Business data push method, device and server

Abstract

This specification provides a method, device and server for pushing business data. Based on this method, combined with artificial intelligence technology, after predicting the push success rate of a data object by using a preset push result prediction model, and generating a first push list with a corresponding ranking based on the push success rate, the difference in the push effectiveness between different data object type groups, and the unbiased estimator of the correlation between the push rules and the data objects are further determined, and the ranking in the first push list is adjusted by using the above data to eliminate the error caused by the preset push result prediction model not taking into account the fairness between different data object type groups and the correlation between the data objects and the push rules, and obtain a second push list with a relatively more accurate ranking; and then, according to the second push list, the target business data can be pushed to multiple data objects, so that a better push effect can be obtained.

Description

Technical Field

[0001] This specification belongs to the field of artificial intelligence technology, and in particular, relates to a method, device and server for pushing business data. Background Art

[0002] When pushing business data to customers, we often first use a prediction model trained based on artificial intelligence technology (for example, a supervised learning algorithm) to predict the customer's push success rate; then we sort the customers based on the predicted push success rate and generate a push list for the customers; and then we can push the corresponding business data to the customers based on the push list.

[0003] However, due to the limitations of the training mechanism and the factors considered, the above prediction model itself will have deviations in the learning and training process. As a result, the push list generated by sorting based on the push success rate predicted by the above prediction model will also have errors, affecting the subsequent push effect.

[0004] To address the above problems, no effective solution has been proposed yet. Summary of the invention

[0005] This specification provides a method, device and server for pushing business data to generate a second push list with a relatively more accurate and reasonable sorting, and based on the second push list, more accurately push target business data to the data object to achieve a better push effect.

[0006] This specification provides a method for pushing business data, including:

[0007] Acquire attribute data of multiple data objects and historical push records; wherein the historical push records record historical business data of the data objects pushed based on historical push rules and corresponding historical push results;

[0008] Calling a preset push result prediction model to determine a push success rate of pushing target business data to each of the multiple data objects according to the attribute data of the multiple data objects;

[0009] According to the push success rate, the identity identifiers of the multiple data objects are sorted to obtain a first push list;

[0010] Based on the attribute data of multiple data objects and historical push records, determine the difference in push effectiveness between different data object type groups and an unbiased estimator of the correlation between push rules and data objects;

[0011] According to the difference degree of the push effectiveness and the unbiased estimator of the correlation, the order of the identity identifiers of the data objects in the first push list is adjusted to obtain a second push list;

[0012] According to the second push list and target push rules, target business data is pushed to multiple data objects.

[0013] In one embodiment, the attribute data of the data object includes at least one of the following: the education level of the data object, the monthly income of the data object, and the occupation of the data object.

[0014] In one embodiment, the historical push results include one or more of the following results: pushed and determined to be successful; pushed and determined to be a failure; not pushed.

[0015] In one embodiment, based on the attribute data of multiple data objects and the historical push records, determining the difference in push effectiveness between different data object type groups includes:

[0016] Determining, according to the attribute data of the data object, the data object type group to which each data object in the plurality of data objects belongs;

[0017] According to the historical push records of the data objects in each data object type group, respectively calculate the fairness value parameter of each data object type group; wherein the fairness value parameter is used to characterize the degree of partiality of the preset push result prediction model to the data object type group;

[0018] According to the historical push records of the data objects in each data object type group, respectively calculate the average push results of the data objects in each data object type group for the historical push rules;

[0019] The difference in push effectiveness between different data object type groups is determined based on the fairness value parameters of each data object type group and the average value of push effectiveness of the data objects in each data object type group according to the historical push rules.

[0020] In one embodiment, determining the 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 includes:

[0021] According to the attribute data of the multiple data objects, clustering processing is performed on the multiple data objects to determine the data object type group to which each data object in the multiple data objects belongs.

[0022] In one embodiment, according to the fairness value parameters of each data object type group and the average push effectiveness of the data objects in each data object type group according to the historical push rules, determining the difference degree of push effectiveness between different data object type groups includes:

[0023] The difference in push results between the data object type group numbered i and the data object type group numbered j is calculated using the following formula:

[0024]

[0025] in, Info is the difference in push results 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 numbered i for the historical push rules, Info τ (G j ) is the average value of the push results of the data objects in the data object type group numbered j for the historical push rules, Value(G i ) is the fairness value parameter of the data object type group numbered i, Value(G j ) is the fairness value parameter of the data object type group numbered j.

[0026] In one embodiment, based on the attribute data of multiple data objects and historical push records, an unbiased estimator of the association between the push rule and the data object is determined, including:

[0027] According to the attribute data of the data object and the historical push rules, the corresponding unbiased estimator is constructed to solve the loss function;

[0028] According to the historical push results, determine the explicit relevance parameters of the data object relative to the historical push rules;

[0029] According to the push success rate and explicit correlation parameter of the data objects, a global unbiased estimator of the data object ranking is calculated;

[0030] According to the global unbiased estimator of the data object and the loss function solved by the unbiased estimator, an unbiased estimator of the association between the corresponding push rule and the data object is calculated.

[0031] In one embodiment, according to the difference degree of the push effectiveness and the unbiased estimator of the correlation, the order of the identity identifiers of the data objects in the first push list is adjusted to obtain the second push list, including:

[0032] The ranking position of each data object in the second push list is determined according to the following formula:

[0033]

[0034] Among them, σ τ is the ranking position of the data object in the second push list, d is the data object, D is the data object set consisting of multiple data objects, τ is time, v is the historical push rule, V is the historical push rule set consisting of multiple historical push rules, λ is the adjustment parameter, It is an unbiased estimator of the association between push rules and data objects.

[0035] In one embodiment, after obtaining the second push list, the method further includes:

[0036] Determine the diversity distribution probability of data objects based on the attribute data of multiple data objects and historical push records;

[0037] According to the diversity distribution probability of the data objects, the order of the identity identifiers of the data objects in the second push list is adjusted to obtain a third push list;

[0038] Accordingly,

[0039] According to the third push list and target push rules, target business data is pushed to multiple data objects.

[0040] In one embodiment, adjusting the order of the identities 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:

[0041] 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 representation vector through a hidden layer in the processing model;

[0042] Calculating an average diversity index according to the diversity distribution probability of the data object and the abstract feature representation vector;

[0043] According to the average diversity index, the order of the identity identifiers of the data objects in the second push list is adjusted to obtain a third push list.

[0044] In one embodiment, the data object includes a customer object, the target push rule includes a currently adopted marketing plan, and the target business data includes a business product or business service to be promoted.

[0045] This specification also provides a device for pushing business data, including:

[0046] An acquisition module, used to acquire attribute data of multiple data objects and historical push records; wherein the historical push records record historical business data of the data objects pushed based on historical push rules and corresponding historical push results;

[0047] A first determination module is used to call a preset push result prediction model to determine a push success rate of pushing target business data to each of the multiple data objects according to the attribute data of the multiple data objects;

[0048] A first sorting module, configured to sort the identities of the plurality of data objects according to the push success rate to obtain a first push list;

[0049] A second determination module is used to determine the difference in push effectiveness between different data object type groups and an unbiased estimator of the correlation between the push rules and the data objects based on the attribute data of the multiple data objects and the historical push records;

[0050] A second sorting module is used to adjust the sorting of the identity identifiers of the data objects in the first push list according to the difference degree of the push effectiveness and the unbiased estimator of the correlation to obtain a second push list;

[0051] The push module is used to push target business data to multiple data objects according to the second push list and target push rules.

[0052] In one embodiment, the device further includes a third determining module and a third sorting module, wherein:

[0053] The third determination module is used to determine the diversity distribution probability of the data objects according to the attribute data of the multiple data objects and the historical push records;

[0054] The third sorting module is used to adjust the sorting 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;

[0055] Accordingly,

[0056] The push module is used to push target business data to multiple data objects according to the third push list and target push rules.

[0057] The present specification provides a server, comprising a processor and a memory for storing processor executable instructions, wherein when the processor executes the instructions, the processor obtains attribute data of multiple data objects and historical push records; wherein the historical push records record historical business data of the data objects pushed based on historical push rules and corresponding historical push results; a preset push result prediction model is called to determine the push success rate of pushing target business data to each of the multiple data objects according to the attribute data of the multiple data objects; according to the push success rate, the identity identifiers of the multiple data objects are sorted to obtain a first push list; according to the attribute data of the multiple data objects and the historical push records, the difference degree of push effectiveness between different data object type groups and the unbiased estimator of the correlation between the push rules and the data objects are determined; according to the difference degree of push effectiveness and the unbiased estimator of the correlation, the sorting of the identity identifiers of the data objects in the first push list is adjusted to obtain a second push list; according to the second push list and the target push rules, the target business data is pushed to the multiple data objects.

[0058] The present specification also provides a computer-readable storage medium having computer instructions stored thereon, which, when executed, are used to obtain attribute data of multiple data objects and historical push records; wherein the historical push records record historical business data of the data objects pushed based on historical push rules, and corresponding historical push results; calling a preset push result prediction model, and determining a push success rate of pushing target business data to each of the multiple data objects based on the attribute data of the multiple data objects; sorting the identity identifiers of the multiple data objects based on the push success rate to obtain a first push list; determining the degree of difference in push effectiveness between different data object type groups and an unbiased estimator of the correlation between push rules and data objects based on the attribute data of the multiple data objects and the historical push records; adjusting the sorting of the identity identifiers of the data objects in the first push list based on the degree of difference in push effectiveness and the unbiased estimator of the correlation to obtain a second push list; and pushing target business data to multiple data objects based on the second push list and the target push rules.

[0059] The present specification provides a method, device and server for pushing business data. Before pushing target business data to multiple data objects, the attribute data of the data objects and the historical push records can be obtained first; and a preset push result prediction model is called to predict the push success rate of the data objects according to the attributes of the data objects, and the identity identifiers of the data objects are sorted based on the push success rate to obtain the corresponding first push list; further, the difference degree of push effectiveness between different data object type groups and the unbiased estimator of the correlation between the push rules and the data objects can be determined according to the attribute data of the data objects and the historical push records, and the above data can be used to predict the first push success rate of the data objects. The sorting in the push list is adjusted accordingly to eliminate the error caused by the preset push result prediction model not taking into account the fairness of prediction between different data object type groups and the correlation between data objects and push rules, so as to obtain a second push list with a relatively more accurate and reasonable sorting; and then, according to the above-mentioned second push list, the target business data can be pushed to multiple data objects more accurately and specifically, so as to obtain a better push effect, improve the push effectiveness, and solve the technical problems of inaccurate and unreasonable sorting of data objects in the generated push list and poor push effect when pushing business data to data objects based on the push list in the existing method. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of this specification, the drawings required for use in the embodiments will be briefly introduced below. The drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0061] Figure 1 It is a schematic diagram of an embodiment of the structural composition of a system using the method for pushing business data provided in the embodiments of this specification;

[0062] Figure 2 It is a flowchart of a method for pushing business data provided by an embodiment of this specification;

[0063] Figure 3 It is a schematic diagram of the structure of a server provided by an embodiment of this specification;

[0064] Figure 4 This is a schematic diagram of the structure of a device for pushing service data provided by an embodiment of this specification;

[0065] Figure 5 It is a schematic diagram of an embodiment of a method for pushing business data provided by an embodiment of this specification, in a scenario example;

[0066] Figure 6 It is a schematic diagram of an embodiment of a method for pushing business data provided by an embodiment of this specification, in a scenario example;

[0067] Figure 7 It is a schematic diagram of an embodiment of a method for pushing business data provided by an embodiment of this specification, in a scenario example;

[0068] Figure 8 It is a schematic diagram of an embodiment of a method for pushing business data provided by an embodiment of this specification, in a scenario example;

[0069] Fig. 9 It is a schematic diagram of an embodiment of a method for pushing business data provided by an embodiment of this specification, in a scenario example;

[0070] Fig.10 It is a schematic diagram of an embodiment of a method for pushing business data provided by an embodiment of this specification, in a scenario example;

[0071] Fig.11 It is a schematic diagram of an embodiment of a method for pushing business data provided by an embodiment of this specification, in a scenario example. DETAILED DESCRIPTION

[0072] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0073] Considering that the existing methods often use the trained prediction model to predict the push success rate of each data object when pushing target business data to data objects (for example, customers, etc.); then sort the data objects according to the push success rate of each data object to obtain the corresponding push list; finally, push the target business data to the data object according to the above push list.

[0074] However, through creative analysis, it was found that the above-mentioned prediction model used was restricted by the training mechanism and the limitations of the factors considered, and there were deviations in the learning and training process itself.

[0075] Specifically, first, the above prediction model is usually trained using positive sample data (for example, customer data that has been successfully pushed) and negative sample data (for example, customer data that has been unsuccessfully pushed), so that data objects that are similar to the positive sample data and belong to the same type group will continue to receive higher prediction probabilities; while data objects that are similar to the negative sample data will continue to receive lower prediction probabilities. This results in the trained prediction model not taking into account the fairness of the prediction results, and when generating prediction results, it will tend to favor data objects that are similar to the positive sample data, resulting in bias and errors in the prediction results, such as the phenomenon of "the rich get richer and the poor get poorer."

[0076] Secondly, the above prediction models do not consider the specific push rules when pushing the target business data during the training and learning process. For example, when marketing a business product to a customer, the customer is marketed based on a specific marketing plan. Therefore, the above prediction models often ignore the relevance and fit between the data object and the push rules in the interaction case, resulting in deviations in the generated prediction results.

[0077] Based on the above two reasons, it is bound to be unreasonable and erroneous if data objects are sorted directly based on the prediction results of the prediction model.

[0078] It is precisely because of the discovery and attention to the above-mentioned problems that, through creative labor, the present specification considers that after predicting the push success rate of data objects by using a preset push result prediction model and generating a first push list based on the push success rate, the degree of difference in push effectiveness between different data object type groups and an unbiased estimator of the correlation between push rules and data objects can be further determined, and the above-mentioned data can be used to adjust the first push list to eliminate the errors caused by the preset push result prediction model due to the failure to take into account the fairness between different data object type groups and the correlation between data objects and push rules, so as to obtain a second push list with relatively more accurate and reasonable sorting; and then, according to the second push list, the target business data can be pushed to multiple data objects more accurately, thereby obtaining a better push effect and improving the push effectiveness.

[0079] The embodiments of the present specification provide a method for pushing business data, which can be specifically applied to a system including a first server, a second server and a terminal device.

[0080] For details, please refer to Figure 1As shown, the first server, the second server and the terminal device can be connected by wire or wirelessly to perform data interaction. The terminal device is responsible for collecting relevant data of the data object. The first server is responsible for sorting the data objects and generating a corresponding push list. The second server is responsible for pushing target business data to the data object.

[0081] In specific implementation, the terminal device can collect attribute data of multiple data objects and historical push records. The historical push records record the historical business data of the data object pushed based on the historical push rules and the corresponding historical push results. The attribute data of the multiple data objects and the historical push records are sent to the first server.

[0082] Correspondingly, the first server receives the attribute data of the multiple data objects sent by the terminal device, and the historical push records are sent to the first server.

[0083] The first server first calls a preset push result prediction model, and determines a push success rate for pushing target business data to each of the multiple data objects based on the attribute data of the multiple data objects; and sorts the identity identifiers of the multiple data objects based on the push success rate to obtain a first push list.

[0084] Furthermore, the first server can determine the difference in push effectiveness between different data object type groups and the unbiased estimator of the association between push rules and data objects based on the attribute data of multiple data objects and historical push records; and adjust the order of the identity identifiers of the data objects in the first push list based on the difference in push effectiveness and the unbiased estimator of the association to obtain a second push list. Then send the second push list to the second server.

[0085] Correspondingly, the second server receives and pushes the target business data to multiple data objects according to the second push list and corresponding target push rules.

[0086] In this embodiment, the first server and the second server may specifically include a background server applied to one side of the business data processing platform, which can realize functions such as data transmission and data processing. Specifically, the first server and the second server may be, for example, an electronic device with data calculation, storage and network interaction functions. Alternatively, the first server and the second server may also be software programs running in the electronic device to provide support for data processing, storage and network interaction. In this embodiment, the number of servers included in the first server and the second server is not specifically limited. The first server and the second server may specifically be one server, or several servers, or a server cluster formed by several servers.

[0087] In this embodiment, the terminal device may specifically include a front-end device applied to the user side and capable of realizing functions such as data collection 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.

[0088] See also Figure 2 As shown, the embodiment of this specification provides a method for pushing business data. Among them, the method is specifically applied to the server side. When implemented specifically, the method may include the following contents.

[0089] S21: Acquire attribute data of multiple data objects and historical push records; wherein the historical push records record historical business data of the data objects pushed based on historical push rules and corresponding historical push results.

[0090] S22: calling a preset push result prediction model, and determining a push success rate of pushing the target business data to each of the multiple data objects according to the attribute data of the multiple data objects.

[0091] S23: Sort the identities of the multiple data objects according to the push success rate to obtain a first push list.

[0092] S24: Determine the difference in push effectiveness between different data object type groups and an unbiased estimator of the correlation between the push rules and the data objects based on the attribute data of the multiple data objects and the historical push records.

[0093] S25: According to the difference degree of the push effectiveness and the unbiased estimator of the correlation, the order of the identity identifiers of the data objects in the first push list is adjusted to obtain a second push list.

[0094] S26: Push target business data to multiple data objects according to the second push list and target push rules.

[0095] Through the above embodiments, the degree of difference in push effectiveness between different data object type groups and the unbiased estimator of the correlation between push rules and data objects can be determined and utilized to eliminate the errors caused by the preset push result prediction model not taking into account the fairness between different data object type groups and the correlation between data objects and push rules, thereby obtaining a second push list with relatively more accurate and reasonable sorting, and then based on the second push list, the target business data can be pushed to multiple data objects more accurately, thereby obtaining a better push effect and improving the push effectiveness.

[0096] In this embodiment, the data object may specifically include a user object. In specific implementation, the user object may be different types of user objects according to specific application scenarios.

[0097] Specifically, for example, in a scenario where a bank promotes financial products, the user object may be a customer object of the bank. For another example, in a scenario where an e-commerce company pushes products, the user object may be a shopper of a shopping website. Of course, the data objects listed above are only a schematic illustration. In specific implementations, in combination with specific application scenarios, the data objects may also include other types of user objects. This specification does not limit this.

[0098] In this embodiment, the attribute data of the data object may be specifically understood as parameter data that can reflect certain attribute characteristics of the data object.

[0099] In one embodiment, the attribute data of the data object may specifically include at least one of the following: the education level of the data object, the monthly income of the data object, the occupation of the data object, etc. Of course, the attribute data listed above is only a schematic illustration. In combination with specific application scenarios and processing requirements, other types of attribute data may also be introduced.

[0100] Through the above embodiments, the server can obtain and use more types of more diverse attribute data, so as to be able to sort multiple data objects more accurately and comprehensively.

[0101] In this embodiment, the above historical push record may specifically record the historical business data of the data object pushed based on the historical push rule, as well as the corresponding historical push result and other information.

[0102] In one embodiment, the historical push results include one or more of the following results: pushed and determined to be successful; pushed and determined to be failed; not pushed, etc. Of course, the historical push results listed above are only a schematic illustration. In specific implementation, further refinement can be made according to specific application scenarios and specific processing conditions to obtain more refined push results.

[0103] In this embodiment, the above method can avoid simply considering the two historical push results of successful push corresponding to positive sample data and failed push corresponding to negative sample data as in the existing method, which leads to ignoring the detailed information in the process of pushing historical business data to the data object according to the historical push rules. Richer and more detailed historical push results can be recorded and obtained, so that the correlation between the data object and the push rules and the pushed business data can be further analyzed according to the above historical push results to make up for the shortcomings of the preset push result prediction model.

[0104] In this embodiment, it should be noted that, usually when pushing business data, in order to improve the success rate of the push and make the data object more willing to accept and participate in the business activities involved in the business data, the business data is often pushed to the data object based on certain specific push rules.

[0105] Specifically, for example, when pushing financial products or financial services launched by the bank to bank customers (which can be understood as a kind of business data), specific push marketing can be carried out to bank customers based on a specific marketing plan or marketing strategy (which can be understood as a kind of push rule) to encourage bank customers to be willing to purchase the financial products or financial services and increase the order conversion rate.

[0106] In this embodiment, further, the above-mentioned historical push records can also record the historical push time of the data object, the initiator of the historical business data being pushed, historical push rules and other information related to the historical push experience of the data object, so that the above-mentioned historical push records can be used to more comprehensively and accurately analyze the acceptance degree of the data object when the corresponding target business data is pushed based on the target push rules, as well as the push effectiveness.

[0107] In one embodiment, the above-mentioned preset push result prediction model can be specifically understood as a model trained with positive sample data and negative sample data, which can predict the push success rate of pushing business data to the data object based on the attribute data of the data object.

[0108] In one embodiment, during specific implementation, the attribute data of each data object may be processed using a preset push result prediction model to predict the push success rate of each data object. Then, according to the push success rate of each data object, the identity identifiers of the data objects (e.g., the name, account name, object number, etc. of the data object) are sorted in descending order according to the push success rate to obtain the corresponding first push list.

[0109] In one embodiment, the above-mentioned preset push result prediction model can be trained specifically in the following manner: obtain multiple sample data; wherein each sample data at least includes attribute data of the corresponding sample data object and a push result; according to the push result of the sample data, label the sample data with a push result of successful push as positive sample data, and label the sample data with a push result of failed push as negative sample data, to obtain labeled sample data; use the labeled sample data to perform model training to obtain the preset push result prediction model.

[0110] In this embodiment, it should be noted that the preset push result prediction model trained in the above manner does not take into account the fairness of the prediction results, resulting in a tendency to favor data objects that are similar to the positive sample data and belong to the same data object type group when using the model for prediction, causing prediction bias and affecting the model accuracy.

[0111] In one embodiment, after obtaining the first push list, the fairness between different data object type groups and the errors caused by the correlation between data objects and push rules are also taken into consideration, and the degree of difference in push effectiveness between different data object type groups and the unbiased estimator of the correlation between push rules and data objects are further determined and adjusted in a targeted manner to the sorting of the identity identifiers of the data objects in the above-mentioned first push list, so as to obtain a second push list with more reasonable and accurate sorting.

[0112] In this embodiment, the multiple data objects included in each data object type group have a common feature in at least one attribute data.

[0113] In this embodiment, the difference in push results between the above-mentioned different data object type groups can be specifically understood as parameter data that can quantitatively reflect the degree of unfairness of the preset push result prediction model for different data object type groups during prediction.

[0114] In this embodiment, by determining and utilizing the differences in push effectiveness between the above-mentioned different data object type groups, the deviation introduced by using a preset push result prediction model that does not take into account the fairness of the prediction results can be compensated, thereby obtaining a push list with more reasonable and accurate sorting.

[0115] In one embodiment, the above-mentioned determination of the difference in push effectiveness between different data object type groups based on the attribute data of multiple data objects and historical push records may include the following contents during specific implementation.

[0116] S1: determining the data object type group to which each data object among the plurality of data objects belongs according to attribute data of the data object;

[0117] S2: Calculate the fairness value parameter of each data object type group according to the historical push records of the data objects in each data object type group; wherein the fairness value parameter is used to characterize the degree of partiality of the preset push result prediction model to the data object type group;

[0118] S3: Calculate the average push results of the data objects in each data object type group according to the historical push records of the data objects in each data object type group according to the historical push rules;

[0119] S4: Determine the degree of difference in push effectiveness between different data object type groups based on the fairness value parameters of each data object type group and the average value of push effectiveness of the data objects in each data object type group according to the historical push rules.

[0120] Through the above embodiments, the difference in push effectiveness between different data object type groups can be determined more accurately in a quantitative manner.

[0121] In one embodiment, the above-mentioned determining the data object type group to which each data object among the multiple data objects belongs based on the attribute data of the data objects may include the following contents during specific implementation: determining the data object type group to which each data object among the multiple data objects belongs by clustering the multiple data objects based on the attribute data of the multiple data objects.

[0122] Through the above embodiments, multiple data objects can be efficiently divided into corresponding data object type groups.

[0123] Of course, the above-listed method for determining the data object type group to which a data object belongs is only an illustrative description. In specific implementation, the division rules of different data object type groups can also be set according to specific application scenarios; and then the data object type group to which each data object belongs can be determined according to the division rules and the attribute data of the data object.

[0124] In one embodiment, the above-mentioned determination of the difference in push effectiveness between different data object type groups based on the fairness value parameters of each data object type group and the average value of push effectiveness of the data objects in each data object type group for the historical push rules may include the following contents during specific implementation:

[0125] The difference in push results between the data object type group numbered i and the data object type group numbered j is calculated using the following formula:

[0126]

[0127] in, Specifically, it can be expressed as the difference in push results between the data object type group numbered i and the data object type group numbered j. τ (G i ) can be specifically represented as the average value of the push results of the data objects in the data object type group numbered i for the historical push rules, Info τ (G j ) can be specifically represented as the average value of the push results of the data objects in the data object type group numbered j for the historical push rules, Value(G i ) can be specifically expressed as the fairness value parameter of the data object type group numbered i, Value(G j ) can be specifically expressed as the fairness value parameter of the data object type group numbered j.

[0128] In this embodiment, the unbiased estimator of the correlation between the above-mentioned push rules and data objects can be specifically understood as a parameter data that can quantitatively reflect the impact of the degree of fit between the data object and the push rules on the push effectiveness that is not taken into account by the preset push result prediction model during prediction.

[0129] In this embodiment, by determining and utilizing an unbiased estimator of the association between the above-mentioned push rules and data objects, the deviation introduced by using a preset push result prediction model that does not take into account the association between the data objects and the push rules can be compensated, thereby obtaining a push list with a further more reasonable and accurate sorting.

[0130] In one embodiment, the above-mentioned determination of the unbiased estimator of the association between the push rule and the data object based on the attribute data of the plurality of data objects and the historical push records may include the following contents during specific implementation.

[0131] S1: Based on the attribute data of the data object and the historical push rules, the corresponding unbiased estimator is constructed to solve the loss function;

[0132] S2: Determine the explicit relevance parameter of the data object relative to the historical push rule based on the historical push result;

[0133] S3: Calculate the global unbiased estimator of the data object ranking based on the push success rate and explicit correlation parameter of the data object;

[0134] S4: According to the global unbiased estimator of the data object and the loss function solved by the unbiased estimator, an unbiased estimator of the association between the corresponding push rule and the data object is calculated.

[0135] Through the above embodiments, an unbiased estimate of the correlation between the push rule and the data object can be determined more accurately in a quantitative manner.

[0136] In one embodiment, according to the degree of difference in the push effectiveness and the unbiased estimator of the correlation, targeted adjustments can be made based on the sorting position of the identity identifiers of the data objects in the first push list; and then, based on the sorting of the identity identifiers of the adjusted data objects, a second push list with a more reasonable and accurate sorting can be obtained.

[0137] In one embodiment, the above adjustment of the order of the identity identifiers of the data objects in the first push list according to the difference degree of the push effectiveness and the unbiased estimator of the correlation to obtain the second push list may include the following contents during specific implementation:

[0138] The ranking position of any data object d among the data objects in the second push list is determined according to the following formula:

[0139]

[0140] Among them, σ τ Specifically, it can be represented as the sorting position of the data object in the second push list, d can be represented as a data object, D can be represented as a data object set composed of multiple data objects, τ can be represented as time, v can be represented as a historical push rule, V can be represented as a historical push rule set composed of multiple historical push rules, λ can be represented as an adjustment parameter, Specifically, it can be expressed as an unbiased estimator of the association between the push rule and the data object.

[0141] In this embodiment, t represents a specific time point, Represents the attribute data of data object d at time point t, represents the rule characteristics of the historical push rule v used to push historical business data to the data object d at the time point t, and m represents the number of combinations between different data object type groups.

[0142] Through the above embodiment, the difference degree of the push effect and the unbiased estimator of the correlation can be used to adjust the ranking position of the identity identifiers of the data objects in the first push list in a targeted manner to obtain the corresponding second push list. The ranking of the identity identifiers of the data objects in the second push list is more reasonable and accurate than that of the first push list, and has a higher reference value.

[0143] In this embodiment, the data objects with higher identity identifiers in the second push list will have a relatively higher push success rate when the target business data is pushed. For example, customers with higher identity identifiers in the second push list are relatively more likely to be successfully marketed and purchase the promoted financial products.

[0144] In one embodiment, multiple data objects may be divided into multiple different push levels according to the second push list; data objects at different push levels are then distinguished, and target business data are pushed separately according to target push rules.

[0145] Specifically, for example, a group of data objects with front identities (for example, in the top 30%) in the second push list may be classified as a first push level, and the remaining data objects in the second push list may be classified as a second push level.

[0146] Since the push success rate of data objects in the first push layer is relatively high, according to the target push rules, the data objects in the first push layer can be pushed once based on a combination of push method 1 (for example, SMS push) and push method 2 (for example, email push).

[0147] Since the push success rate of data objects in the second push layer is relatively low, the data objects in the second push layer are pushed three times in a row based on a combination of push methods 1, push method 2, and push method 3 (for example, telephone push), so as to obtain a better push effect.

[0148] In this embodiment, before pushing target business data to multiple data objects, the attribute data of the data objects and the historical push records can be obtained first; then the preset push result prediction model can be called to predict the push success rate of the data objects according to the attributes of the data objects, and the identity identifiers of the data objects can be sorted based on the push success rate to obtain a first push list; further, the difference in push effectiveness between different data object type groups and the unbiased estimator of the correlation between the push rules and the data objects can be determined according to the attribute data of the data objects and the historical push records, and the first push list can be adjusted accordingly using the above data to eliminate the error caused by the preset push result prediction model not taking into account the fairness between different data object type groups and the correlation between the data objects and the push rules, so as to obtain a second push list with a relatively more accurate and reasonable sorting; and then, according to the second push list, the target business data can be pushed to multiple data objects more accurately, so as to obtain a better push effect, improve the push effect, and solve the technical problem that the sorting of the data objects in the generated push list is inaccurate and unreasonable, resulting in poor push effect when pushing business data to the data objects based on the push list in the existing method.

[0149] In one embodiment, further, considering that the preset push result processing model does not take into account the influence of the diversity of individual data objects on the prediction results during learning and training, the model may also have deviations in prediction. In order to obtain a push list with a relatively more accurate and reasonable ranking based on the second push list, the diversity distribution probability of the data objects can be further determined and used to adjust the ranking in the second push list again, so as to obtain a third push list with a relatively more accurate and reasonable ranking.

[0150] In one embodiment, after obtaining the second push list, the method may also include the following contents when implemented: determining the diversity distribution probability of data objects based on the attribute data of multiple data objects and historical push records; adjusting the sorting of the identity identifiers of the data objects in the second push list based on the diversity distribution probability of the data objects to obtain a third push list; accordingly, pushing target business data to multiple data objects based on the third push list and target push rules.

[0151] Through the above embodiment, a third push list with a relatively more accurate and reasonable sorting can be obtained, and the third push list can be used to replace the second push list to push target business data to the data object, thereby further improving the push effect of the target business data.

[0152] In one embodiment, the above-mentioned adjusting the order 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 the third push list may include the following contents during specific implementation.

[0153] 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 representation vector through a hidden layer in the processing model;

[0154] S2: Calculating an average diversity index according to the diversity distribution probability of the data object and the abstract feature representation vector;

[0155] S3: According to the average diversity index, the order of the identity identifiers of the data objects in the second push list is adjusted to obtain a third push list.

[0156] In this embodiment, during specific implementation, an abstract feature representation vector with relatively good effect can be extracted through the last hidden layer in the processing model for subsequent use.

[0157] In this embodiment, the above average diversity index can be specifically understood as parameter data that can quantitatively reflect the impact of diversity factors of data objects that are not considered by the preset push result prediction model during prediction on the push results.

[0158] Through the above embodiment, the average diversity index can be determined more accurately in a quantitative manner, and then the above diversity index can be used to adjust the sorting in the second push list based on the diversity factors of the data objects, so as to obtain a third push list with relatively more accurate and reasonable sorting and better push effect.

[0159] In one embodiment, the above method can be applied to a variety of different application scenarios to effectively push relevant business data in a variety of different application scenarios and obtain better push effects.

[0160] In one embodiment, the above method can be specifically applied to a product marketing scenario. In a product marketing scenario, the data object can specifically include a customer object, such as a financial management customer of a bank. The target business data includes a business product or service to be promoted, such as a newly launched financial management service of a bank. The target push rule can specifically include a currently adopted marketing plan, such as a marketing strategy currently designed by a bank for the financial management service.

[0161] Furthermore, the method provided in this specification can be used to determine the third push list of customer objects in the above-mentioned marketing scenario, and then the customer objects can be differentiated according to the third push list, and the corresponding marketing plan can be used to promote business products, thereby obtaining better marketing results and increasing the order conversion rate.

[0162] Of course, the marketing scenarios listed above are only schematic illustrations. In specific implementation, the method can also be applied to other types of application scenarios. Accordingly, the above data objects, push rules, and target business data can also include relevant data in other types of application scenarios. This specification does not limit this.

[0163] In one embodiment, during specific implementation, the third push list can be compared with the ranking in the first push list to obtain a comparison result; according to the comparison result, the preset push result prediction model can be adjusted in a targeted manner and updated. This can reduce the error of the preset push result prediction model and improve the model accuracy. Subsequently, the updated preset push result prediction model can be used to obtain a relatively more accurate and reasonable push success rate.

[0164] As can be seen from the above, the method for pushing business data provided by the embodiment of the present specification can first obtain the attribute data of the data object and the historical push records before pushing the target business data to multiple data objects; then call the preset push result prediction model to predict the push success rate of the data object according to the attribute of the data object, and sort the identity identifiers of the data object based on the push success rate to obtain a first push list; further, the difference degree of push effectiveness between different data object type groups and the unbiased estimator of the correlation between the push rule and the data object can be determined according to the attribute data of the data object and the historical push record, and the first push list can be adjusted accordingly using the above data to eliminate the error caused by the preset push result prediction model not taking into account the fairness between different data object type groups and the correlation between the data object and the push rule, so as to obtain a second push list with a relatively more accurate and reasonable sorting; and then according to the second push list, the target business data can be pushed to multiple data objects more accurately, so as to obtain a better push effect, improve the push effect, and solve the technical problem that the sorting of the data objects in the generated push list is inaccurate and unreasonable, resulting in poor push effect when pushing business data to the data object based on the push list in the existing method. After determining the second push list, the diversity factor of the data objects is taken into consideration, and the average diversity index is determined according to the diversity distribution probability of the data objects; the above average diversity index is then used to adjust the sorting of the data objects in the second push list again, so as to obtain a third push list with a relatively more accurate and reasonable sorting and better push effect; and the above third push list can be used to push target business data to the data objects, thereby further improving the push effect.

[0165] The embodiment of the present specification also provides a server, including a processor and a memory for storing processor executable instructions, wherein the processor can perform the following steps according to the instructions when implemented: obtaining attribute data of multiple data objects and historical push records; wherein the historical push records record historical business data of the data objects pushed based on historical push rules, and corresponding historical push results; calling a preset push result prediction model, and determining a push success rate of pushing target business data to each of the multiple data objects according to the attribute data of the multiple data objects; sorting the identity identifiers of the multiple data objects according to the push success rate to obtain a first push list; determining the difference degree of push effectiveness between different data object type groups and an unbiased estimator of the correlation between the push rules and the data objects according to the attribute data of the multiple data objects and the historical push records; adjusting the sorting of the identity identifiers of the data objects in the first push list according to the difference degree of push effectiveness and the unbiased estimator of the correlation to obtain a second push list; and pushing the target business data to the multiple data objects according to the second push list and the target push rules.

[0166] In order to complete the above instructions more accurately, refer to Figure 3 As shown, the embodiment of this specification also provides another specific server, wherein the server includes a network communication port 31, a processor 32 and a memory 33, and the above structures are connected through internal cables so that each structure can perform specific data interaction.

[0167] The network communication port 31 can be used to obtain attribute data of multiple data objects and historical push records; wherein the historical push records record historical business data of the data objects pushed based on historical push rules and corresponding historical push results.

[0168] The processor 32 can be specifically used to call a preset push result prediction model, and determine the push success rate of pushing target business data to each of the multiple data objects according to the attribute data of the multiple data objects; sort the identity identifiers of the multiple data objects according to the push success rate to obtain a first push list; determine the degree of difference in push effectiveness between different data object type groups and an unbiased estimator of the association between push rules and data objects according to the attribute data of the multiple data objects and historical push records; adjust the sorting of the identity identifiers of the data objects in the first push list according to the degree of difference in push effectiveness and the unbiased estimator of the association to obtain a second push list; push the target business data to the multiple data objects according to the second push list and the target push rules.

[0169] The memory 33 may be specifically used to store corresponding instruction programs.

[0170] In this embodiment, the network communication port 31 can be a virtual port that is bound to different communication protocols so that different data can be sent or received. For example, the network communication port can be a port responsible for web data communication, a port responsible for FTP data communication, or a port responsible for email data communication. In addition, the network communication port can also be a physical communication interface or communication chip. For example, it can be a wireless mobile network communication chip, such as GSM, CDMA, etc.; it can also be a Wifi chip; it can also be a Bluetooth chip.

[0171] In this embodiment, the processor 32 may be implemented in any suitable manner. For example, the processor may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) executable by the (micro)processor, a logic gate, a switch, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc. This specification does not limit this.

[0172] In this embodiment, the memory 33 may include multiple levels. In a digital system, anything that can store binary data can be a memory; in an integrated circuit, a circuit with a storage function but no physical form is also called a memory, such as RAM, FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory stick, TF card, etc.

[0173] The embodiment of the present specification also provides a computer storage medium based on the above-mentioned business data push method, wherein the computer storage medium stores computer program instructions, and when the computer program instructions are executed, the following are implemented: obtaining attribute data of multiple data objects and historical push records; wherein the historical push records record historical business data of the data objects pushed based on historical push rules, and corresponding historical push results; calling a preset push result prediction model, and determining a push success rate of pushing target business data to each of the multiple data objects according to the attribute data of the multiple data objects; sorting the identity identifiers of the multiple data objects according to the push success rate to obtain a first push list; determining the difference degree of push effectiveness between different data object type groups and an unbiased estimator of the correlation between the push rules and the data objects according to the attribute data of the multiple data objects and the historical push records; adjusting the sorting of the identity identifiers of the data objects in the first push list according to the difference degree of push effectiveness and the unbiased estimator of the correlation to obtain a second push list; and pushing the target business data to the multiple data objects according to the second push list and the target push rules.

[0174] In this embodiment, the storage medium includes, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a cache, a hard disk (HDD), or a memory card. The memory may be used to store computer program instructions. The network communication unit may be an interface for network connection communication set in accordance with the standard specified by the communication protocol.

[0175] In this embodiment, the functions and effects specifically implemented by the program instructions stored in the computer storage medium can be explained in comparison with other implementations and will not be described in detail here.

[0176] See also Figure 4 As shown, at the software level, the embodiment of this specification also provides a device for pushing business data, which may specifically include the following structural modules.

[0177] The acquisition module 41 may be specifically used to acquire attribute data of multiple data objects and historical push records; wherein the historical push records record historical business data of the data objects pushed based on historical push rules and corresponding historical push results;

[0178] The first determination module 42 may be specifically configured to call a preset push result prediction model to determine a push success rate of pushing target business data to each of the plurality of data objects according to the attribute data of the plurality of data objects;

[0179] A first sorting module 43 may be specifically configured to sort the identity identifiers of the plurality of data objects according to the push success rate to obtain a first push list;

[0180] The second determination module 44 may be specifically used to determine the difference in push effectiveness between different data object type groups and an unbiased estimate of the correlation between the push rules and the data objects based on the attribute data of the multiple data objects and the historical push records;

[0181] The second sorting module 45 may be specifically configured to adjust the sorting of the identity identifiers of the data objects in the first push list according to the difference in the push effectiveness and the unbiased estimator of the correlation to obtain a second push list;

[0182] The push module 46 may be specifically configured to push target business data to multiple data objects according to the second push list and target push rules.

[0183] In one embodiment, the device may further include a third determination module and a third sorting module, wherein:

[0184] The third determination module may be specifically configured to determine the diversity distribution probability of the data objects according to the attribute data of the plurality of data objects and the historical push records;

[0185] The third sorting module may be specifically configured to adjust the 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;

[0186] Accordingly,

[0187] The push module can be specifically used to push target business data to multiple data objects according to the third push list and target push rules.

[0188] It should be noted that the units, devices or modules described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described separately by functions divided into various modules. Of course, when implementing this specification, the functions of each module can be implemented in the same or more software and / or hardware, or the modules that implement the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0189] As can be seen from the above, the business data push device provided in the embodiment of the present specification, after predicting the push success rate of the data object by using the preset push result prediction model and generating the first push list based on the push success rate, also determines the degree of difference in push effectiveness between different data object type groups, and the unbiased estimator of the correlation between the push rules and the data objects, and uses the above data to adjust the first push list to eliminate the errors caused by the preset push result prediction model due to the failure to take into account the fairness between different data object type groups and the correlation between the data objects and the push rules, and obtains a second push list with relatively more accurate sorting; and then, according to the second push list, the target business data can be pushed to multiple data objects, thereby obtaining a better push effect.

[0190] In a specific scenario example, the business data push method provided in this specification can be applied to generate a corresponding customer marketing list, and marketing to customers based on the list can achieve better marketing results. For specific implementation, please refer to the following content.

[0191] For marketing scenarios, during specific implementation, a corresponding system can be built based on the business data push method provided in this specification to be responsible for specific business marketing.

[0192] Based on the functional processing level, the system can specifically include four main functional modules: customer marketing probability prediction module, fair and unbiased dynamic adjustment module, diversity re-ranking module, and customer marketing labeling module.

[0193] Based on the system structure level, you can refer to Figure 5As shown, the system may include two parts: a machine learning system and a customer marketing system. Furthermore, the machine learning system may include three devices, namely: a machine learning model device, a fair and unbiased dynamic adjustment device, and a reordering device. Among them, the customer marketing probability prediction module is implemented in the machine learning model device, the fair and unbiased dynamic adjustment module is implemented in the fair and unbiased dynamic adjustment device, and the diversity reordering module is implemented in the reordering device. The customer marketing system may include a business marketing device. Among them, the customer marketing labeling module is implemented in the business marketing device.

[0194] For details, please refer to Figure 6 As shown. The main functional modules include the following: Customer Marketing Probability Prediction Module 1: Use a supervised machine learning model (e.g., a preset push result prediction model) to predict the probability of a customer (e.g., a data object) being successfully marketed (e.g., push success rate). Fair and Unbiased Dynamic Adjustment Module 2: Dynamically adjust the fairness metric to achieve fair and unbiased dynamic adjustment of the predicted ranking (e.g., determine and adjust the ranking in the first push list based on the degree of difference in push effectiveness between different data object type groups, and the unbiased estimator of the correlation between the push rules and the data objects, to obtain a second push list). Diversity Reordering Module 3: Adjust the diversity of the predicted ranking to achieve predicted reordering (e.g., use the average diversity index determined based on the diversity distribution probability of the data object to adjust the ranking of the identity identifiers of the data objects in the second push list to obtain a third push list). Customer Marketing Annotation Module 4: Annotate the customer marketing results and provide feedback to the machine learning module.

[0195] For specific implementation, see Figure 7 As shown, the customer marketing probability prediction module 1 may include the following contents during specific operation.

[0196] Step S101: Design feature variables (eg, attribute data of a data object). Design model feature variables according to the modeling purpose. The marketing model feature variables can be designed according to common customer characteristics.

[0197] Step S102: Label the customer ID. Collect customer labels, which are mainly divided into successful marketing customers and unsuccessful marketing customers. The modeling data set is collected with customers as the dimension, and a customer record is used as a sample data. The number of successful marketing customers and unsuccessful marketing customer records constitutes the total sample data set for modeling.

[0198] Let the customer group be D. For each customer d, d∈D. Let the customer feature set be X. For each customer's feature variable x d , there are x d ∈X.

[0199] Furthermore, the time factor is also considered to establish time series data. For each time point, there are: t=1,...,τ.

[0200] Let the set of marketing plans (e.g., push rules) be V. For each plan v, v∈V. The set of marketing plan feature variables Z. For each marketing plan feature variable z v , there is v ∈Z.

[0201] For each marketing plan v, the following annotation information can be collected:

[0202]

[0203] Among them, y t (d|v) is the explicit correlation between the degree of fit between marketing strategy and customers, and the implicit correlation behind it is rel t (d|v)∈[0,1]. In an ideal state, y t (d|v)=E[rel t (d|v)].

[0204] Step S103: Generate model prediction time series results. Take the time series data and generate the marketing probability p corresponding to each customer under different marketing plans for each time t=1,...,τ. t (d|v).

[0205] See also Figure 8 As shown, the fair and unbiased dynamic adjustment module 2 may include the following contents when it is specifically run.

[0206] Step S201: Fairness difference measurement. Classify by customer group (for example, determine the type group of data objects to which the data object belongs), G = {G1, ..., G m}.

[0207] The classification method of customer groups can be based on expert rules and customer feature attributes, or it can be based on machine learning models such as clustering models. The main reason is that supervised marketing prediction models have similar probability values ​​when predicting. Therefore, after dividing the customer groups, the ranking of similar customer groups can be improved by using a fair and unbiased dynamic adjustment method.

[0208] Calculate the fairness value (eg, the fairness value parameter):

[0209] Among them, R t(d|v) is the global ranking value of customers in different marketing plans at different times. First, calculate the average global ranking value of each customer under different marketing plans at different times. Then calculate the average fairness value of the same customer group. The higher the fairness value, the higher the ranking of the group on the marketing list on average.

[0210] Calculate the degree of marketing difference. The greater the difference, the greater the difference in marketing probability and effectiveness between different groups. The purpose of adjusting the marketing probability is to narrow the difference.

[0211] Calculate the information value of the marketing probability. Calculate the average marketing probability of each customer at different times, and then calculate the average marketing probability of the customer group:

[0212] For all marketing plans, there are:

[0213]

[0214] Calculate the information value of marketing effectiveness. Calculate the average marketing effectiveness of each customer at different times, and then calculate the average marketing effectiveness of the customer group:

[0215] For all marketing plans, there are:

[0216]

[0217] Using the concept of entropy to calculate the average value of marketing information, we have:

[0218]

[0219] Calculate the degree of marketing differences between different customer groups (for example, the degree of difference in push results between different data object type groups):

[0220]

[0221] get

[0222] Ω τ Indicates the degree of difference between two groups. The larger the value, the greater the difference.

[0223] Step S202: Unbiased measurement. Calculate the correlation between the marketing strategy and the degree of fit between the customer t (d|v) unbiased estimator and global ranking R(d|v) unbiased estimator of customer marketing ranking.

[0224] Calculate the correlation between marketing strategy and customer fit t (d|v) unbiased estimator. Let the corresponding model function form be Solve again

[0225] Its unbiased estimator solves the loss function as:

[0226]

[0227] in, is the characteristic variable of each customer d at time t, The characteristic variables of each marketing plan v at time t.

[0228] The unbiased estimator of the global ranking R(d|v) of customer marketing ranking is:

[0229] Step S203: Fair dynamic adjustment. Calculate the global ranking strategy. Specifically, the marketing ranking (for example, ranking position) of each customer can be calculated according to the following formula:

[0230]

[0231] Among them, σ τ represents the marketing ranking of each customer at time τ, It represents the average degree of association under different marketing plans at different times.

[0232] Among them, the hyperparameter λ is used to adjust the sorting strategy and regulate the influence of marketing difference on sorting. The present invention takes the sum of the marketing difference. According to different usage needs, Ω τ It can be adjusted to the maximum, minimum, or median of the marketing difference. The higher the λ, the greater the adjustment to the original ranking.

[0233] It should be added that the number of the above marketing plans V can be adjusted dynamically. If customer ranking is designed for a marketing plan, V=1 in the above formula.

[0234] See also Fig. 9 As shown, the diversity reordering module 3 may include the following contents when it is specifically run.

[0235] Step 301: Calculate ideal diversity distribution: Calculate the ideal customer diversity distribution pattern.

[0236] The marketing plan set is V. For each plan v, v∈V. Design the characteristic variable set Z of the diversity in the ideal state of marketing plan ideal . For each marketing plan’s characteristic variables have For example, the feature variable set is [education level distribution, asset distribution]. Education level distribution [high school, undergraduate, undergraduate and above], asset distribution [100,000, 100,000-1 million, 1 million-5 million, 5 million and above]. For each marketing plan, design its ideal diversity distribution. For example, marketing plan v1 targets the diversity of customers with different education levels, so the designed one-hot feature vector is [1,1,1], without considering asset distribution. If marketing plan v2 targets the diversity of customers with multiple education levels and assets under 1 million, the designed one-hot feature vector is [1,1,1,1,1,0,0].

[0237] Let the ideal distribution set be F, and for each v there is an ideal distribution f v , there is f v ∈F. Calculate the Hellinger distance between the customer group D and the marketing plan v features:

[0238]

[0239] Among them, the probability distribution of this distance is the ideal distribution of diversity

[0240] Step 302: Calculate the context embedding vector parameters.

[0241] Using LSTM model:

[0242] Among them, {d1,...,d s} is the sequence of customer groups sorted according to the ideal distribution probability, hidden n-1 ,hidden n Indicates the number of hidden layers, cell n It is the state of the LSTM cell at layer n. are the parameters of LSTM.

[0243] The target output is the ideal probability distribution of the aforementioned step S301.

[0244] Calculate the model parameters

[0245] The Context Embedding Vector can be obtained by taking hidden n The output of the layer is obtained.

[0246] Step 303: Diversity measurement: Calculate the average diversity index.

[0247]

[0248] in, p t (d|v) is consistent with the definition of the above steps S202 and S103. Here p t (d|v) represents The weight represents the weighted ranking value under different marketing plans. Indicates the similarity between two customers.

[0249] The Hellinger distance can be used to calculate the similarity:

[0250]

[0251] It should be noted that this formula indicates that diversity is considered when weighted ranking values ​​are considered. Wherein, k represents the k-th dimension Context Embedding Vector feature vector.

[0252] When calculating diversity, the sorted σt in step S203 is connected to the LSTM model in the previous step.

[0253] Get each customer's hidden n As Context Embedding Vector. Among them, γ is a hyperparameter, and the higher the value, the richer the diversity.

[0254] Step 304: Customer re-ranking. The desired degree of fairness and diversity can be adjusted by adjusting the hyperparameters λ and γ.

[0255] According to the following formula, the utility value function of the evaluation ranking is:

[0256] stMLR(D)∈[floor,cap]

[0257] Among them, σ τ (d) indicates that customer d is in σ τ Utility represents the overall ranking utility value. It is required that the utility reaches the expected goal under certain diversity conditions. ExpectedUtility, floor, and cap are set values.

[0258] After obtaining the ranking value of the satisfied utility, the customers who meet the expected results are re-ranked by σ to obtain the final customer marketing list (for example, the third push list).

[0259] See also Fig.10 As shown, the customer marketing labeling module 4 may include the following contents when it is specifically run.

[0260] Step S401: Customer marketing: In the customer marketing system, marketing is carried out according to the customer list.

[0261] Step S402: Expand marketing annotation data. Based on the dynamically adjusted customer marketing list, record whether the customer has been marketed and purchased the corresponding product. Expand the annotation information not captured in the original marketing model, and it can also be used to update the customer marketing model.

[0262] To compare the above process with existing methods, see Fig.11 As shown in the figure, the upper part (corresponding to before implementation) is the system architecture status and data flow based on the existing method. The lower part (corresponding to after implementation) is the system architecture status and data flow based on this method.

[0263] Through comparison, it is found that based on the existing methods, the customer marketing model uses a supervised algorithm model to predict the probability of successful marketing of customers according to the customer's characteristic variables, and obtains the customer marketing list according to the probability. The basic principle of this algorithm is to collect sample data on whether the customer has been successfully marketed, and train the prediction and adjustment model according to the customer's characteristic variables and positive and negative samples. In terms of the algorithm, customers similar to the positive sample (customers who have been successfully marketed) have a higher prediction probability; customers similar to the negative sample (customers who have not been successfully marketed) have a lower prediction probability. The problems with this algorithm are: First, the algorithm does not consider the fairness of the prediction results. The results favor positive sample customers, high-probability customers continue to get high probabilities, and low-probability customers continue to get low probabilities, and there is a state of "the rich get richer and the poor get poorer". Second, the algorithm does not consider the unbiasedness of the prediction results. The algorithm does not calculate the correlation between marketing strategies and customer preferences, and the prediction results are biased. Third, the algorithm does not consider the diversity of customer types. The distribution and types of customers in the marketing strategy are not explicitly considered in the algorithm, which easily leads to the simplification of customer types.

[0264] Compared with the existing methods, this method adds a fair and unbiased adjustment algorithm strategy to the original marketing prediction model results, improves the fairness and unbiasedness of the prediction results, adds a diversified re-ranking strategy, and achieves diversity in results. The algorithm implementation device can adaptively and dynamically adjust the ranking of customers on the marketing list according to data changes, and connect the adjusted and ranked list to the marketing business system. It makes the original business system transparent and does not affect the business operation process of the business system.

[0265] Through the above scenario examples, it is verified that based on the method provided in this manual, a dynamic adjustment mechanism of fairness, unbiasedness and diversity is added to the customer marketing prediction algorithm. First, the algorithm increases the fairness and unbiasedness of the prediction results. Narrow the state of "the rich get richer and the poor get poorer". Second, the algorithm adds a diverse evaluation measurement. Incorporate customer diversity into the algorithm. Third, the algorithm adjustment mechanism can seamlessly connect to the original business process and realize a dynamic closed loop of the entire process. 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 to efficiently realize the corresponding data processing.

[0266] Although the present specification provides method operation steps as described in the embodiments or flow charts, more or less operation steps may be included based on conventional or non-creative means. The order of steps listed in the embodiments is only one way of executing the order of many steps, and does not represent a unique execution order. When the device or client product in practice is executed, it can be executed in sequence or in parallel according to the method shown in the embodiments or the drawings (for example, a parallel processor or a multi-threaded processing environment, or even a distributed data processing environment). The term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, product or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such a process, method, product or device. In the absence of more restrictions, it is not excluded that there are other identical or equivalent elements in the process, method, product or device including the elements. The first, second, etc. words are used to represent the name, and do not represent any particular order.

[0267] Those skilled in the art also know that, in addition to implementing the controller in a purely computer-readable program code, the controller can be made to implement the same function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered as a hardware component, and the devices for implementing various functions included therein can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules for implementing the method and structures within the hardware component.

[0268] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media including storage devices.

[0269] Through the description of the above embodiments, it can be known that those skilled in the art can clearly understand that the present specification can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present specification can essentially be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the methods described in each embodiment of the present specification or some parts of the embodiments.

[0270] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. This specification can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0271] Although the present specification is described through embodiments, those skilled in the art will appreciate that there are many modifications and changes to the present specification without departing from the spirit of the present specification, and it is intended that the appended claims include these modifications and changes without departing from the spirit of the present specification.

Claims

1. A method for pushing business data, characterized in that: include: Acquire attribute data of multiple data objects and historical push records; wherein the historical push records record historical business data of the data objects pushed based on historical push rules and corresponding historical push results; Calling a preset push result prediction model to determine a push success rate of pushing target business data to each of the multiple data objects according to the attribute data of the multiple data objects; According to the push success rate, the identity identifiers of the multiple data objects are sorted to obtain a first push list; Based on the attribute data of multiple data objects and historical push records, determine the difference in push effectiveness between different data object type groups and an unbiased estimator of the correlation between push rules and data objects; According to the difference degree of the push effectiveness and the unbiased estimator of the correlation, the order of the identity identifiers of the data objects in the first push list is adjusted to obtain a second push list; Pushing target business data to multiple data objects according to the second push list and target push rules; Among them, according to the attribute data of multiple data objects and the historical push records, the difference degree of push effectiveness between different data object type groups is determined, including: according to the attribute data of the data objects, determining the data object type group to which each data object among the multiple data objects belongs; according to the historical push records of the data objects in each data object type group, respectively calculating the fairness value parameters of each data object type group; wherein the fairness value parameters are used to characterize the degree of favoritism of the preset push result prediction model to the data object type group; according to the historical push records of the data objects in each data object type group, respectively calculating the average value of the push effectiveness of the data objects in each data object type group for the historical push rules; according to the fairness value parameters 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, the difference degree of push effectiveness between different data object type groups is determined.

2. The method according to claim 1, characterized in that The attribute data of the data object includes at least one of the following: the education level of the data object, the monthly income of the data object, and the occupation of the data object.

3. The method according to claim 1, characterized in that: The historical push results include one or more of the following results: pushed and determined to be successful; pushed and determined to be failed; not pushed.

4. The method according to claim 1, characterized in that: Determining, according to the attribute data of the data object, a data object type group to which each of the plurality of data objects belongs, comprises: According to the attribute data of the multiple data objects, clustering processing is performed on the multiple data objects to determine the data object type group to which each data object in the multiple data objects belongs.

5. The method according to claim 1, characterized in that Determining the difference in push effectiveness between different data object type groups according to the fairness value parameters of each data object type group and the average push effectiveness of the data objects in each data object type group according to the historical push rules includes: The difference in push results between the data object type group numbered i and the data object type group numbered j is calculated using the following formula: Among them, diff τ E (G i ,G j ) is the difference in push results between the data object type group numbered i and the data object type group numbered j, Info τ (G i ) is the average value of the push results of the data objects in the data object type group numbered i for the historical push rules, Info τ (G j ) is the average value of the push results of the data objects in the data object type group numbered j for the historical push rules, Value(G i ) is the fairness value parameter of the data object type group numbered i, Value(G j ) is the fairness value parameter of the data object type group numbered j.

6. The method according to claim 2, characterized in that Based on the attribute data of multiple data objects and historical push records, an unbiased estimator of the association between the push rules and the data objects is determined, including: According to the attribute data of the data object and the historical push rules, the corresponding unbiased estimator is constructed to solve the loss function; According to the historical push results, determine the explicit relevance parameters of the data object relative to the historical push rules; According to the push success rate and explicit correlation parameter of the data objects, a global unbiased estimator of the data object ranking is calculated; According to the global unbiased estimator of the data object and the loss function solved by the unbiased estimator, an unbiased estimator of the association between the corresponding push rule and the data object is calculated.

7. The method according to claim 1, characterized in that According to the difference degree of the push results and the unbiased estimator of the correlation, the order of the identity identifiers of the data objects in the first push list is adjusted to obtain a second push list, including: The ranking position of each data object in the second push list is determined according to the following formula: Among them, σ τ is the ranking position of the data object in the second push list, d is the data object, D is the data object set consisting of multiple data objects, τ is time, v is the historical push rule, V is the historical push rule set consisting of multiple historical push rules, λ is the adjustment parameter, It is an unbiased estimator of the association between push rules and data objects.

8. The method according to claim 1, characterized in that After obtaining the second push list, the method further includes: Determine the diversity distribution probability of data objects based on the attribute data of multiple data objects and historical push records; According to the diversity distribution probability of the data objects, the order of the identity identifiers of the data objects in the second push list is adjusted to obtain a third push list; Accordingly, According to the third push list and target push rules, target business data is pushed to multiple data objects.

9. The method according to claim 8, characterized in that According to the diversity distribution probability of the data objects, the order of the identity identifiers of the data objects in the second push list is adjusted to obtain a third push list, including: 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 representation 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 representation vector; According to the average diversity index, the order of the identity identifiers of the data objects in the second push list is adjusted to obtain a third push list.

10. The method according to claim 1, characterized in that The data object includes a customer object, the target push rule includes a currently adopted marketing plan, and the target business data includes a business product or business service to be promoted.

11. A device for pushing business data, characterized in that: include: An acquisition module, used to acquire attribute data of multiple data objects and historical push records; wherein the historical push records record historical business data of the data objects pushed based on historical push rules and corresponding historical push results; A first determination module is used to call a preset push result prediction model to determine a push success rate of pushing target business data to each of the multiple data objects according to the attribute data of the multiple data objects; A first sorting module, configured to sort the identities of the plurality of data objects according to the push success rate to obtain a first push list; A second determination module is used to determine the difference in push effectiveness between different data object type groups and an unbiased estimator of the correlation between the push rules and the data objects based on the attribute data of the multiple data objects and the historical push records; A second sorting module is used to adjust the sorting of the identity identifiers of the data objects in the first push list according to the difference degree of the push effectiveness and the unbiased estimator of the correlation to obtain a second push list; A push module, configured to push target business data to multiple data objects according to the second push list and target push rules; Among them, the second determination module is specifically used to determine the data object type group to which each data object among the multiple data objects belongs based on the attribute data of the data object; calculate the fairness value parameters of each data object type group based on the historical push records of the data objects in each data object type group; wherein the fairness value parameters are used to characterize the degree of favoritism of the preset push result prediction model to the data object type group; calculate the average value of the push effectiveness of the data objects in each data object type group for the historical push rules based on the historical push records of the data objects in each data object type group; determine the degree of difference in push effectiveness between different data object type groups based on the fairness value parameters 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.

12. The device according to claim 11, characterized in that The device further includes a third determining module and a third sorting module, wherein: The third determination module is used to determine the diversity distribution probability of the data objects according to the attribute data of the multiple data objects and the historical push records; The third sorting module is used to adjust the sorting 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 push module is used to push target business data to multiple data objects according to the third push list and target push rules.

13. A server, characterized in that: The invention comprises a processor and a memory for storing instructions executable by the processor, wherein the steps of the method according to any one of claims 1 to 10 are implemented when the processor executes the instructions.

14. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the instructions are executed, the steps of the method according to any one of claims 1 to 10 are implemented.

Images