pushed based on a historical pushing rule and a corresponding historical pushing result; calling a preset pushing result prediction model, and determining the pushing success rate of pushing target service data to each data object in the plurality of data objects according to the attribute data of the plurality of data objects; according to the push success rate, sequencing the identity identifications of the data objects to obtain a first push list; determining the difference degree of the pushing effect between different data object type groups and the unbiased estimation quantity of the relevance between the pushing rule and the data object according to the attribute data of the data objects and the historical pushing record; adjusting the sequence of the identity identifiers of the data objects in the first push list according to the difference degree of the push effect and the unbiased estimation quantity of the relevance to obtain a second push list; and pushing target business data to a plurality of data objects according to the second pushing list and a target pushing rule.

In this embodiment, the storage medium includes, but is not limited to, a Random Access Memory (RAM), a Read-Only Memory (ROM), a Cache (Cache), a Hard Disk Drive (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.

In this embodiment, the functions and effects specifically realized by the program instructions stored in the computer storage medium can be explained by comparing with other embodiments, and are not described herein again.

Referring to fig. 4, in a software level, an embodiment of the present specification further provides a service data pushing device, which may specifically include the following structural modules.

The obtaining module 41 may be specifically configured to obtain attribute data of a plurality of data objects and a history push record; the historical pushing record records historical service data of a data object pushed based on a historical pushing rule and a corresponding historical pushing result;

the first determining module 42 may be specifically configured to invoke a preset pushing result prediction model, and determine, according to the attribute data of the plurality of data objects, a pushing success rate for pushing the target service data to each data object of the plurality of data objects;

the first sorting module 43 may be specifically configured to sort the identifiers of the multiple data objects according to the push success rate, so as to obtain a first push list;

the second determining module 44 is specifically configured to determine, according to the attribute data of the multiple data objects and the historical pushing records, a difference degree of pushing effectiveness between different data object type groups and an unbiased estimation amount of relevance between a pushing rule and a data object;

the second sorting module 45 may be specifically configured to adjust, according to the difference degree of the push results and the unbiased estimation amount of the relevance, the sorting of the identity identifiers of the data objects in the first push list to obtain a second push list;

the pushing module 46 may be specifically configured to push the target service data to the multiple data objects according to the second push list and the target push rule.

In one embodiment, the apparatus may further include a third determining module and a third sorting module, wherein,

the third determining module may be specifically configured to determine a diversity distribution probability of the data objects according to the attribute data of the plurality of data objects and the history push record;

the third sorting module may be specifically configured to adjust, according to the diversity distribution probability of the data objects, the sorting of the identity identifiers of the data objects in the second push list, so as to obtain a third push list;

accordingly, the method can be used for solving the problems that,

the pushing module may be specifically configured to push target service data to the multiple data objects according to the third push list and the target push rule.

It should be noted that, the units, devices, modules, etc. illustrated in the above embodiments may be implemented by a computer chip or an entity, or implemented by a product with certain functions. For convenience of description, the above devices are described as being divided into various modules by functions, and are described separately. It is to be understood that, in implementing the present specification, functions of each module may be implemented in one or more pieces of software and/or hardware, or a module that implements the same function may be implemented by a combination of a plurality of sub-modules or sub-units, or the like. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only one logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.

As can be seen from the above, in the service data pushing apparatus provided in this specification, after the push success rate of a data object is predicted by using a preset push result prediction model and a first push list is generated based on the push success rate, by determining the difference degree of push effects between different data object type groups and the unbiased estimation amount of the relevance between a push rule and the data object and adjusting the first push list by using the data, an error caused by the fact that fairness between different data object type groups and the relevance between the data object and the push rule are not considered in the preset push result prediction model is eliminated, and a second push list with relatively more accurate ordering is obtained; and then, target service data can be pushed to the plurality of data objects according to the second push list, so that a better pushing effect can be obtained.

In a specific scenario example, the pushing method of the business data provided by the present specification may be applied to generate a corresponding customer marketing list, and the customer is marketed based on the list, so as to obtain a better marketing effect. Specific implementations can be found in the following.

For the marketing scenario, in the specific implementation, a corresponding system can be constructed to take charge of specific business marketing based on the business data pushing method provided by the specification.

Based on the functional processing level, the system may specifically include: the system comprises a client marketing probability prediction module, a fair unbiased dynamic adjustment module, a diversity reordering module and a client marketing marking module, wherein the client marketing probability prediction module, the fair unbiased dynamic adjustment module, the diversity reordering module and the client marketing marking module are 4 main functional modules.

Based on the system structure level, as shown in fig. 5, the system may include two parts, namely a machine learning system and a customer marketing system. Further, the machine learning system may further include three devices, respectively: machine learning model device, fair unbiased dynamic adjusting device and reordering device. The client marketing probability prediction module is realized in a machine learning model device, the fair unbiased dynamic adjustment module is realized in the fair unbiased dynamic adjustment device, and the diversity reordering module is realized in the reordering device. The customer marketing system may include a business marketing apparatus. Wherein, the client marketing marking module is realized in the service marketing device.

In particular, reference may be made to FIG. 6. Customer marketing probability prediction module 1 among the main functional modules: the probability (e.g., push success rate) that a customer (e.g., data object) is successfully marketed is predicted using a supervised machine learning model (e.g., a preset push outcome prediction model). The fairness unbiased dynamic adjustment module 2: and dynamically adjusting the fairness metric to realize fair unbiased dynamic adjustment of the prediction ranking (for example, determining and adjusting the ranking in the first push list according to the difference degree of the push effect between different data object type groups and unbiased estimation quantity of the relevance between the push rule and the data object to obtain a second push list). Diversity reordering module 3: and adjusting the diversity of the prediction ordering to realize prediction reordering (for example, adjusting the ordering of the identity of the data object in the second push list by using an average diversity index determined based on the diversity distribution probability of the data object to obtain a third push list). Customer marketing annotation module 4: and marking a marketing result of the client and feeding back the marketing result to the machine learning module.

In particular implementation, referring to fig. 7, the customer marketing probability prediction module 1 may include the following components in particular operation.

Step S101: design feature variables (e.g., attribute data of the data object). And designing model characteristic variables according to modeling purposes. Marketing model characteristic variables can be designed according to common customer characteristics.

Step S102: and marking the client identification. And collecting customer labels which are mainly divided into successful marketing customers and unsuccessful marketing customers. The collection of the modeling data set takes a customer as a dimension and one customer record as one sample data. The number of successful marketing customers and unsuccessful marketing customers records comprise the total sample dataset modeled.

Let the customer base be D, and for each customer D, there is D ∈ D. The client characteristic set is X, and the characteristic variable X of each client^dHas x^d∈X。

Further, time factors are considered, time sequence data are established, and for each time point, the following data are respectively provided: t 1.,. tau.

The set of marketing scenarios (e.g., push rules) is V, and for each scenario V, there is V ∈ V. Marketing plan feature variable set Z. Characteristic variable z for each marketing plan^vHaving z of^v∈Z。

For each marketing scenario v, the following annotation information may be collected:

wherein, y_t(d | v) is the explicit association of the degree of engagement between marketing strategy and customer, the implicit association behind which is rel_t(d|v)∈[0，1]. In an ideal state, y_t(d|v)＝E[rel_t(d|v)]。

Step S103: and generating a model prediction time sequence result. And (3) taking time sequence data, and generating a marketing probability p corresponding to each customer under different marketing schemes for each time t 1_t(d|v)。

Referring to fig. 8, the operation of the fair unbiased dynamic adjustment module 2 may include the following.

Step S201: a fairness discrepancy metric. Classified by customer population (e.g., determining the set of types of data objects to which the data object belongs), G ═ G₁,...,G_m}。

The classification method of the client group can classify the client group according to the characteristic attributes of the client according to expert rules, and can also classify the client group according to machine learning models such as a clustering model and the like. The main reasons are: the probability values obtained by the supervised marketing prediction model with similar characteristics in prediction are relatively close. Therefore, after the client groups are divided, the sequencing of similar client groups can be promoted according to a fair unbiased dynamic adjustment method.

Computing fairness value (e.g., fairness value parameter):

wherein R is_t(d | v) is the global ranking value of customers in different marketing scenarios at different times. The average global ranking value of each customer under different marketing schemes at different times is calculated. And calculating the average fairness value of the same type of client group. The higher the fairness value is, the higher the average ranking on the marketing list of the group is.

And calculating the marketing difference degree, wherein the larger the difference is, the larger the marketing probability and the effect difference among different groups are, and the purpose of the adjusted marketing probability is to reduce the difference distance.

And calculating the information value of the marketing probability. Calculating the average marketing probability of each customer at different moments, and then calculating the average marketing probability of the customer group:

for all marketing programs, there are:

and calculating the information value of the marketing effect. Calculating the average marketing effect of each client at different time, and calculating the average marketing effect of the client group:

for all marketing programs, there are:

the concept of entropy is used for calculating the average value of the marketing information, and comprises the following steps:

calculate the degree of marketing difference between different customer groups (e.g., the degree of difference in push success between different sets of data object types):

to obtain

Ω_τIndicating the degree of difference between the two populations. The larger the value, the greater the degree of difference.

Step S202: unbiased measurement. Calculating relevance rel of fit degree of marketing strategy and client_t(d | v) unbiased estimate and global rank R (d | v) unbiased estimate of customer marketing rank.

Calculating relevance rel of fit degree of marketing strategy and client_t(d | v) unbiased estimate. Setting the corresponding model function form as

Re-solving

The unbiased estimate solving loss function is:

wherein the content of the first and second substances,

is the characteristic variable of each client d at time t,

characteristic variables of each marketing plan v at time t.

The global rank R (d | v) unbiased estimate of the customer marketing rank is:

step S203: and (4) performing fair dynamic adjustment. A global ordering policy is computed. The marketing ranking (e.g., ranking position) of each customer may be specifically calculated according to the following equation:

wherein σ_τRepresenting the marketing ranking for each customer at time instant tau,

indicating the average degree of association under different marketing schemes at different times.

And the super-parameter lambda is used for adjusting the sequencing strategy and adjusting the influence degree of the marketing diversity on the sequencing. The invention sums the marketing difference degree difference degrees. According to different use requirements, omega_τCan be adjusted to the maximum value, the minimum value and the median of the marketing difference degree. Higher λ indicates a larger adjustment to the original ordering.

It is to be added that the number of marketing schemes V, described above, can be dynamically adjusted. If the customer ranking is designed for a certain marketing plan, let V be 1 in the above formula.

Referring to fig. 9, the detailed operation of the diversity reordering module 3 may include the following.

Step 301: and calculating the ideal distribution of diversity. And calculating the client diversity distribution shape of the ideal state.

The set of marketing schemes is V, and for each scheme V, there is V ∈ V. Designing feature variable set Z of marketing scheme diversity under ideal state_ideal. Characteristic variables for each marketing plan

Is provided with

For example, the feature variable set is [ education level distribution, asset distribution ]]. Distribution of education level (high school, this department, above this department)]Assets distribution [10 ten thousand, 10 ten thousand-100 ten thousand, 100 ten thousand-500 ten thousand, 500 ten thousand or more]. The distribution of the diversity of the ideal state is designed for each marketing plan.Such as marketing scheme v₁The designed one-hot feature vector is [1,1 ] according to the customer diversity of different education degree distributions of the target]Regardless of asset distribution. If marketing scheme v₂The target simultaneously considers the customer diversity of multiple education degrees and assets within 100 ten thousand, and the designed one-hot feature vector is [1,1,1,1,1,0 ]]。

Let the set of ideal distributions be F, there being an ideal distribution F for each v^vHas f^vE.g. F. Calculating the Hellinger distance between the client group D and the marketing scheme v characteristics:

wherein the probability distribution of the distance is a diversity ideal distribution

Step 302: context Embedding Vector parameter calculation.

Using the LSTM model:

wherein, { d₁,...,d_sH is a sequence, hidden, of client groups ordered according to ideal distribution probability_n-1,hidden_nIndicating the number of hidden layers, cell_nIs the state of the cell of the LSTM at the nth layer.

Is a parameter of the LSTM.

The target output is the ideal probability distribution of the aforementioned step S301.

Calculating to obtain model parameters

Wherein the Context Embedding Vector can be specifically selected from the hidden_nThe output of the layer is obtained.

Step 303: and (4) measuring the diversity. An average diversity index is calculated.

Wherein the content of the first and second substances,

p_t(d | v) is defined in accordance with the above-described step S202 and step S103. Where p is_t(d | v) represents

Represents the weighted ranking value under different marketing schemes.

Representing the similarity between two clients.

The Hellinger distance can be used to calculate the similarity:

it should be noted that the formula indicates that the consideration of diversity is added when considering the weighted ranking value. Wherein k represents the k-dimension Context Embedding Vector feature Vector.

When calculating the diversity, the LSTM model in step S203 is selected by accessing the sequence σ t to the previous step.

Get the hide of each customer_nAs the Context Embedding Vector. Where γ is a hyper-parameter, higher indicates more diversity.

Step 304: the clients are reordered. The desired degree of fairness, unbiased and diversity is adjusted by adjusting the hyper-parameters λ and γ.

Evaluating the sorted utility value function according to the following equation:

s.t.MLR(D)∈[floor,cap]

wherein σ_τ(d) Indicates that customer d is at σ_τThe rank value of (c). Utility represents the overall ranking Utility value. It is required that the utility is brought to a desired target under a condition of satisfying a certain diversity. expectedUtility, floor, cap are set values.

After the ranking values that satisfy utility are found, the customers that satisfy the desired results are reordered by σ. Resulting in a final customer marketing manifest (e.g., a third push list).

Referring to fig. 10, the specific operation of the customer marketing annotation module 4 may include the following.

Step S401: and (5) marketing by the customer. In a customer marketing system, marketing is conducted according to a customer list.

Step S402: and expanding the marketing annotation data. And recording whether the customer is marketed and purchasing a corresponding product based on the dynamically adjusted customer marketing list. And the marking information which is not captured in the original marketing model is expanded, and the marking information can also be used for updating the customer marketing model.

Comparing the above process with the prior art method, reference can be made to fig. 11. Where the upper part (before corresponding implementation) is based on the system architecture state and data flow of existing methods. The lower part (after corresponding implementation) is based on the system architecture state and data flow of the method.

Through comparison, based on the existing method, the client marketing model predicts the probability of successful marketing of the client according to the characteristic variables of the client by adopting a supervised algorithm model, and acquires a client marketing list according to the probability. The basic principle of the algorithm is to collect sample data of whether a customer is successfully marketed, and train a prediction and adjustment model according to the characteristic variables of the customer and the positive and negative samples. Algorithmically, customers who are similar to the positive sample (customers who have been successfully marketed) get a higher prediction probability; customers who are similar to the negative examples (customers who have not yet successfully marketed) get a low probability of prediction. The algorithm has the following problems: one is that the algorithm does not take into account prediction fairness. The results favor over the positive sample customers, the high-probability customers continue to obtain high probabilities, the low-probability customers continue to obtain low probabilities, and a state of "the richest is, the poorer is" exists. Secondly, the algorithm does not consider the unbiased prediction result. The algorithm does not calculate the correlation between the marketing strategy and the preference of the client, and the prediction result has bias. Thirdly, the algorithm does not take into account the diversity of the client types. The distribution and types of customers in the marketing strategy are not explicitly considered in the algorithm, so that the simplification of the types of the customers is easily caused.

Compared with the existing method, the method increases the fairness unbiased adjustment algorithm strategy on the original marketing prediction model result, improves the fairness and unbiased prediction result, increases the diversified reordering strategy, and realizes the diversity of the result. The algorithm implementation device can self-adaptively and dynamically adjust the sequencing of the customers on the marketing list according to the data change, and the sequenced list is connected with the marketing business system. The original business system is transparent, and the business operation process of the business system is not influenced.

Through the scene example, the method provided by the specification is verified, and a fair, unbiased and diversified dynamic adjustment mechanism is added to a customer marketing prediction algorithm. One is that the algorithm increases the fairness and unbiased performance of the prediction results. The situation of "richness of the richen person and the poor person" is narrowed. Secondly, the evaluation measurement of diversity is increased by the algorithm. Customer diversity is incorporated into the algorithm. Thirdly, the algorithm adjusting mechanism can seamlessly joint the original service flow, and dynamic closed loop of the whole flow is realized. And each functional module adopts the design concept of low coupling and high cohesion. In application, nodes and data flows can be added to the original system architecture, and corresponding data processing can be efficiently realized.

Although the present specification provides method steps as described in the examples or flowcharts, additional or fewer steps may be included based on conventional or non-inventive means. The order of steps recited in the embodiments is merely one manner of performing the steps in a multitude of orders and does not represent the only order of execution. When an apparatus or client product in practice executes, it may execute sequentially or in parallel (e.g., in a parallel processor or multithreaded processing environment, or even in a distributed data processing environment) according to the embodiments or methods shown in the figures. 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, the presence of additional identical or equivalent elements in a process, method, article, or apparatus that comprises the recited elements is not excluded. The terms first, second, etc. are used to denote names, but not any particular order.

Those skilled in the art will also appreciate that, in addition to implementing the controller as pure computer readable program code, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may therefore be considered as a hardware component, and the means included therein for performing the various functions may also be considered as a structure within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.

This 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 the embodiments, it is clear to those skilled in the art that the present specification can be implemented by software plus necessary general hardware platform. With this understanding, the technical solutions in the present specification may be essentially embodied 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 includes several instructions for enabling a computer device (which may be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments in the present specification.

The embodiments in the present specification are described in a progressive manner, and the same or similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. The description is operational with numerous general purpose or special purpose computing system environments or configurations. For example: personal computers, server computers, hand-held or portable devices, tablet-type 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.

While the specification has been described with examples, those skilled in the art will appreciate that there are numerous variations and permutations of the specification that do not depart from the spirit of the specification, and it is intended that the appended claims include such variations and modifications that do not depart from the spirit of the specification.

Claims

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

2. The method of claim 1, wherein the attribute data of the data object comprises at least one of: a scholarly calendar of the data object, a monthly income of the data object, a profession of the data object.

3. The method of claim 1, wherein the historical push results comprise one or more of the following: the pushing is carried out, and the pushing is determined to be successful; the push is already carried out, and the push failure is determined; and not pushed.

4. The method of claim 1, wherein determining a degree of difference in push performance between different data object type groups based on attribute data of a plurality of data objects and a historical push record comprises:

5. The method of claim 4, wherein determining a data object type group to which each of the plurality of data objects belongs based on attribute data of the data objects comprises:

6. The method according to claim 4, wherein determining the degree of difference in the push effectiveness between different data object type groups according to the fairness value parameter of each data object type group and the average value of the push effectiveness of the data objects in each data object type group for the historical push rules comprises:

wherein the content of the first and second substances,

info, the degree of difference in push success between the data object type group numbered i and the data object type group numbered j_τ(G_i) For data object in data object type group numbered iAverage, Info of push results for history push rules_τ(G_j) Is the average Value, Value (G), of the push results of the data objects in the data object type group numbered j against the history push rule_i) A fair Value parameter, Value (G), for the set of data object types numbered i_j) A fairness value parameter for the set of data object types numbered j.

7. The method of claim 2, wherein determining an unbiased estimate of the association of push rules with data objects based on attribute data of a plurality of data objects and historical push records comprises:

8. The method of claim 1, wherein adjusting the ordering of the ids of the data objects in the first push list according to the difference degree of the push success and the unbiased estimation of the correlation to obtain a second push list comprises:

wherein σ_τRanking of data objects in a second push listAn order position, D is a data object set formed by a plurality of data objects, tau is time, V is a historical pushing rule set formed by a plurality of historical pushing rules, lambda is an adjustment parameter,

is an unbiased estimate of the relevance of the push rule to the data object.

9. The method of claim 1, wherein after obtaining the second push list, the method further comprises:

accordingly, the method can be used for solving the problems that,

10. The method of claim 9, wherein adjusting the ordering of the ids of the data objects in the second push list according to the distribution probability of diversity of the data objects to obtain a third push list comprises:

11. The method of claim 1, wherein the data object comprises a customer object, the targeted push rule comprises a currently adopted marketing plan, and the targeted business data comprises a business product or business service currently to be promoted.

12. A device for pushing service data, comprising:

13. The apparatus of claim 12, further comprising a third determination module and a third ordering module, wherein,

accordingly, the method can be used for solving the problems that,

14. 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.

15. A computer-readable storage medium having stored thereon computer instructions which, when executed, implement the steps of the method of any one of claims 1 to 11.