CN115345653A

CN115345653A - Gain value acquisition method and device, storage medium and electronic equipment

Info

Publication number: CN115345653A
Application number: CN202210952101.3A
Authority: CN
Inventors: 范芳芳; 赵叶宇; 方彦明
Original assignee: Alipay Hangzhou Information Technology Co Ltd
Current assignee: Alipay Hangzhou Information Technology Co Ltd
Priority date: 2022-08-09
Filing date: 2022-08-09
Publication date: 2022-11-15

Abstract

The application discloses a gain value obtaining method, a gain value obtaining device, a storage medium and electronic equipment, wherein the method comprises the following steps: the method comprises the steps of obtaining user characteristic data, obtaining popularization affair vectors corresponding to the popularization affair characteristic data based on an embedded layer in a gain model, obtaining coding characteristic data corresponding to the user characteristic data based on a coding layer in the gain model, adopting a cross layer in the gain model, and obtaining gain values corresponding to users under the influence of popularization affairs according to the popularization affair vectors and the coding characteristic data. By adopting the method and the device, the gain value of the user under the influence of the promotion affair can be obtained by cross fusion of the characteristics of the promotion affair and the characteristics of the user, the influence of the promotion affair on the user is obtained, and the accuracy of pushing the promotion affair is improved.

Description

Gain value acquisition method and device, storage medium and electronic equipment

Technical Field

The present disclosure relates to the field of computer technologies, and in particular, to a method and an apparatus for obtaining a gain value, a storage medium, and an electronic device.

Background

According to historical data generated by various browsing and consuming behaviors of a user on the network, the user can be depicted, so that the user can know the favorite information of the user, the user behavior and the like can be influenced by promotion transactions, for example, the consuming behavior of the user can be influenced by promotion transactions such as advertisements and discounts, but the influence of different users is different, and a method for acquiring the influence of the promotion transactions of the user is needed.

Disclosure of Invention

The embodiment of the application provides a gain value obtaining method and device, a storage medium and an electronic device, which can obtain a gain value of a user under the influence of a promotion transaction by performing cross fusion on the characteristics of the promotion transaction and the characteristics of the user, obtain the influence of the promotion transaction on the user, and improve the accuracy of pushing the promotion transaction. The technical scheme is as follows:

in a first aspect, an embodiment of the present application provides a method for obtaining a gain value, where the method includes:

acquiring user characteristic data and acquiring popularization transaction characteristic data;

acquiring a promotion transaction vector corresponding to the promotion transaction characteristic data based on an embedded layer in a gain model, and acquiring coding characteristic data corresponding to the user characteristic data based on a coding layer in the gain model;

and acquiring a gain value corresponding to the user under the influence of the promotion affair by adopting a cross layer in the gain model and according to the promotion affair vector and the coding feature data.

In a second aspect, an embodiment of the present application provides a gain value obtaining apparatus, where the apparatus includes:

the characteristic data acquisition module is used for acquiring user characteristic data and acquiring popularization transaction characteristic data;

the embedding module is used for acquiring a promotion transaction vector corresponding to the promotion transaction characteristic data based on an embedding layer in a gain model and acquiring coding characteristic data corresponding to the user characteristic data based on a coding layer in the gain model;

and the gain value acquisition module is used for acquiring the gain value corresponding to the user under the influence of the promotion affair according to the promotion affair vector and the coding feature data by adopting the cross layer in the gain model.

In a third aspect, embodiments of the present application provide a computer storage medium storing a plurality of instructions adapted to be loaded by a processor and to perform the above-mentioned method steps.

In a fourth aspect, the present application provides a computer program product, which stores a plurality of instructions adapted to be loaded by a processor and execute the above method steps.

In a fifth aspect, an embodiment of the present application provides an electronic device, which may include: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to perform the above-mentioned method steps.

In one or more embodiments of the present application, user feature data are obtained, promotion transaction feature data are obtained, a promotion transaction vector corresponding to the promotion transaction feature data is obtained based on an embedded layer in a gain model, coding feature data corresponding to the user feature data are obtained based on a coding layer in the gain model, a cross layer in the gain model is adopted, and a gain value corresponding to the user under the influence of the promotion transaction is obtained according to the promotion transaction vector and the coding feature data. The method and the device have the advantages that the gain value of the user under the influence of the promotion affair is obtained by cross fusion of the characteristics of the promotion affair and the characteristics of the user, the influence of the user on the promotion affair is obtained, and the accuracy of pushing the promotion affair is improved.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings used in the embodiments or the prior art descriptions will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative efforts.

Fig. 1 is a schematic flowchart of a gain value obtaining method according to an embodiment of the present application;

fig. 2 is a schematic flowchart of a gain value obtaining method according to an embodiment of the present application;

fig. 3 is a schematic diagram illustrating an example of obtaining an actual gain value according to an embodiment of the present application;

FIG. 4 is a schematic structural diagram of an initial model provided in an embodiment of the present application;

fig. 5 is a schematic structural diagram of a gain value obtaining apparatus according to an embodiment of the present application;

fig. 6 is a schematic structural diagram of a gain value obtaining apparatus according to an embodiment of the present application;

fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

The gain value acquisition device is used for realizing the gain value acquisition method, the gain value acquisition device can acquire user characteristic data and promotion transaction characteristic data, the user characteristic data is characteristic data reflecting characteristic information of a user, the user characteristic data can comprise characteristic data on at least one characteristic type of the user, the gain value acquisition device can record operation data of the user when the user carries out consumption behaviors and browsing behaviors in the past, the operation data of the user can be stored in a historical database, and the gain value acquisition device can acquire the operation data of the user from the historical database and extract the user characteristic data according to the operation data. The feature types may include the years, occupation, whether to marry, and other basic features reflecting the user attributes of the user, and may further include window features, timing features, cross-combination features, and the like. It will be appreciated that the window characteristic may be a time-sensitive characteristic, such as the amount of money the user has spent in the most recent period of time, the time-series characteristic may be a characteristic that reflects the time-series, such as different amounts of money the user has spent at the same time each year, and the cross-combination characteristic is at least two associated characteristics. The promotion transaction is a promotion transaction pushed by a user, for example, an advertisement and a discount offer pushed by the user, the promotion transaction characteristic data is characteristic data of the promotion transaction, and may reflect a type of the promotion transaction, for example, the advertisement or the discount offer, and may also represent specific contents of the promotion transaction, for example, when the promotion transaction is the discount offer, the promotion transaction characteristic data may also reflect that several discount offers are pushed to the user. The gain value acquisition device can input the user characteristic data and the promotion transaction characteristic data into the gain model to obtain the corresponding gain value of the user under the influence of the promotion transaction. The gain value may reflect the influence of the promotion transaction on the user, for example, the gain value may be a difference between the amount of money consumed by the user after being influenced by the promotion transaction and the amount of money consumed without being influenced by the promotion transaction, for example, if the user consumes 1000 yuan in the month without being influenced by the promotion transaction, but the user consumes 1500 yuan in the month after being influenced by the promotion transaction such as advertisement, the gain value may consume 500 yuan.

The gain value acquisition device can acquire sample characteristic data of at least one sample user, wherein the sample characteristic data comprises target characteristic data of the sample user and an actual gain value of the sample user, the actual gain value is a gain value which is really generated after the sample user is influenced by a popularization affair, the target characteristic data is the characteristic data of the sample user, and the gain value acquisition device can take the sample characteristic data and the popularization affair characteristic data as input, take the actual gain value as output and train an initial model to obtain the gain model.

The following describes the gain value obtaining method provided in the present application in detail with reference to specific embodiments.

Referring to fig. 1, a schematic flow chart of a gain value obtaining method according to an embodiment of the present application is provided. As shown in fig. 1, the method of the embodiment of the present application may include the following steps S102-S106.

And S102, acquiring user characteristic data and acquiring popularization transaction characteristic data.

Specifically, the gain value obtaining device may obtain past operation data of the user from the historical database, obtain user characteristic data of the user from the operation data of the user, and obtain popularization affair characteristic data, where the popularization affair characteristic data may be set by a relevant worker in the gain value obtaining device, or may be provided by a company or a manufacturer that needs to push popularization affairs.

And S104, acquiring a promotion transaction vector corresponding to the promotion transaction characteristic data based on the embedded layer in the gain model, and acquiring coding characteristic data corresponding to the user characteristic data based on the coding layer in the gain model.

Specifically, the gain value obtaining device may input the promotional transaction feature data and the user feature data into the gain model, the gain model has an embedded layer, the embedded layer may map the high-dimensional data into a low-dimensional vector, the gain value obtaining device may obtain the promotional transaction vector corresponding to the promotional transaction feature data based on the embedded layer in the gain model, the promotional transaction feature data is input into the embedded layer, and the data output by the embedded layer is the promotional transaction vector corresponding to the promotional transaction feature data.

The gain model is provided with a coding layer which can map a natural language sequence into mathematical expression, the gain value acquisition device can acquire coding characteristic data corresponding to user characteristic data based on the coding layer in the gain model, the user characteristic data can be input into the embedding layer, the embedding layer outputs user characteristic vectors corresponding to the user characteristic data, then the user characteristic vectors output by the embedding layer are input into the coding layer, and the data output by the coding layer is the coding characteristic data corresponding to the user characteristic data.

And S106, acquiring a gain value corresponding to the user under the influence of the promotion affairs by adopting a cross layer in the gain model according to the promotion affair vector and the coding characteristic data.

Specifically, a Cross layer exists in the gain model, the Cross layer can perform interactive processing on the promotion transaction vector output by the embedded layer and the coding characteristic data output by the coding layer, so that the influence of promotion transactions on a user is calculated, and a corresponding gain value under the influence of the user promotion transaction is obtained.

In the embodiment of the application, user feature data are obtained, promotion transaction vectors corresponding to the promotion transaction feature data are obtained based on an embedded layer in a gain model, coding feature data corresponding to the user feature data are obtained based on a coding layer in the gain model, a cross layer in the gain model is adopted, and a corresponding gain value of the user under the influence of the promotion transaction is obtained according to the promotion transaction vectors and the coding feature data. The method and the device have the advantages that the gain value of the user under the influence of the promotion affair is obtained by cross fusion of the characteristics of the promotion affair and the characteristics of the user, the influence of the user on the promotion affair is obtained, and the accuracy of pushing the promotion affair is improved.

Please refer to fig. 2, which provides a schematic flow chart of a gain value obtaining method according to an embodiment of the present application. As shown in fig. 2, the method of the embodiment of the present application may include the following steps S201 to S214.

S202, obtaining initial characteristic data of all sample users in at least one sample user, obtaining target characteristic data with information value larger than a value threshold value from the initial characteristic data, and obtaining a target characteristic type corresponding to the target characteristic data.

Specifically, the gain value obtaining device may obtain at least one sample user from the historical database, obtain past operation data of the sample user, and obtain initial feature data of the sample user according to the operation data. It is understood that the initial feature data of the sample user includes feature data corresponding to a plurality of feature types, the gain Value acquiring means may calculate an Information Value (IV) of each of all the initial feature data, and acquire target feature data having an IV Value greater than a Value threshold Value from the initial feature data, the Value threshold Value is used to select the target feature data from the initial feature data, and the Value threshold Value may be an initial setting of the gain Value acquiring means, or may be set and stored by a relevant worker, for example, the Value threshold Value may be 0.95. And then the gain value acquisition device can acquire the target characteristic type corresponding to the target characteristic data, the gain value acquisition device can screen the characteristic data of the user according to the target characteristic type, the information value of the characteristic data corresponding to the target characteristic type of the user is higher, the importance is higher, and the accuracy of the gain model is improved.

Optionally, the at least one sample user may include at least one first sample user and at least one second sample user, the number of the first sample users may be the same as that of the second sample users, and the number of the first sample users may also be different from that of the second sample users. The gain value acquisition device can push and release promotion transactions to a first sample user, but not push promotion transactions to a second sample user, and then acquire actual gain values of the first sample user and the second sample user after a preset time period, wherein the preset time period can be initial setting of the gain value acquisition device, or can be set by related staff, for example, the gain value acquisition device can acquire monthly consumption amounts of the first sample user and the second sample user, monthly consumption amounts of the first sample user and the second sample user are acquired after one month, and a difference value between the monthly consumption amounts and the monthly consumption amount is an actual gain value.

Please refer to fig. 3 together, which provides an exemplary schematic diagram of obtaining an actual gain value for the embodiment of the present application, where if the preset time period is one month, the amount of consumption of the first sample user from 2 months 15 days to 3 months 15 days may be obtained, then a promotion transaction is pushed to the first sample user, and the amount of consumption of the first sample user in the next month, that is, between 3 months 15 days and 4 months 15 days is obtained, and a difference between the amount of consumption in the previous month and the amount of consumption in the next two months is a gain value a, that is, an actual gain value of the first sample user. Similarly, the consumption amount of the second sample user from 2 months 15 days to 3 months 15 days is obtained, but the promotion affair is not pushed to the second sample user, the consumption amount of the second sample user between 3 months 15 days and 4 months 15 days is obtained, and the difference between the consumption amount of the second sample user in the previous month and the consumption amount in the next month is the gain value B, namely the actual gain value of the second sample user.

Optionally, the old user and the new user in the at least one sample user may be distinguished, for example, the sample user whose registration duration of the user account is longer than the preset registration duration is determined as the old user, otherwise, the sample user whose registration duration of the user account is shorter than or equal to the preset registration duration is determined as the new user, and the preset registration duration is used to distinguish the new user and the old user in the sample users, and may be the initial setting of the gain value obtaining apparatus, or may be set by a relevant worker, for example, may be set to three months. The gain value acquisition device can perform different sample processing on the new user and the old user, can perform regression model training on the old user according to the consumption amount of the old user, achieves the purpose of rejecting the abnormally disturbed sample user in the old user, and can perform up-sampling processing on the new user, so that the user category of the new user is more balanced. The gain value acquisition means may train the initial model using at least one sample user after the sample processing.

And S204, acquiring training gain values corresponding to all sample users in the sample characteristic data based on the initial model.

Specifically, the gain value obtaining device may use the sample characteristic data and the popularization transaction characteristic data as inputs, use the actual gain value as an output, and train the initial model to obtain the gain model. After the gain value acquisition device inputs the target characteristic data of each sample user in the sample characteristic data into the initial model, the initial model can output the training gain value corresponding to each sample user, namely the gain value of the initial model under the influence of the popularization affairs, which is predicted by the target characteristic data and the popularization affair characteristic data.

And S206, calculating a loss value based on the training gain value and the actual gain value.

Specifically, the gain value obtaining device may optimize the trained initial model by calculating a loss value between a training gain value predicted by the initial model and an actual gain value. The Loss value may represent the degree of difference between a predicted value and a true value obtained by the model, the Loss value may be calculated by a Loss function from a training gain value and an actual gain value, and the Loss function may be a Mean Square Error function (MSE), an Exponential Loss function (explicit Loss), a perceptual Loss function (Perceptron Loss), or the like.

Optionally, the first sample user is a sample user affected by the promotion transaction, and the second sample user is a sample user not affected by the promotion transaction, it may be understood that, even if the user is not pushed the promotion transaction, the user still has a gain value due to other factors, for example, due to reasons such as a user's salary, needing to add clothes for season change, etc., even if the user is not pushed an advertisement or discount, the consumption amount of the user in the current month may still be greater than the consumption amount in the previous month, that is, there is a gain value. The gain value acquisition device can separately calculate a first loss value corresponding to the first sample user and a second loss value corresponding to the second sample user, so that the initial model can learn the gain value of the user without the influence of promotion affairs in the training process and the optimization process, and the accuracy of the gain model is further improved. The gain value obtaining device may obtain, based on the initial model, a first loss value from the training gain value corresponding to each first sample user and the actual gain value corresponding to each first sample user, and obtain a second loss value from the training gain value corresponding to each second sample user and the actual gain value corresponding to each second sample user.

And S208, optimizing the initial model based on the loss value to obtain a gain model.

Specifically, for each input of a Back Propagation (BP) optimization network, a suitable optimizer is selected according to a loss value to perform reverse optimization on the entire network, and the gain value obtaining device may perform optimization processing on the initial model based on the loss value to obtain a gain model.

Optionally, the gain value obtaining device may perform optimization processing on the initial model according to the first loss value and the second loss value, and it can be understood that the sample users affected by the promotion transaction and the sample users not affected by the promotion transaction are distinguished in the optimization processing, so that the influence of the promotion transaction on the gain value is fully learned, monotonicity of different promotion transactions can be ensured, and accuracy of the gain model is further improved.

Optionally, the initial model may be optimized and iterated multiple times in the training process until the gain model is obtained. An iterative model is obtained after each iteration, the gain value obtaining device may calculate a Correlation Coefficient corresponding to the iterative model, when the Correlation Coefficient of the iterative model is greater than a preset Coefficient, the gain value obtaining device may determine the iterative model as the gain model, the Correlation Coefficient is a Correlation Coefficient between a training gain value obtained by the iterative model and an actual gain value, for example, a Pearson Product-moment Correlation Coefficient (PPMCC) between the training gain value obtained by the iterative model and the actual gain value, the preset Coefficient is used to determine whether the iterative model can be determined as the gain model, the preset Coefficient may be an initial setting of the gain value obtaining device, and may be set by a relevant worker, for example, 0.99.

And S210, acquiring user characteristic data and acquiring popularization transaction characteristic data.

Optionally, the gain value obtaining device may obtain, in the historical database, user feature data corresponding to the user in the target feature type based on the target feature type.

S212, acquiring a promotion transaction vector corresponding to the promotion transaction characteristic data based on the embedded layer in the gain model, and acquiring coding characteristic data corresponding to the user characteristic data based on the coding layer in the gain model.

Specifically, the gain value obtaining device may input the promotional transaction feature data and the user feature data into the gain model, the gain model has an embedded layer, the embedded layer may map high-dimensional data into a low-dimensional vector, the gain value obtaining device may obtain the promotional transaction vector corresponding to the promotional transaction feature data based on the embedded layer in the gain model, the promotional transaction feature data is input into the embedded layer, and data output by the embedded layer is the promotional transaction vector corresponding to the promotional transaction feature data. The gain model is provided with a coding layer which can map a natural language sequence into mathematical expression, the gain value acquisition device can acquire coding characteristic data corresponding to user characteristic data based on the coding layer in the gain model, the user characteristic data can be input into the embedding layer, the embedding layer outputs user characteristic vectors corresponding to the user characteristic data, then the user characteristic vectors output by the embedding layer are input into the coding layer, and the data output by the coding layer is the coding characteristic data corresponding to the user characteristic data.

Optionally, the gain value obtaining device may obtain a sparse feature vector and a dense feature vector in the user feature data, perform coding processing on the sparse feature vector according to a coding layer in the gain model to obtain sparse feature data, perform coding processing on the dense feature vector to obtain dense feature data, for example, perform multiple dimensionality reduction and/or dimensionality enhancement and discarding (Dropout) processing on the dense feature vector through a full connection layer to obtain dense feature data, and then perform deep interaction processing on the sparse feature data and the dense feature data to obtain coding feature data corresponding to the user feature data, where the deep interaction processing may be convolution processing or convolution-like processing, so that hidden layer information between features may be fully mined, and the prediction capability of the gain model is improved.

And S214, acquiring a gain value corresponding to the user under the influence of the promotion affairs by adopting the cross layer in the gain model according to the promotion affair vector and the coding characteristic data.

Specifically, the gain model has a Cross layer, and the Cross layer may perform interactive processing on the promotion transaction vector output by the embedded layer and the coding feature data output by the coding layer, thereby calculating the influence of the promotion transaction on the user, and obtaining a corresponding gain value under the influence of the user re-promotion transaction, for example, the promotion transaction vector and the coding feature data may be subjected to interactive processing through a Cross Attention mechanism (Cross Attention), and then the output of the gain model is obtained through a remolding (resume) function and a full connection layer, that is, the corresponding gain value of the user under the influence of the promotion transaction.

Optionally, after the gain value of the user is predicted through the gain model, the influence of the user on the promotion affairs can be known according to the gain value of the user, and the promotion affairs for the user can be formulated according to the gain value, such as whether to push the promotion affairs to the user, what kind of promotion affairs to push, and the like. The gain value obtaining device may store the gain values of the users in a gain value set, where the gain value set includes gain values corresponding to at least one user, and perform bucket sorting based on the gain values of the users, for example, the users may be sorted from large to small according to the gain values and divided into ten buckets. The gain value obtaining device may perform transaction sensitivity classification processing on at least one user according to the bucket sorting result, and divide the at least one user into a preset number of sensitivity classes, where the preset number may be an initial setting of the gain model, and may also be set by a related worker, for example, the at least one user may be divided into three sensitivity classes: for example, in the bucket sorting, users are divided into ten buckets from large to small according to gain values, the users in the first three buckets can be divided into high-sensitive users, the users in the fourth bucket to the seventh bucket can be divided into medium-sensitive users, and the users in the last three buckets can be divided into low-sensitive users. The transaction sensitivity level grading processing result may be used to formulate a promotion transaction for a user, for example, a small number of promotion transactions may be pushed to a high-sensitivity user, a large number of promotion transactions may be pushed to a medium-sensitivity user, and no promotion transaction may be pushed to a low-sensitivity user.

In the embodiment of the application, the initial feature data of all sample users in at least one sample user is obtained, the target feature data with the information value larger than the value threshold value and the target feature type corresponding to the target feature data are obtained from the initial feature data, and the information value and the importance of the feature data corresponding to the target feature type of the user are higher, so that the accuracy of the gain model is improved. The method comprises the steps of obtaining training gain values corresponding to all sample users based on an initial model, calculating loss values according to the training gain values and actual gain values, optimizing the initial model based on the loss values to obtain a gain model, distinguishing sample users affected by popularization affairs from sample users unaffected by the popularization affairs in optimization, fully learning the influence of the popularization affairs on the gain values, ensuring monotonicity of different popularization affairs, and further improving accuracy of the gain model. Then, user feature data are obtained according to a target feature type, promotion transaction feature data are obtained, promotion transaction vectors corresponding to the promotion transaction feature data are obtained based on an embedded layer in a gain model, coding feature data corresponding to the user feature data are obtained based on a coding layer in the gain model, a cross layer in the gain model is adopted, and a gain value corresponding to the user under the influence of the promotion transaction is obtained according to the promotion transaction vectors and the coding feature data. The method and the device have the advantages that the gain value of the user under the influence of the promotion affair is obtained by cross fusion of the characteristics of the promotion affair and the characteristics of the user, the influence of the user on the promotion affair is obtained, and the accuracy of pushing the promotion affair is improved. And the gain values of the users can be stored in the gain value set, transaction sensitivity grading processing is carried out on at least one user according to bucket sorting, and then the grading processing result according to the sensitivity of the user can be used for formulating promotion transactions aiming at the user, so that the promotion transactions are pushed more accurately and intelligently.

Referring to fig. 4, a schematic structural diagram of an initial model is provided in the present embodiment. It can be understood that the initial model is trained by the sample feature data of at least one sample user and the promotion transaction feature data to obtain the gain model, so that the structures of the initial model and the gain model are the same. An embedding layer, an encoding layer, a crossing layer, a no transaction effect output layer, a loss calculation layer, and an optimization layer may be included in the initial model.

One end of the embedded layer is connected with the coding layer, the other end of the embedded layer is connected with the cross layer, and the coding layer is connected with the cross layer. The embedding layer can map high-dimensional data into low-dimensional vectors, in the training process, the gain value acquisition device inputs popularization affair feature data into the embedding layer, the embedding layer can obtain popularization affair vectors corresponding to the popularization affair feature data, and the popularization affair vectors are transmitted to the crossing layer. The gain value acquisition device can input target feature data corresponding to a sample user into the embedding layer, the embedding layer can obtain target feature vectors corresponding to the target feature data, and the target feature vectors are transmitted to the coding layer. The coding layer can map the natural language sequence into mathematical expression, and the coding layer can obtain target coding feature data corresponding to the target feature data from the target feature vectors.

The coding layer is connected with the cross layer, the coding layer is connected with the transaction-free influence output layer, the cross layer is connected with the transaction-free influence output layer, the transaction-free influence output layer is connected with the transaction-free influence output layer, the coding layer can transmit target coding characteristic data corresponding to the target characteristic data to the cross layer, and can also transmit the target coding characteristic data to the transaction-free influence output layer. The cross layer can perform interactive processing on the promotion transaction vector output by the embedded layer and the target coding characteristic data output by the coding layer, for example, the promotion transaction vector and the target coding characteristic data are performed interactive processing through a cross attention mechanism, then a training gain value of the sample user under the influence of the promotion transaction is obtained through a remodeling function and the full connection layer, so that the influence of the promotion transaction on the user is calculated, and the cross layer can transmit the training gain value of the sample user obtained through calculation to the transaction influence output layer. The transaction-free influence output layer may output a first output value, where the first output value is an output value of the sample user without influence of the promotion transaction, for example, a consumption amount spent by the sample user in a next month when the sample user is not influenced by the promotion transaction. The non-transaction-influence output layer can perform dimension reduction processing on the target coding characteristic data transmitted by the coding layer to obtain a first output value of the sample user under the condition that the popularization transaction is not influenced, and the non-transaction-influence output layer can transmit the first output value to the transaction-influence output layer. The transaction influence output layer may output a second output value, where the second output value is an output value of the sample user under the influence of the promotion transaction, for example, a consumption amount spent by the sample user in a next month under the influence of the promotion transaction such as advertisement and discount offer. The transaction influence output layer can add the first output value transmitted by the non-transaction influence output layer and the training gain value of the sample user transmitted by the cross layer to obtain a second output value of the sample user under the influence of the promotion transaction.

The transaction-free influence output layer is connected with the loss calculation layer, the transaction influence output layer is connected with the loss calculation layer, and the loss calculation layer is connected with the optimization layer. The loss calculation layer may calculate a loss value between a training gain value predicted by the initial model and an actual gain value, where the loss value may represent a degree of difference between a predicted value and a true value obtained by the model, the loss value may be calculated from the training gain value and the actual gain value through a loss function, and the loss function may be a mean square error function, an exponential loss function, a perceptual loss function, or the like. The loss calculation layer can transmit the obtained loss value to the optimization layer, the optimization layer can optimize the initial model through the loss value, and the reverse propagation optimization network selects a proper optimizer according to the loss value for each input to reversely optimize the whole network, so that the gain model is obtained.

Optionally, the first sample user and the second sample user are included in the sample users, the first sample user is a sample user affected by the promotion transaction, and the second sample user is a sample user not affected by the promotion transaction. The loss calculation layer can separately calculate a first loss value corresponding to the first sample user and a second loss value corresponding to the second sample user, so that the initial model can learn the gain value of the user without the influence of promotion transactions in the training process and the optimization process, and the accuracy of the gain model is further improved. The loss calculation layer can acquire a first loss value through a training gain value corresponding to each first sample user and an actual gain value corresponding to each first sample user, acquire a second loss value through a training gain value corresponding to each second sample user and an actual gain value corresponding to each second sample user, transmit the first loss value and the second loss value to the optimization layer, and the optimization layer can optimize the initial model according to the first loss value and the second loss value respectively, distinguish sample users affected by promotion affairs from sample users not affected by the promotion affairs in optimization, fully learn the influence of the promotion affairs on the gain values, ensure monotonicity of different promotion affairs, and further improve accuracy of the gain model.

Optionally, the initial model may be optimized and iterated for multiple times in the training process until the gain model is obtained. The gain value obtaining device can calculate a correlation coefficient corresponding to the iterative model, when the correlation coefficient of the iterative model is greater than a preset coefficient, the gain value obtaining device can determine the iterative model as the gain model, the correlation coefficient is a correlation coefficient between a training gain value and an actual gain value obtained by the iterative model, for example, a pearson product moment correlation coefficient between the training gain value and the actual gain value obtained by the iterative model, the preset coefficient is used for judging whether the iterative model can be determined as the gain model, the preset coefficient can be initial setting of the gain value obtaining device, and can be set by related workers, for example, the preset coefficient can be 0.99.

In the embodiment of the application, the initial feature data of all sample users in at least one sample user is obtained, the target feature data with the information value larger than the value threshold value and the target feature type corresponding to the target feature data are obtained from the initial feature data, and the information value and the importance of the feature data corresponding to the target feature type of the user are higher, so that the accuracy of the gain model is improved. The method comprises the steps of obtaining training gain values corresponding to all sample users based on an initial model, calculating loss values according to the training gain values and actual gain values, optimizing the initial model based on the loss values to obtain a gain model, distinguishing sample users affected by popularization affairs from sample users unaffected by the popularization affairs in optimization, fully learning the influence of the popularization affairs on the gain values, ensuring monotonicity of different popularization affairs, and further improving accuracy of the gain model. The method comprises the steps of obtaining user characteristic data, obtaining popularization affair vectors corresponding to the popularization affair characteristic data based on an embedded layer in a gain model, obtaining coding characteristic data corresponding to the user characteristic data based on a coding layer in the gain model, adopting a cross layer in the gain model, and obtaining gain values corresponding to users under the influence of popularization affairs according to the popularization affair vectors and the coding characteristic data. The method and the device have the advantages that the gain value of the user under the influence of the promotion affair is obtained by cross fusion of the characteristics of the promotion affair and the characteristics of the user, the influence of the user on the promotion affair is obtained, and the accuracy of pushing the promotion affair is improved.

The gain value obtaining apparatus provided in the embodiments of the present application will be described in detail below with reference to fig. 5 to 6. It should be noted that, the gain value obtaining apparatus in fig. 5-fig. 6 is used for executing the method of the embodiment shown in fig. 1 and fig. 2 of the present application, for convenience of description, only the portion related to the embodiment of the present application is shown, and details of the technology are not disclosed, please refer to the embodiment shown in fig. 1 and fig. 2 of the present application.

Please refer to fig. 5, which shows a schematic structural diagram of a gain value obtaining apparatus according to an exemplary embodiment of the present application. The gain value acquisition means may be implemented as all or part of the apparatus by software, hardware or a combination of both. The apparatus 1 comprises a feature data acquisition module 11, an embedding module 12 and a gain value acquisition module 13.

The characteristic data acquisition module 11 is used for acquiring user characteristic data and acquiring popularization transaction characteristic data;

the embedding module 12 is configured to obtain a promotion transaction vector corresponding to the promotion transaction feature data based on an embedding layer in a gain model, and obtain coding feature data corresponding to the user feature data based on a coding layer in the gain model;

and a gain value obtaining module 13, configured to use the cross layer in the gain model, and obtain, according to the promotion transaction vector and the coding feature data, a gain value corresponding to the user under the influence of the promotion transaction.

In this embodiment, user feature data and popularization transaction feature data are obtained, a popularization transaction vector corresponding to the popularization transaction feature data is obtained based on an embedded layer in a gain model, coding feature data corresponding to the user feature data is obtained based on a coding layer in the gain model, a cross layer in the gain model is adopted, and a gain value corresponding to the user under the influence of the popularization transaction is obtained according to the popularization transaction vector and the coding feature data. The method and the device have the advantages that the gain value of the user under the influence of the popularization affair is obtained by cross fusion of the characteristics of the popularization affair and the characteristics of the user, the influence of the user on the popularization affair is obtained, and the accuracy of pushing the popularization affair is improved.

Please refer to fig. 6, which shows a schematic structural diagram of a gain value obtaining apparatus according to an exemplary embodiment of the present application. The gain value acquisition means may be implemented as all or part of an apparatus by software, hardware or a combination of both. The device 1 comprises a sample data acquisition module 14, a training processing module 15, a characteristic data acquisition module 11, an embedding module 12, a gain value acquisition module 13 and a grading processing module 16.

A sample data obtaining module 14, configured to obtain initial feature data of all sample users in at least one sample user;

and acquiring target characteristic data with the information value larger than a value threshold value from the initial characteristic data, and acquiring a target characteristic type corresponding to the target characteristic data.

A training processing module 15, configured to perform training processing on the initial model based on sample feature data and promotional transaction feature data to obtain a gain model, where the sample feature data includes target feature data corresponding to each sample user of at least one sample user, and an actual gain value corresponding to each sample user;

the at least one sample user comprises at least one first sample user and at least one second sample user, the first sample user is a sample user affected by the promotion transaction, and the second sample user is a sample user not affected by the promotion transaction.

Optionally, the training processing module 15 is specifically configured to obtain, based on an initial model, a training gain value corresponding to each sample user in the sample feature data;

calculating a loss value based on the training gain value and the actual gain value;

and optimizing the initial model based on the loss value to obtain a gain model.

Optionally, the training processing module 15 is specifically configured to, based on the initial model, obtain a first loss value according to the training gain value corresponding to each first sample user and the actual gain value corresponding to each first sample user;

based on the initial model, acquiring a second loss value according to the training gain value corresponding to each second sample user and the actual gain value corresponding to each second sample user;

optimizing the initial model based on the loss value to obtain a gain model, including:

and optimizing the initial model based on the first loss value and the second loss value to obtain a gain model.

The characteristic data acquisition module 11 is used for acquiring user characteristic data and acquiring promotion transaction characteristic data;

optionally, the feature data obtaining module 11 is specifically configured to obtain, based on the target feature type, user feature data corresponding to the user from a historical database.

The embedding module 12 is configured to obtain a popularization transaction vector corresponding to the popularization transaction feature data based on an embedding layer in a gain model, and obtain coding feature data corresponding to the user feature data based on a coding layer in the gain model;

optionally, the embedding module 12 is specifically configured to obtain a sparse feature vector and a dense feature vector of the user feature data based on an embedding layer in the gain model;

based on a coding layer in the gain model, coding the sparse feature vector to obtain sparse feature data, and coding the dense feature vector to obtain dense feature data;

and performing deep interaction processing on the sparse feature data and the dense feature data to obtain coding feature data corresponding to the user feature data.

And a gain value obtaining module 13, configured to use a cross layer in the gain model, and obtain, according to the promotion transaction vector and the coding feature data, a gain value corresponding to the user under the influence of the promotion transaction.

A hierarchical processing module 16, configured to store the gain values of the users in a set of gain values, where the set of gain values includes a gain value corresponding to each user in at least one user;

performing bucket sorting based on the gain value of each user;

and performing transaction sensitivity grading processing on the at least one user according to the bucket sorting result, wherein the transaction sensitivity grading processing result is used for formulating popularization transactions aiming at the user.

In the embodiment of the application, the initial feature data of all sample users in at least one sample user is obtained, the target feature data with the information value larger than the value threshold value and the target feature type corresponding to the target feature data are obtained from the initial feature data, and the information value and the importance of the feature data corresponding to the target feature type of the user are higher, so that the accuracy of the gain model is improved. The method comprises the steps of obtaining training gain values corresponding to all sample users based on an initial model, calculating loss values according to the training gain values and actual gain values, optimizing the initial model based on the loss values to obtain a gain model, distinguishing sample users affected by popularization affairs from sample users unaffected by the popularization affairs in optimization, fully learning the influence of the popularization affairs on the gain values, ensuring monotonicity of different popularization affairs, and further improving accuracy of the gain model. Then, user feature data are obtained according to a target feature type, promotion transaction feature data are obtained, promotion transaction vectors corresponding to the promotion transaction feature data are obtained based on an embedded layer in a gain model, coding feature data corresponding to the user feature data are obtained based on a coding layer in the gain model, a cross layer in the gain model is adopted, and a gain value corresponding to the user under the influence of the promotion transactions is obtained according to the promotion transaction vectors and the coding feature data. The method and the device have the advantages that the gain value of the user under the influence of the promotion affair is obtained by cross fusion of the characteristics of the promotion affair and the characteristics of the user, the influence of the user on the promotion affair is obtained, and the accuracy of pushing the promotion affair is improved. And the gain values of the users can be stored in the gain value set, transaction sensitivity grading processing is carried out on at least one user according to bucket sorting, and then the grading processing result according to the sensitivity of the user can be used for formulating promotion transactions aiming at the user, so that the promotion transactions are pushed more accurately and intelligently.

It should be noted that, when the gain value obtaining apparatus provided in the foregoing embodiment executes the gain value obtaining method, only the division of the above functional modules is taken as an example, and in practical applications, the above functions may be distributed to different functional modules according to needs, that is, the internal structure of the device may be divided into different functional modules to complete all or part of the above described functions. In addition, the gain value obtaining device and the gain value obtaining method provided by the above embodiments belong to the same concept, and details of implementation processes thereof are referred to in the method embodiments and are not described herein again.

The above-mentioned serial numbers of the embodiments of the present application are merely for description and do not represent the merits of the embodiments.

An embodiment of the present application further provides a computer storage medium, where the computer storage medium may store a plurality of instructions, and the instructions are suitable for being loaded by a processor and being executed by the method for obtaining a gain value according to the embodiment shown in fig. 1 to fig. 4, and a specific execution process may refer to specific descriptions of the embodiment shown in fig. 1 to fig. 4, which is not described herein again.

The present application further provides a computer program product, where at least one instruction is stored, and the at least one instruction is loaded by the processor and executes the gain value obtaining method according to the embodiment shown in fig. 1 to 4, where a specific execution process may refer to specific descriptions of the embodiment shown in fig. 1 to 4, and is not described herein again.

Referring to fig. 7, a block diagram of an electronic device according to an exemplary embodiment of the present application is shown. The electronic device in the present application may comprise one or more of the following components: a processor 110, a memory 120, an input device 130, an output device 140, and a bus 150. The processor 110, memory 120, input device 130, and output device 140 may be coupled by a bus 150.

Processor 110 may include one or more processing cores. The processor 110 connects various parts within the overall electronic device using various interfaces and lines, performs various functions of the terminal 100 and processes data by executing or executing instructions, programs, code sets, or instruction sets stored in the memory 120 and calling data stored in the memory 120. Alternatively, the processor 110 may be implemented in hardware using at least one of Digital Signal Processing (DSP), field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor 110 may integrate one or a combination of a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), a modem, and the like. The CPU mainly processes an operating system, a user page, an application program and the like; the GPU is used for rendering and drawing display content; the modem is used to handle wireless communications. It is understood that the modem may not be integrated into the processor 110, but may be implemented by a communication chip.

The Memory 120 may include a Random Access Memory (RAM) or a Read-Only Memory (ROM). Optionally, the memory 120 includes a Non-Transitory Computer-Readable Medium (Non-transient Computer-Readable Storage Medium). The memory 120 may be used to store instructions, programs, code, sets of codes, or sets of instructions. The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playing function, an image playing function, and the like), instructions for implementing the above method embodiments, and the like, and the operating system may be an Android (Android) system, including a system based on Android system depth development, an IOS system developed by apple, including a system based on IOS system depth development, or other systems.

The memory 120 may be divided into an operating system space, where an operating system runs, and a user space, where native and third-party applications run. In order to ensure that different third-party application programs can achieve a better operation effect, the operating system allocates corresponding system resources for the different third-party application programs. However, the requirements of different application scenarios in the same third-party application program on system resources are different, for example, in a local resource loading scenario, the third-party application program has a higher requirement on the disk reading speed; in the animation rendering scene, the third-party application program has a high requirement on the performance of the GPU. The operating system and the third-party application program are independent from each other, and the operating system cannot sense the current application scene of the third-party application program in time, so that the operating system cannot perform targeted system resource adaptation according to the specific application scene of the third-party application program.

In order to enable the operating system to distinguish a specific application scenario of the third-party application program, data communication between the third-party application program and the operating system needs to be opened, so that the operating system can acquire current scenario information of the third-party application program at any time, and further perform targeted system resource adaptation based on the current scenario.

The input device 130 is used for receiving input instructions or data, and the input device 130 includes, but is not limited to, a keyboard, a mouse, a camera, a microphone, or a touch device. The output device 140 is used for outputting instructions or data, and the output device 140 includes, but is not limited to, a display device, a speaker, and the like. In one example, the input device 130 and the output device 140 may be combined, and the input device 130 and the output device 140 are touch display screens.

The touch display screen can be designed as a full-face screen, a curved screen or a profiled screen. The touch display screen can also be designed to be a combination of a full-face screen and a curved-face screen, and a combination of a special-shaped screen and a curved-face screen, which is not limited in the embodiment of the present application.

In addition, those skilled in the art will appreciate that the configurations of the electronic devices illustrated in the above-described figures are not meant to be limiting, and that the electronic devices may include more or fewer components than those shown, or some components may be combined, or different arrangements of components may be used. For example, the electronic device further includes a radio frequency circuit, an input unit, a sensor, an audio circuit, a Wireless Fidelity (WiFi) module, a power supply, a bluetooth module, and other components, which are not described herein again.

In the electronic device shown in fig. 7, the processor 110 may be configured to invoke the gain value obtaining application program stored in the memory 120, and specifically perform the following operations:

In one embodiment, the processor 110 further performs the following operations before performing the steps of obtaining the user characteristic data and obtaining the promotional transaction characteristic data:

training the initial model based on sample characteristic data and promotion transaction characteristic data to obtain a gain model, wherein the sample characteristic data comprises target characteristic data corresponding to each sample user in at least one sample user and an actual gain value corresponding to each sample user;

In one embodiment, the processor 110 further performs the following operations before performing the training process on the initial model based on the sample feature data and the generalized transaction feature data to obtain the gain model:

acquiring initial characteristic data of all sample users in at least one sample user;

and acquiring target characteristic data with information value larger than a value threshold value from the initial characteristic data, and acquiring a target characteristic type corresponding to the target characteristic data.

In an embodiment, when the processor 110 performs a training process on the initial model based on the sample data set and the promoted transaction feature data set to obtain the gain model, the following operations are specifically performed:

based on an initial model, acquiring a training gain value corresponding to each sample user in the sample characteristic data;

In one embodiment, the processor 110 when executing the calculating the loss value based on the training gain value and the actual gain value specifically performs the following operations:

based on the initial model, acquiring a first loss value according to the training gain value corresponding to each first sample user and the actual gain value corresponding to each first sample user;

In an embodiment, the processor 110 specifically performs the following operations when executing the step of acquiring the user feature data:

and acquiring user characteristic data corresponding to the user from a historical database based on the target characteristic type.

In an embodiment, when the processor 110 performs the following operation to obtain the coding feature data corresponding to the user feature data based on the coding layer in the gain model:

acquiring sparse feature vectors and dense feature vectors of the user feature data based on an embedded layer in the gain model;

In one embodiment, after the processor 110 executes the cross layer in the gain model and obtains the gain value corresponding to the user under the influence of the promotion transaction according to the promotion transaction vector and the coding feature data, the following operations are further executed:

storing the gain values of the users in a gain value set, wherein the gain value set comprises gain values corresponding to all users in at least one user;

performing bucket sorting based on the gain value of each user;

and performing transaction sensitivity grading processing on the at least one user according to the bucket sorting result, wherein the transaction sensitivity grading processing result is used for formulating a promotion transaction for the user.

In the embodiment, the initial feature data of all sample users in at least one sample user is obtained, the target feature data with the information value larger than the value threshold value and the target feature type corresponding to the target feature data are obtained from the initial feature data, and the information value and the importance of the feature data corresponding to the target feature type of the user are higher, so that the accuracy of the gain model is improved. The method comprises the steps of obtaining training gain values corresponding to all sample users based on an initial model, calculating loss values according to the training gain values and actual gain values, optimizing the initial model based on the loss values to obtain a gain model, distinguishing sample users affected by popularization affairs from sample users unaffected by the popularization affairs in optimization, fully learning the influence of the popularization affairs on the gain values, ensuring monotonicity of different popularization affairs, and further improving accuracy of the gain model. Then, user feature data are obtained according to a target feature type, promotion transaction feature data are obtained, promotion transaction vectors corresponding to the promotion transaction feature data are obtained based on an embedded layer in a gain model, coding feature data corresponding to the user feature data are obtained based on a coding layer in the gain model, a cross layer in the gain model is adopted, and a gain value corresponding to the user under the influence of the promotion transactions is obtained according to the promotion transaction vectors and the coding feature data. The method and the device have the advantages that the gain value of the user under the influence of the promotion affair is obtained by cross fusion of the characteristics of the promotion affair and the characteristics of the user, the influence of the user on the promotion affair is obtained, and the accuracy of pushing the promotion affair is improved. And the gain value of the user can be stored in the gain value set, transaction sensitivity grading processing is carried out on at least one user according to the bucket sorting, and then the grading processing result according to the sensitivity of the user can be used for formulating promotion transactions aiming at the user, so that the promotion transactions are pushed more accurately and intelligently.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium can be a magnetic disk, an optical disk, a read-only memory or a random access memory.

The above disclosure is only for the purpose of illustrating the preferred embodiments of the present application and should not be taken as limiting the scope of the present application, so that the present application will be covered by the appended claims.

Claims

1. A method for obtaining a gain value, the method comprising:

acquiring user characteristic data and acquiring promotion transaction characteristic data;

2. The method of claim 1, wherein before obtaining the user characteristic data and the promotional transaction characteristic data, further comprising:

3. The method of claim 2, wherein before training the initial model based on the sample feature data and the generalized transaction feature data to obtain the gain model, the method further comprises:

4. The method of claim 2, wherein the training of the initial model based on the sample data set and the promotional transaction feature data set to obtain a gain model comprises:

5. The method of claim 4, wherein calculating a loss value based on the training gain value and the actual gain value comprises:

6. The method of claim 3, wherein the obtaining user characteristic data comprises:

7. The method according to claim 1, wherein said obtaining the coding characteristic data corresponding to the user characteristic data based on the coding layer in the gain model comprises:

8. The method according to claim 1, wherein after the obtaining, by using the cross layer in the gain model and according to the promotion transaction vector and the coding feature data, a gain value corresponding to the user under the influence of the promotion transaction, further comprises:

performing bucket sorting based on the gain value of each user;

9. A gain value acquisition apparatus, characterized in that the apparatus comprises:

10. A computer storage medium, characterized in that it stores a plurality of instructions adapted to be loaded by a processor and to perform the method steps according to any one of claims 1 to 8.

11. A computer program product having stored thereon a plurality of instructions adapted to be loaded by a processor and to carry out the method steps according to any of claims 1 to 8.

12. An electronic device, comprising: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to perform the method steps of any of claims 1 to 8.