Extended customer data generation method and device, equipment, medium and product

Abstract

The embodiment of the invention discloses an extended customer data generation method and device, equipment, a medium and a product, and relates to the field of artificial intelligence, and the method comprises the steps: generating synthetic customer reason data through a first target generator, the synthetic customer reason data comprising a synthetic customer identifier and synthetic customer behavior data corresponding to the synthetic customer identifier; the synthesized customer reason data are input into a second target generator, synthesized customer result data matched with the synthesized customer reason data are obtained, and the synthesized result data comprise synthesized customer preference data; combining the synthesized customer reason data with the synthesized customer result data to obtain extended customer data; wherein the first target generator is a generator in a pre-trained reason data generative adversarial network, and the second target generator is a generator in a pre-trained result data generative adversarial network, so that the internal connectivity of the extended customer data is increased, and the authenticity of the extended customer data is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, equipment, medium and product for generating extended customer data. Background Art

[0002] With the rapid development of information technology, the application scope of data mining technology and artificial intelligence technology is becoming more and more extensive. At present, data mining technology is usually used to mine huge amounts of customer data, and artificial intelligence technology is used to recommend financial products that are more in line with customer expectations based on the data mining results. Therefore, the amount of customer data has become an important factor affecting the accuracy of the recommendation results.

[0003] In the prior art, the extended customer data is usually directly generated by a generative adversarial network. However, the internal connectivity of the extended customer data generated by the prior art is not strong, resulting in the possibility of inconsistency in the extended customer data. Summary of the invention

[0004] The present invention provides a method, device, equipment, medium and product for generating extended customer data, which increases the internal connection of the extended customer data and improves the authenticity of the extended customer data.

[0005] In a first aspect, an embodiment of the present invention provides a method for generating extended customer data, comprising:

[0006] Generate synthetic customer reason data by a first target generator, the synthetic customer reason data including: a synthetic customer identifier and synthetic customer behavior data corresponding to the synthetic customer identifier;

[0007] Inputting the synthesized customer reason data into the second target generator to obtain synthesized customer result data matching the synthesized customer reason data, the synthesized customer result data including the synthesized customer preference data;

[0008] combining the synthesized customer reason data with the synthesized customer result data to obtain extended customer data;

[0009] Among them, the first target generator is a generator in a pre-trained cause data generation adversarial network, and the second target generator is a generator in a pre-trained result data generation adversarial network.

[0010] Optionally, before generating synthetic customer reason data through the first target generator, it also includes: forming a reason data generation adversarial network with the first generator model and the first discriminator model, and constructing a first generation loss function corresponding to the first generator model, and a first discriminant loss function corresponding to the first discriminator model; fixing the parameters of the first discriminator model, and iteratively optimizing the parameters of the first generator model through the first generation loss function until the number of optimizations of the first generator model is equal to the set first number threshold, and obtaining the first generator model after one optimization; fixing the parameters of the first generator model after one optimization, and iteratively optimizing the parameters of the first discriminator model through the first discriminant loss function. Iterative optimization is performed until the optimization times of the first discriminator model is equal to the set second times threshold, and the first discriminator model after optimization is obtained; wherein, the ratio between the first times threshold and the second times threshold is the first update ratio, and the first update ratio is greater than 1; the first alternating training times of the first generator model and the first discriminator model are increased by one; if the first alternating training times is less than the set first alternating training times threshold, then the operation of fixing the parameters of the first discriminator model and iteratively optimizing the parameters of the first generator model through the first generation loss function is returned until the first alternating training times is equal to the set first alternating training times threshold, and the first target generator model is obtained.

[0011] Optionally, the parameters of the first generator model are iteratively optimized through the first generation loss function until the number of optimizations of the first generator model is equal to the set first number threshold, and the first generator model after one optimization is obtained, including: inputting random noise data into the first generator model for processing to obtain simulated customer reason data, and inputting the simulated customer reason data and the real customer reason data into the first discriminator model for discrimination to obtain a first discrimination result; back-propagating the first discrimination result to the first generator model through the first generation loss function, optimizing the parameters of the first generator model, and increasing the number of optimizations of the first generator model by one; if the number of optimizations of the first generator model is less than the set first number threshold, returning to execute the operation of inputting random noise data into the first generator model for processing until the number of optimizations of the first generator model is equal to the set first number threshold, and the first generator model after one optimization is obtained.

[0012] Optionally, before inputting the synthetic customer reason data into the second target generator to obtain the synthetic customer result data matching the synthetic customer reason data, it also includes: forming a result data generation adversarial network with the second generator model and the second discriminator model, and constructing a second generation loss function corresponding to the second generator model; fixing the parameters of the second discriminator model, and iteratively optimizing the parameters of the second generator model through the second generation loss function until the number of optimizations of the second generator model is equal to the set third number threshold, to obtain an optimized second generator model; wherein the second generation loss function includes a true and false loss function and a causal loss function, the true and false loss function is used to reflect the similarity between the synthetic customer result data and the real result data, and the causal loss function is used to reflect the similarity between the synthetic customer result data and the simulated customer reason data the degree of causal relationship between them; fixing the parameters of the second generator model after one optimization, iteratively optimizing the parameters of the second discriminator model until the number of optimizations of the second discriminator model is equal to the set fourth number threshold, and obtaining the second discriminator model after one optimization; wherein, the ratio between the third number threshold and the fourth number threshold is the second update ratio, and the second update ratio is greater than 1; adding one to the second alternating training times of the second generator model and the second discriminator model; if the second alternating training times is less than the set second alternating training times threshold, returning to execute the operation of fixing the parameters of the second discriminator model, iteratively optimizing the parameters of the second generator model through the second generation loss function, until the second alternating training times is equal to the set second alternating training times threshold, and obtaining the second target generator model.

[0013] Optionally, the parameters of the second generator model are iteratively optimized through a second generation loss function until the number of optimizations of the second generator model is equal to a set third number threshold, thereby obtaining a second generator model that has been optimized once, including: inputting random noise data and real customer reason data into the second generator model for processing to obtain simulated customer result data, and inputting the real customer reason data, the real customer result data, and the simulated customer result data into the second discriminator model for discrimination to obtain a second discrimination result; back-propagating the second discrimination result to the second generator model through the second generation loss function, iteratively optimizing the parameters of the second generator model, and increasing the number of optimizations of the second generator model by one; if the number of optimizations of the second generator model is less than the set third number threshold, returning to execute the operation of inputting random noise data and real customer reason data into the second generator model for processing until the number of optimizations of the second generator model is equal to the set third number threshold, thereby obtaining a second generator model that has been optimized once.

[0014] Optionally, constructing a second generation loss function corresponding to the second generator model includes: constructing a true and false loss function based on the expected amount of generated data and the probability that the second discriminator model identifies the simulated customer result data as real customer result data; calculating the first Pearson correlation coefficient between each column of result data of the real customer result data and the corresponding column of cause data in the real customer reason data; calculating the second Pearson correlation coefficient between each column of result data of the simulated customer result data and the corresponding column of cause data in the real customer reason data; constructing a causal loss function based on the number of columns of the real customer result data, the number of columns of the real customer reason data, the first Pearson correlation coefficient and the second Pearson correlation coefficient; and performing weighted summation of the true and false loss function and the causal loss function to obtain the second generation loss function.

[0015] In a second aspect, an embodiment of the present invention further provides a device for generating extended customer data, including:

[0016] A reason data generating module, configured to generate synthetic customer reason data through a first target generator, wherein the synthetic customer reason data includes: a synthetic customer identifier and synthetic customer behavior data corresponding to the synthetic customer identifier;

[0017] A result data generating module, used for inputting the synthesized customer reason data into the second target generator to obtain synthesized customer result data matching the synthesized customer reason data, wherein the synthesized customer result data includes the synthesized customer preference data;

[0018] An extended data generation module, used for combining the synthesized customer reason data with the synthesized customer result data to obtain extended customer data;

[0019] Among them, the first target generator is a generator in a pre-trained cause data generation adversarial network, and the second target generator is a generator in a pre-trained result data generation adversarial network.

[0020] In a third aspect, an embodiment of the present invention further provides an electronic device, the electronic device comprising:

[0021] at least one processor; and

[0022] a memory communicatively connected to at least one processor; wherein,

[0023] The memory stores a computer program that can be executed by at least one processor. The computer program is executed by at least one processor so that the at least one processor can execute the method for generating extended customer data provided by any embodiment of the present invention.

[0024] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a processor to implement the method for generating extended customer data provided by any embodiment of the present invention when executed.

[0025] In a fifth aspect, an embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the method for generating extended customer data provided by any embodiment of the present invention.

[0026] The technical solution of the embodiment of the present invention generates synthetic customer reason data through a first target generator, generates synthetic customer result data according to the synthetic customer reason data through a second target generator, and combines the synthetic customer reason data with the synthetic customer result data to obtain extended customer data, thereby avoiding the situation where the extended customer data generated by the prior art has weak internal connectivity, resulting in inconsistencies in the extended customer data (for example, in the same extended customer data, the customer's consumption habits and the customer's financial management preferences are inconsistent), increasing the internal connectivity of the extended customer data, and improving the authenticity of the extended customer data, so that the extended customer data can be more put into use in real scenarios.

[0027] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0029] Figure 1 is a flow chart of a method for generating extended customer data provided according to Embodiment 1 of the present invention;

[0030] Figure 2 is a flowchart of another method for generating extended customer data provided in Embodiment 2 of the present invention;

[0031] Figure 3 is a schematic diagram of the structure of a device for generating extended customer data provided according to Embodiment 3 of the present invention;

[0032] Figure 4 It is a schematic diagram of the structure of an electronic device provided by Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0035] Embodiment 1

[0036] Figure 1 1 is a flowchart of a method for generating extended customer data according to Embodiment 1 of the present invention. This embodiment is applicable to the case where customer data is to be expanded. The method can be executed by a device for generating extended customer data. The device for generating extended customer data can be implemented in the form of hardware and / or software. The device for generating extended customer data can be configured in an electronic device such as a computer.

[0037] like Figure 1 As shown, a method for generating extended customer data disclosed in this embodiment includes:

[0038] S110 . Generate synthetic customer reason data through a first target generator, where the synthetic customer reason data includes: a synthetic customer identifier and synthetic customer behavior data corresponding to the synthetic customer identifier.

[0039] In this embodiment, the synthetic customer reason data may be data obtained by simulating the real customer reason data through the first target generator. The synthetic customer identifier may be used to uniquely identify a new customer corresponding to the synthetic customer behavior data. The synthetic customer behavior data may be used to reflect the characteristic information of the customer, such as the customer address and the customer consumption habits.

[0040] In this step, specifically, a pre-constructed variable that obeys a normal distribution can be sampled to obtain random noise data, and the random noise data can be input into a pre-trained first target generator for processing to obtain at least one new customer and synthetic customer reason data corresponding to each new customer. The first target generator is a generator in a pre-trained reason data generation adversarial network.

[0041] S120, inputting the synthesized customer reason data into the second target generator to obtain synthesized customer result data matching the synthesized customer reason data, wherein the synthesized customer result data includes the synthesized customer preference data.

[0042] In this embodiment, the synthetic customer result data may be data obtained by simulating the real customer result data that has a causal relationship with the real customer cause data through the second target generator. The synthetic customer result data has a causal relationship with the synthetic customer cause data. The synthetic customer preference data may be used to reflect various preferences of newly added customers, such as financial management preferences.

[0043] In this step, specifically, the random noise data and the synthetic customer cause data can be input into the second target generator for processing to obtain the synthetic customer result data that has a causal relationship with the synthetic customer cause data. The second target generator is a generator in a pre-trained result data generation adversarial network. The second target generator is used to generate synthetic customer result data corresponding to the synthetic customer cause data according to the causal relationship between the customer cause data and the customer result data after receiving the synthetic customer cause data.

[0044] Exemplarily, assuming that the synthetic customer reason data generated by the first goal generator includes customer consumption habits, the second goal generator can generate the financial management preferences of new customers based on the causal relationship between customer consumption habits and financial management preferences.

[0045] S130 , combining the synthesized customer reason data with the synthesized customer result data to obtain extended customer data.

[0046] In this step, specifically, the synthesized customer reason data and the synthesized customer result data belonging to the same newly added customer may be combined to obtain extended customer data corresponding to each newly added customer, wherein the extended customer data may be table data including multiple columns.

[0047] The technical solution of this embodiment is to generate synthetic customer reason data through a first target generator, and the synthetic customer reason data includes: a synthetic customer identification and synthetic customer behavior data corresponding to the synthetic customer identification; the synthetic customer reason data is input into a second target generator to obtain synthetic customer result data matching the synthetic customer reason data, and the synthetic customer result data includes synthetic customer preference data; the synthetic customer reason data is combined with the synthetic customer result data to obtain extended customer data; wherein the first target generator is a generator in a pre-trained cause data generation adversarial network, and the second target generator is a generator in a pre-trained result data generation adversarial network. The technical means solves the problem that the internal connectivity of the extended customer data generated by the prior art is not strong, resulting in inconsistencies in the extended customer data, increases the internal connectivity of the extended customer data, and improves the authenticity of the extended customer data.

[0048] Embodiment 2

[0049] Figure 2 1 is a flow chart of another method for generating extended customer data provided according to Embodiment 2 of the present invention. This embodiment is a further optimization and extension based on the above embodiment.

[0050] like Figure 2 As shown, a method for generating extended customer data disclosed in this embodiment includes:

[0051] S210. The first generator model and the first discriminator model are combined into a cause data generation adversarial network, and a first generation loss function corresponding to the first generator model and a first discriminant loss function corresponding to the first discriminator model are constructed.

[0052] In this step, specifically, after constructing a value function corresponding to the cause data generative adversarial network, a first generation loss function corresponding to the first generator model and a first discriminant loss function corresponding to the first discriminator model can be constructed according to the value function.

[0053] Exemplarily, the value function corresponding to the cause data generation adversarial network can be expressed by the following formula:

[0054]

[0055] The first generation loss function corresponding to the first generator model constructed according to the value function is:

[0056]

[0057] Where G(z) represents the simulated customer reason data generated by the first generator model, z represents random noise, D(G(z)) represents the probability that the first discriminator module recognizes the simulated customer reason data as true, z~Pz (z) represents sampling from random noise data.

[0058] The first discriminant loss function corresponding to the first discriminator model constructed according to the value function is:

[0059]

[0060] Where D(x) represents the probability that the first discriminator model recognizes the real customer reason data as true, D(G(z)) represents the probability that the first discriminator module recognizes the simulated customer reason data as true, x~P data (x) represents sampling from real customer reason data, z~P z (z) represents sampling from random noise data input to the first generator model.

[0061] S220. Fix the parameters of the first discriminator model, and iteratively optimize the parameters of the first generator model through the first generation loss function until the number of optimizations of the first generator model is equal to the set first number threshold, so as to obtain the first optimized generator model.

[0062] In this step, specifically, after optimizing the parameters of the first generator model through the first generation loss function, it can be determined whether the number of optimizations of the first generator model is equal to the set first number threshold. If so, it can be considered that the optimization of the first generator model is completed in stages, and the first generator model after optimization is obtained. If not, return to perform the operation of optimizing the parameters of the first generator model through the first generation loss function until the number of optimizations of the first generator model is equal to the set first number threshold. Among them, the first number threshold can be set according to user needs, such as 3.

[0063] Optionally, the parameters of the first generator model are iteratively optimized by the first generation loss function until the number of optimizations of the first generator model is equal to the set first number threshold, and the first generator model after optimization is obtained, including: inputting random noise data into the first generator model for processing to obtain simulated customer reason data, and inputting the simulated customer reason data and the real customer reason data into the first discriminator model for discrimination to obtain a first discrimination result; back-propagating the first discrimination result to the first generator model by the first generation loss function, optimizing the parameters of the first generator model, and adding one to the number of optimizations of the first generator model; if the number of optimizations of the first generator model is less than the set first number threshold, returning to execute the operation of inputting random noise data into the first generator model for processing until the number of optimizations of the first generator model is equal to the set first number threshold, and obtaining the first generator model after optimization. Wherein, the real customer reason data includes: a real customer identifier and real customer behavior data corresponding to the real customer identifier. The real customer identifier can be used to uniquely identify the real customer, and the real customer can be a customer in a certain area or a customer who sent a login request within a certain time period. The simulated customer reason data may be data obtained by simulating real customer reason data through the first generator model.

[0064] Specifically, when the number of optimization times of the first generator model is less than the set first number threshold, the first generator model can be used to generate simulated customer reason data similar to the real customer reason data based on random noise data. Then, the simulated customer reason data and the real customer reason data can be discriminated by the first discriminator model to obtain a first discrimination result. Afterwards, the first discrimination result can be back-propagated to the first generator model through the first generation loss function, the parameters of the first generator model can be optimized, and the number of optimization times of the first generator model can be increased by one. Finally, it can be determined whether the number of optimization times of the first generator model is less than the set first number threshold. If so, it returns to execute the operation of inputting the random noise data into the first generator model for processing. Otherwise, the first generator model after one optimization is obtained.

[0065] The advantage of this setting is that, through the first generation loss function and the first discrimination result, the parameters of the first generator model are iteratively optimized to obtain the first optimized generator model, which can improve the similarity between the simulated customer reason data generated by the first generator model and the real customer reason data, thereby improving the authenticity of the extended customer data.

[0066] S230. Fix the parameters of the first generator model after the first optimization, and iteratively optimize the parameters of the first discriminator model through the first discriminant loss function until the number of optimizations of the first discriminator model is equal to the set second number threshold, so as to obtain the first discriminator model after the first optimization.

[0067] In this step, specifically, after optimizing the parameters of the first discriminant model through the first discriminant loss function, it can be determined whether the number of optimizations of the first discriminator model is equal to the set second number threshold. If so, it can be considered that the optimization of the first discriminator model is completed in stages, and the first discriminator model after optimization is obtained; if not, return to execute the operation of optimizing the parameters of the first discriminator model through the first discriminant loss function until the number of optimizations of the first discriminator model is equal to the set second number threshold. Among them, the ratio between the first number threshold and the second number threshold is the first update ratio. The first update ratio is greater than 1, that is, the first number threshold is greater than the second number threshold. The first update ratio can be set according to user needs.

[0068] Exemplarily, assuming that the first update ratio is 3, it can be determined that the first count threshold is 3 and the second count threshold is 1. At this time, the parameters of the first discriminator model can be fixed, and the parameters of the first generator model can be iteratively optimized through the first generation loss function until the number of optimizations of the first generator model is equal to 3, and the first generator model after optimization is obtained. After obtaining the first generator model after optimization, the parameters of the first generator model after optimization can be fixed, and the parameters of the first discriminant loss function can be iteratively optimized until the number of optimizations of the first discriminator model is equal to 1, and the first discriminant model after optimization is obtained.

[0069] The advantage of this setting is that since the update speed of the generator model is slow and the update speed of the discriminator model is fast, by setting the ratio between the first number threshold and the second number threshold to be greater than 1, that is, setting the optimization times of the first generator model to be greater than the optimization times of the first discriminator model, the training stability of the cause data generation adversarial network can be improved.

[0070] S240, adding one to the first alternating training times of the first generator model and the first discriminator model.

[0071] In this embodiment, the first number of alternating training times may be the number of times the first generator model and the first discriminator model are optimized as a whole, that is, the number of times the cause data generation adversarial network is optimized.

[0072] S250, determining whether the first alternating training times is equal to the set first alternating training times threshold, if so, executing S260, if not, returning to executing S220.

[0073] In this embodiment, the first target generator model can be used to generate synthetic customer reason data. In this step, specifically, it can be determined whether the first alternating training times is less than the set first alternating training times threshold. If so, the operation of fixing the parameters of the first discriminator model and iteratively optimizing the parameters of the first generator model through the first generation loss function is returned. If not, it can be considered that the first generator model training is completed, the first target generator model is obtained, and the synthetic customer reason data is generated through the first target generator model.

[0074] The advantage of this setting is that by alternately training the first generator model and the first discriminator model, the generation capability of the first generator model can be continuously improved with the help of the first discriminator model, so that the synthetic customer reason data generated by the first generator model becomes more and more realistic, thereby improving the authenticity of the extended customer data.

[0075] S260: Generate synthetic customer reason data through a first target generator, where the synthetic customer reason data includes: a synthetic customer identifier and synthetic customer behavior data corresponding to the synthetic customer identifier.

[0076] S270: Input the synthesized customer reason data into the second target generator to obtain synthesized customer result data matching the synthesized customer reason data, wherein the synthesized customer result data includes the synthesized customer preference data.

[0077] Optionally, before inputting the synthetic customer reason data into the second target generator to obtain the synthetic customer result data matching the synthetic customer reason data, it also includes: forming a result data generation adversarial network with the second generator model and the second discriminator model, and constructing a second generation loss function corresponding to the second generator model; fixing the parameters of the second discriminator model, and iteratively optimizing the parameters of the second generator model through the second generation loss function until the number of optimizations of the second generator model is equal to the set third number threshold, to obtain an optimized second generator model; wherein the second generation loss function includes a true and false loss function and a causal loss function, the true and false loss function is used to reflect the similarity between the synthetic customer result data and the real result data, and the causal loss function is used to reflect the similarity between the synthetic customer result data and the simulated customer reason data the degree of causal relationship between them; fixing the parameters of the second generator model after one optimization, iteratively optimizing the parameters of the second discriminator model until the number of optimizations of the second discriminator model is equal to the set fourth number threshold, and obtaining the second discriminator model after one optimization; wherein, the ratio between the third number threshold and the fourth number threshold is the second update ratio, and the second update ratio is greater than 1; adding one to the second alternating training times of the second generator model and the second discriminator model; if the second alternating training times is less than the set second alternating training times threshold, returning to execute the operation of fixing the parameters of the second discriminator model, iteratively optimizing the parameters of the second generator model through the second generation loss function, until the second alternating training times is equal to the set second alternating training times threshold, and obtaining the second target generator model.

[0078] Specifically, after determining the second generation loss function according to the true and false loss function and the causal loss function, the parameters of the second generator model can be optimized by the second generation loss function, and it is determined whether the number of optimizations of the second generator model is equal to the set third number threshold. If the number of optimizations of the second generator model is equal to the set third number threshold, it can be considered that the optimization of the second generator model is completed in stages, and the second generator model after one optimization is obtained. If the number of optimizations of the second generator model is less than the set third number threshold, the operation of optimizing the parameters of the second generator model by the second generation loss function is returned until the number of optimizations of the second generator model is equal to the set third number threshold. When the number of optimizations of the second generator model is equal to the set third number threshold, the parameters of the second generator model after one optimization are fixed, and the parameters of the second discriminator model are iteratively optimized until the number of optimizations of the second discriminator model is equal to the set fourth number threshold, and the second discriminator model after one optimization is obtained. Among them, the ratio between the third number threshold and the fourth number threshold is the second update ratio. The second update ratio is greater than 1, that is, the third number threshold is greater than the fourth number threshold. The advantage of this setting is that by alternately training the second generator model and the second discriminator model, the generation capability of the second generator model can be continuously improved with the help of the second discriminator model, so that the synthetic customer result data generated by the second generator model becomes more and more realistic, thereby improving the authenticity of the extended customer data. Secondly, since the update speed of the generator model is slow and the update speed of the discriminator model is fast, the training stability of the result data generation adversarial network can be improved by setting the ratio between the third number threshold and the fourth number threshold to be greater than 1, that is, setting the optimization number of the second generator model to be greater than the optimization number of the second discriminator model.

[0079] Further, the parameters of the second generator model are iteratively optimized by the second generation loss function until the number of optimizations of the second generator model is equal to the set third number threshold, and the second generator model after one optimization is obtained, including: inputting random noise data and real customer reason data into the second generator model for processing to obtain simulated customer result data, and inputting real customer reason data, real customer result data and simulated customer result data into the second discriminator model for discrimination to obtain a second discrimination result; back-propagating the second discrimination result to the second generator model by the second generation loss function, iteratively optimizing the parameters of the second generator model, and adding one to the number of optimizations of the second generator model; if the number of optimizations of the second generator model is less than the set third number threshold, returning to execute the operation of inputting random noise data and real customer reason data into the second generator model for processing until the number of optimizations of the second generator model is equal to the set third number threshold, and obtaining the second generator model after one optimization. Wherein, the simulated customer result data can be data obtained by simulating the real customer result data by the second generator model.

[0080] Specifically, after iteratively optimizing the parameters of the second generator model and increasing the number of optimizations of the second generator model by one, it is determined whether the number of optimizations of the second generator model is less than the set third number threshold. If the number of optimizations of the second generator model is less than the set third number threshold, the operation of inputting random noise data and real customer reason data into the second generator model for processing is returned. If the number of optimizations of the second generator model is equal to the set third number threshold, it can be considered that the optimization of the second generator model is completed in stages, and a second generator model after one optimization is obtained.

[0081] The advantage of such a setting is that by inputting the real customer reason data, the real customer result data and the simulated customer result data into the second discriminator model, the second discriminator model can judge the real situation of the simulated customer result data. By feeding back the causal relationship between the customer reason data and the customer result data, as well as the real situation of the simulated customer result data to the second generator model, the second generator model can more accurately generate simulated customer result data that has a causal relationship with the simulated customer reason and is more similar to the real customer result data.

[0082] Optionally, constructing a second generation loss function corresponding to the second generator model includes: constructing a true-false loss function according to the expected amount of generated data and the probability that the second discriminator model identifies the simulated customer result data as real customer result data; calculating the first Pearson correlation coefficient between each column of result data of the real customer result data and the corresponding column of reason data in the real customer reason data; calculating the second Pearson correlation coefficient between each column of result data of the simulated customer result data and the corresponding column of reason data in the real customer reason data; determining the causal loss function according to the number of columns of the real customer result data, the number of columns of the real customer reason data, the first Pearson correlation coefficient and the second Pearson correlation coefficient; performing weighted summation of the true-false loss function and the causal loss function to obtain the second generation loss function. Wherein, the number of columns of the real customer result data is the same as the number of columns of the simulated customer result data, and the number of columns of the real customer reason data is the same as the number of columns of the simulated customer reason data.

[0083] Specifically, as far as true and false loss is concerned, the worse the discrimination effect of the second discriminator model on the simulated customer result data generated by the second generator model is, the smaller the true and false loss of the second generator model is, and the more similar the simulated customer result data is to the real customer result data, the true and false loss function can be determined based on the probability that each simulated customer result data is identified as real customer result data by the second discriminator model, and the expected amount of generated data.

[0084] Since for causal loss, the smaller the causal loss, the more similar the causal relationship between the simulated customer result data and the real customer cause data is to the causal relationship between the real customer result data and the real customer cause data, the first Pearson correlation coefficient between each column of the real customer result data and the corresponding column of the real customer cause data, as well as the second Pearson correlation coefficient between each column of the simulated customer result data and the corresponding column of the real customer cause data can be calculated, and the causality between the simulated customer result data and the real customer cause data can be fed back and adjusted by back propagation of the sum of the differences between multiple columns of Pearson correlation coefficients.

[0085] Exemplarily, the true and false loss function can be expressed by the following specific formula:

[0086]

[0087] Among them, lossa is the true and false loss, Q is the number of simulated customer result data generated in each batch during the result data generation adversarial network training process, that is, the expected amount of generated data, is the probability that the second discriminator model identifies the i-th simulated customer result data as the real customer result data.

[0088] The causal loss function is expressed by the following specific calculation formula:

[0089]

[0090] Wherein, lossb is the causal loss, N is the number of columns of the real customer result data, M is the number of columns of the real customer reason data, cij is the first Pearson correlation coefficient between the result data in the i-th column of the real customer result data and the reason data in the j-th column of the real customer reason data; is the second Pearson correlation coefficient between the i-th column result data in the simulated customer result data and the j-th column reason data in the real customer reason data.

[0091] Finally, the second generation loss function can be obtained by weighted summing the true and false loss function and the causal loss function, wherein the weight coefficients of the true and false loss function and the causal loss function can be adjusted according to historical experience.

[0092] The advantage of this setting is that by generating the second generation loss function through the true and false loss function and the causal loss function, and iteratively optimizing the parameters of the second generator model through the second generation loss function, the second generator model can generate synthetic customer result data that is more similar to the real customer result data and has a causal relationship with the synthetic customer cause data.

[0093] S280: Combine the synthesized customer reason data with the synthesized customer result data to obtain extended customer data.

[0094] Among them, the first target generator is a generator in a pre-trained cause data generation adversarial network, and the second target generator is a generator in a pre-trained result data generation adversarial network.

[0095] Optionally, after generating the extended customer data, the model that uses only the real customer data as training samples can be used as a control group, and the model that uses both the real customer data and the extended customer data as training samples can be used as an experimental group, and then the next experiment can be carried out. During the experiment, the use effect of the experimental group model can be compared with the use effect of the control group model to obtain the degree of usability of the extended customer data in downstream tasks.

[0096] The technical solution of this embodiment, by alternately training the generator model and the discriminator model, can continuously improve the generation ability of the generator model with the help of the discriminator model, so that the data generated by the generator model becomes more and more realistic, thereby improving the authenticity of the extended customer data. Secondly, since the update speed of the generator model is slow and the update speed of the discriminator model is fast, the training stability of the cause data generation adversarial network can be improved by setting the ratio between the first number threshold and the second number threshold to be greater than 1, that is, setting the optimization number of the first generator model to be greater than the optimization number of the first discriminator model. Finally, by setting the ratio between the third number threshold and the fourth number threshold to be greater than 1, that is, setting the optimization number of the second generator model to be greater than the optimization number of the second discriminator model, the training stability of the result data generation adversarial network can be improved.

[0097] It should be noted that the customer-related information collected by the present invention is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0098] Embodiment 3

[0099] Figure 3 1 is a schematic diagram of the structure of an extended customer data generation device provided according to Embodiment 3 of the present invention. This embodiment is applicable to the case where customer data is to be expanded. The extended customer data generation device can be implemented in the form of hardware and / or software and can be configured in an electronic device such as a computer.

[0100] like Figure 3 As shown, the device for generating extended customer data disclosed in this embodiment includes: a cause data generating module 31, a result data generating module 32 and an extended data generating module 33, wherein:

[0101] The reason data generating module 31 is used to generate synthetic customer reason data through the first target generator, and the synthetic customer reason data includes: a synthetic customer identifier and synthetic customer behavior data corresponding to the synthetic customer identifier;

[0102] A result data generating module 32, for inputting the synthesized customer reason data into the second target generator to obtain synthesized customer result data matching the synthesized customer reason data, the synthesized customer result data including the synthesized customer preference data;

[0103] An extended data generating module 33, used to combine the synthesized customer reason data with the synthesized customer result data to obtain extended customer data;

[0104] Among them, the first target generator is a generator in a pre-trained cause data generation adversarial network, and the second target generator is a generator in a pre-trained result data generation adversarial network.

[0105] The technical solution in this embodiment solves the problem that the internal connectivity of the extended customer data generated by the prior art is not strong, resulting in inconsistencies in the extended customer data, through the mutual cooperation of the cause data generation module, the result data generation module and the extended data generation module, increases the internal connectivity of the extended customer data, and improves the authenticity of the extended customer data.

[0106] Optionally, the device also includes a first generator training module, which is used to: combine the first generator model and the first discriminator model to form a causal data generation adversarial network, and construct a first generation loss function corresponding to the first generator model, and a first discriminant loss function corresponding to the first discriminator model; fix the parameters of the first discriminator model, and iteratively optimize the parameters of the first generator model through the first generation loss function until the number of optimizations of the first generator model is equal to a set first number threshold, so as to obtain the first generator model after optimization; fix the parameters of the first generator model after optimization, and iteratively optimize the parameters of the first discriminator model through the first discriminant loss function. until the optimization times of the first discriminator model is equal to the set second times threshold, and the first discriminator model after one optimization is obtained; wherein, the ratio between the first times threshold and the second times threshold is the first update ratio, and the first update ratio is greater than 1; the first alternating training times of the first generator model and the first discriminator model are increased by one; if the first alternating training times is less than the set first alternating training times threshold, then return to execute the operation of fixing the parameters of the first discriminator model, and iteratively optimizing the parameters of the first generator model through the first generation loss function, until the first alternating training times is equal to the set first alternating training times threshold, and the first target generator model is obtained.

[0107] Optionally, when the first generator training module iteratively optimizes the parameters of the first generator model through the first generation loss function until the number of optimizations of the first generator model is equal to the set first number threshold, and the first generator model is obtained after one optimization, it is specifically used to: input random noise data into the first generator model for processing to obtain simulated customer reason data, and input the simulated customer reason data and the real customer reason data into the first discriminator model for discrimination to obtain a first discrimination result; back-propagate the first discrimination result to the first generator model through the first generation loss function, optimize the parameters of the first generator model, and increase the number of optimizations of the first generator model by one; if the number of optimizations of the first generator model is less than the set first number threshold, return to execute the operation of inputting random noise data into the first generator model for processing until the number of optimizations of the first generator model is equal to the set first number threshold, and the first generator model is obtained after one optimization.

[0108] Optionally, the device also includes a second generator training module, which is used to: combine the second generator model and the second discriminator model into a result data generation adversarial network, and construct a second generation loss function corresponding to the second generator model; fix the parameters of the second discriminator model, and iteratively optimize the parameters of the second generator model through the second generation loss function until the number of optimization times of the second generator model is equal to the set third number threshold, so as to obtain the second generator model after one optimization; wherein the second generation loss function includes a true and false loss function and a causal loss function, the true and false loss function is used to reflect the similarity between the synthetic customer result data and the real result data, and the causal loss function is used to reflect the degree of causal correlation between the synthetic customer result data and the simulated customer cause data; the one-time optimization The parameters of the optimized second generator model are fixed, and the parameters of the second discriminator model are iteratively optimized until the number of optimizations of the second discriminator model is equal to the set fourth number threshold, so as to obtain the second discriminator model after one optimization; wherein, the ratio between the third number threshold and the fourth number threshold is the second update ratio, and the second update ratio is greater than 1; the second alternating training times of the second generator model and the second discriminator model are increased by one; if the second alternating training times is less than the set second alternating training times threshold, then return to execute the operation of fixing the parameters of the second discriminator model, and iteratively optimizing the parameters of the second generator model through the second generation loss function until the second alternating training times is equal to the set second alternating training times threshold, so as to obtain the second target generator model.

[0109] Optionally, when the second generator training module iteratively optimizes the parameters of the second generator model through the second generation loss function until the number of optimizations of the second generator model is equal to the set third number threshold, and the second generator model after one optimization is obtained, it is specifically used to: input random noise data and real customer reason data into the second generator model for processing to obtain simulated customer result data, and input the real customer reason data, the real customer result data and the simulated customer result data into the second discriminator model for discrimination to obtain a second discrimination result; back propagate the second discrimination result to the second generator model through the second generation loss function, iteratively optimize the parameters of the second generator model, and increase the number of optimizations of the second generator model by one; if the number of optimizations of the second generator model is less than the set third number threshold, return to execute the operation of inputting random noise data and real customer reason data into the second generator model for processing until the number of optimizations of the second generator model is equal to the set third number threshold, and the second generator model after one optimization is obtained.

[0110] Optionally, when constructing the second generation loss function corresponding to the second generator model, the second generator training module is specifically used to: construct a true and false loss function based on the expected amount of generated data and the probability that the second discriminator model identifies the simulated customer result data as real customer result data; calculate the first Pearson correlation coefficient between each column of result data of the real customer result data and the corresponding column of cause data in the real customer reason data; calculate the second Pearson correlation coefficient between each column of result data of the simulated customer result data and the corresponding column of cause data in the real customer reason data; construct a causal loss function based on the number of columns of the real customer result data, the number of columns of the real customer reason data, the first Pearson correlation coefficient and the second Pearson correlation coefficient; and perform weighted summation of the true and false loss function and the causal loss function to obtain the second generation loss function.

[0111] The device for generating extended customer data provided in the embodiment of the present invention can execute the method for generating extended customer data provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method. For the contents not described in detail in this embodiment, please refer to the description in any method embodiment of the present application.

[0112] Embodiment 4

[0113] Figure 4 FIG. 1 is a schematic diagram showing the structure of an electronic device 10 that can be used to implement an embodiment of the present invention. Figure 4As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0114] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0115] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a method for generating extended customer data.

[0116] In some embodiments, the method for generating extended customer data may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method for generating extended customer data described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform the method for generating extended customer data in any other appropriate manner (e.g., by means of firmware).

[0117] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0118] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0119] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0120] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0121] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0122] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0123] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0124] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for generating extended customer data, characterized in that: The method comprises: Generate synthetic customer reason data by a first target generator, the synthetic customer reason data comprising: a synthetic customer identifier and synthetic customer behavior data corresponding to the synthetic customer identifier; Inputting the synthesized customer reason data into a second target generator to obtain synthesized customer result data matching the synthesized customer reason data, wherein the synthesized customer result data includes synthesized customer preference data; combining the synthesized customer reason data with the synthesized customer result data to obtain extended customer data; Among them, the first target generator is a generator in a pre-trained cause data generation adversarial network, and the second target generator is a generator in a pre-trained result data generation adversarial network.

2. The method according to claim 1, characterized in that Before generating the synthetic customer reason data by the first target generator, it also includes: The first generator model and the first discriminator model form a cause data generation adversarial network, and construct a first generation loss function corresponding to the first generator model and a first discriminant loss function corresponding to the first discriminator model; The parameters of the first discriminator model are fixed, and the parameters of the first generator model are iteratively optimized by using the first generation loss function until the number of optimizations of the first generator model is equal to the set first number threshold, thereby obtaining the first generator model after one optimization; The parameters of the first generator model after the optimization are fixed, and the parameters of the first discriminator model are iteratively optimized by the first discriminant loss function until the number of optimizations of the first discriminator model is equal to the set second number threshold, thereby obtaining the first discriminator model after the optimization; The ratio between the first number threshold and the second number threshold is the first update ratio, and the first update ratio is greater than 1; Adding one to the first alternating training times of the first generator model and the first discriminator model; If the first number of alternating training times is less than the set first alternating training times threshold, the operation returns to fix the parameters of the first discriminator model, and iteratively optimizes the parameters of the first generator model through the first generation loss function until the first number of alternating training times is equal to the set first alternating training times threshold, thereby obtaining the first target generator model.

3. The method according to claim 2, characterized in that Iteratively optimizing the parameters of the first generator model through the first generation loss function until the number of optimizations of the first generator model is equal to the set first number threshold, and obtaining the first generator model after one optimization, including: Inputting the random noise data into the first generator model for processing to obtain simulated customer reason data, and inputting the simulated customer reason data and the real customer reason data into the first discriminator model for discrimination to obtain a first discrimination result; Back-propagating the first discrimination result to the first generator model through the first generation loss function, optimizing the parameters of the first generator model, and increasing the number of optimization times of the first generator model by one; If the optimization times of the first generator model are less than the set first times threshold, the operation of inputting random noise data into the first generator model for processing is returned until the optimization times of the first generator model is equal to the set first times threshold, thereby obtaining the first generator model after one optimization.

4. The method according to claim 1, characterized in that: Before inputting the synthesized customer reason data into the second target generator to obtain the synthesized customer result data matching the synthesized customer reason data, the method further includes: The second generator model and the second discriminator model form a result data generation adversarial network, and construct a second generation loss function corresponding to the second generator model; The parameters of the second discriminator model are fixed, and the parameters of the second generator model are iteratively optimized through the second generation loss function until the number of optimizations of the second generator model is equal to the set third number threshold, thereby obtaining the second generator model after one optimization; The second generated loss function includes a true-false loss function and a causal loss function, wherein the true-false loss function is used to reflect the similarity between the synthetic customer result data and the real result data, and the causal loss function is used to reflect the degree of causal association between the synthetic customer result data and the simulated customer cause data; The parameters of the first optimized second generator model are fixed, and the parameters of the second discriminator model are iteratively optimized until the number of optimizations of the second discriminator model is equal to a set fourth number threshold, thereby obtaining the first optimized second discriminator model; The ratio between the third number threshold and the fourth number threshold is the second update ratio, and the second update ratio is greater than 1; Add one to the second alternating training times of the second generator model and the second discriminator model; If the second number of alternating training times is less than the set second alternating training times threshold, return to execute the operation of fixing the parameters of the second discriminator model, and iteratively optimizing the parameters of the second generator model through the second generation loss function until the second number of alternating training times is equal to the set second alternating training times threshold, so as to obtain the second target generator model.

5. The method according to claim 4, characterized in that The parameters of the second generator model are iteratively optimized through the second generation loss function until the number of optimizations of the second generator model is equal to the set third number threshold, and the second generator model after one optimization is obtained, including: Inputting the random noise data and the real customer reason data into the second generator model for processing to obtain simulated customer result data, and inputting the real customer reason data, the real customer result data and the simulated customer result data into the second discriminator model for discrimination to obtain a second discrimination result; Back-propagating the second discrimination result to the second generator model through the second generation loss function, iteratively optimizing the parameters of the second generator model, and increasing the number of optimization times of the second generator model by one; If the optimization times of the second generator model are less than the set third number threshold, the operation of inputting random noise data and real customer reason data into the second generator model for processing is returned until the optimization times of the second generator model are equal to the set third number threshold, thereby obtaining the second generator model after one optimization.

6. The method according to claim 4, characterized in that Constructing a second generation loss function corresponding to the second generator model, including: Constructing a true-false loss function according to the estimated amount of generated data and the probability that the second discriminator model identifies the simulated customer result data as real customer result data; Calculate the first Pearson correlation coefficient between each column of result data in the real customer result data and the corresponding column of reason data in the real customer reason data; Calculate the second Pearson correlation coefficient between each column of result data in the simulated customer result data and the corresponding column of reason data in the real customer reason data; Construct a causal loss function based on the number of columns of real customer result data, the number of columns of real customer cause data, the first Pearson correlation coefficient and the second Pearson correlation coefficient; The true and false loss function and the causal loss function are weighted summed to obtain the second generation loss function.

7. A device for generating extended customer data, characterized in that: The device comprises: A reason data generating module, configured to generate synthetic customer reason data through a first target generator, wherein the synthetic customer reason data includes: a synthetic customer identifier and synthetic customer behavior data corresponding to the synthetic customer identifier; A result data generating module, used for inputting the synthesized customer reason data into a second target generator to obtain synthesized customer result data matching the synthesized customer reason data, wherein the synthesized customer result data includes synthesized customer preference data; An extended data generating module, used for combining the synthesized customer reason data with the synthesized customer result data to obtain extended customer data; Among them, the first target generator is a generator in a pre-trained cause data generation adversarial network, and the second target generator is a generator in a pre-trained result data generation adversarial network.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for generating extended customer data according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for generating extended customer data according to any one of claims 1 to 6 when the instructions are executed.

10. A computer program product, characterized in that The computer program product comprises a computer program, which, when executed by a processor, implements the method for generating extended customer data according to any one of claims 1 to 6.

Images