Model training method, prediction probability information generation method, device and electronic equipment

Abstract

Embodiments of the present disclosure disclose a model training method, a prediction probability information generation method, an apparatus and an electronic device. A specific implementation of the method comprises: obtaining a training data set; inputting a variable data set into an initial feature data vector representation model to output an initial variable vector set; performing model training on each initial discriminant model in an initial discriminant model set included in an initial target value payment method probability generation model to obtain a discriminant model set; and performing model training on an initial generative model set and the initial feature data vector representation model included in the trained target value payment method probability generation model according to the initial variable vector set, an actual value reduction information set and an actual value payment method information set to obtain the target value payment method probability generation model. The implementation is related to artificial intelligence, and the trained target value payment method probability generation model can accurately generate target value payment method prediction probability information.

Description

Technical Field

[0001] The embodiments of this disclosure relate to the field of computer technology, specifically to model training methods, prediction probability information generation methods, apparatus, and electronic devices. Background Technology

[0002] Currently, in the context of goods value transfer, certain intervention methods (such as value push) can significantly influence users' decision-making behavior. The common approach to generating the probability of a target value payment method is to use causal inference methods (DeR-CFR, Decomposed Representations for Conterfactual Regression) to determine the predicted probability information of the target value payment method.

[0003] However, the inventors discovered that the following technical problems often arise when using the above method:

[0004] The problem of insufficient accuracy in representing individual causal effects leads to insufficient accuracy in predicting the probability of the generated target value payment method.

[0005] The information disclosed in this background section is only intended to enhance the understanding of the background of the inventive concept, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0007] Some embodiments of this disclosure propose model training methods, prediction probability information generation methods, apparatuses, and electronic devices to address the technical problems mentioned in the background section above.

[0008] In a first aspect, some embodiments of this disclosure provide a model training method, comprising: acquiring a training dataset, wherein the training data in the training dataset includes: variable data, actual value reduction information, and actual value payment method information; inputting the variable dataset into an initial feature data vector representation model included in an initial target value payment method probability generation model to output an initial variable vector set; and training each initial discriminant model in the initial discriminant model set included in the initial discriminant model set according to the initial variable vector set, the actual value reduction information set, and the actual value payment method information set to obtain a discriminant model set, wherein the discriminant model set includes: a first value reduction information generation model, a first value payment method probability... Generative model and second value payment method probability generation model; fixing the model parameters of the first value reduction information generation model, the first value payment method probability generation model and the second value payment method probability generation model, and training the initial generative model set and the initial feature data vector representation model included in the trained target value payment method probability generation model according to the initial variable vector set, the actual value reduction information set and the actual value payment method information set, to obtain the target value payment method probability generation model, wherein the generative model set in the target value payment method probability generation model includes: third value payment method probability generation model, fourth value payment method probability generation model and second value reduction information generation model.

[0009] Optionally, the aforementioned variable data includes: instrumental variable data, adjustment variable data, and confounding variable data; and the aforementioned input of the variable dataset into the initial feature data vector representation model included in the initial target value expenditure mode probability generation model to output an initial variable vector set includes: inputting the instrumental variable dataset, adjustment variable dataset, and confounding variable dataset into the aforementioned initial feature data vector representation model to generate an initial instrumental variable vector set, an initial adjustment variable vector set, and an initial confounding variable vector set; and performing vector set fusion on the aforementioned initial instrumental variable vector set, the aforementioned initial adjustment variable vector set, and the aforementioned initial confounding variable vector set to obtain the initial variable vector set.

[0010] Optionally, the above-mentioned training of each initial discriminant model in the initial discriminant model set included in the initial target value payment method probability generation model, based on the initial variable vector set, the actual value reduction information set, and the actual value payment method information set, to obtain a discriminant model set, includes: selecting an initial adjustment variable vector from the initial adjustment variable vector set as the first initial adjustment variable vector, and performing the following first training step: inputting the first initial adjustment variable vector into the first initial value reduction information generation model included in the initial target value payment method probability generation model to generate first initial value reduction information; determining the actual value reduction information corresponding to the first initial adjustment variable vector as the first actual value reduction information. The process involves: determining first loss information for the aforementioned first initial value reduction information and the aforementioned first actual value reduction information; in response to determining that the aforementioned first loss information is greater than a first value, determining the aforementioned first initial value reduction information generation model as the first value reduction information generation model; in response to determining that the aforementioned first loss information is less than or equal to the aforementioned first value, updating the model parameters of the aforementioned first initial value reduction information generation model based on the aforementioned first loss information to obtain an updated first value reduction information generation model, and reselecting a first initial adjustment variable vector from the aforementioned initial adjustment variable vector set, determining the aforementioned updated first value reduction information generation model as the first initial value reduction information generation model, and continuing to execute the aforementioned first training step.

[0011] Optionally, the above-mentioned training of each initial discriminant model in the initial discriminant model set included in the initial target value payment method probability generation model based on the initial variable vector set, the actual value reduction information set, and the actual value payment method information set to obtain a discriminant model set includes: selecting an initial confusion variable vector from the initial confusion variable vector set as a first initial confusion variable vector, and performing the following second training steps: determining the actual value reduction information corresponding to the first initial confusion variable vector as second actual value reduction information, and determining the initial instrumental variable vector corresponding to the first initial confusion variable vector as a first initial instrumental variable vector; generating a first actual value reduction vector for the second actual value reduction information; inputting the first initial confusion variable vector and the first actual value reduction vector into the first initial value payment method probability generation model to generate first initial value payment method probability information; inputting the first initial instrumental variable vector, the first initial confusion variable vector, and the first actual value reduction vector into the second initial value payment method probability generation model to generate second initial value payment. Method probability information; generating second loss information for the aforementioned first initial value expenditure method probability information and the aforementioned second initial value expenditure method probability information; in response to determining that the aforementioned second loss information is greater than a second value, determining the aforementioned first initial value expenditure method probability generation model as the first value expenditure method probability generation model, and determining the aforementioned second initial value expenditure method probability generation model as the second value expenditure method probability generation model; in response to determining that the aforementioned second loss information is less than or equal to the aforementioned second value, updating the model parameters of the aforementioned first initial value expenditure method probability generation model and the aforementioned second initial value expenditure method probability generation model according to the aforementioned second loss information, obtaining the updated first value expenditure method probability generation model and the updated first value expenditure method probability generation model, and reselecting the first initial confusion variable vector from the aforementioned initial confusion variable vector set, using the aforementioned updated first value expenditure method probability generation model as the first initial value expenditure method probability generation model, and using the aforementioned updated second value expenditure method probability generation model as the second initial value expenditure method probability generation model, to continue executing the aforementioned second training step.

[0012] Optionally, the above-mentioned training of the target value payment method probability generation model, including the initial generative model set and the initial feature data vector representation model, to obtain the target value payment method probability generation model includes: selecting an initial confounding variable vector from the initial confounding variable vector set as the second initial confounding variable vector, and performing the following third training step: determining the initial adjustment variable vector, initial instrumental variable vector, corresponding actual value reduction vector, and corresponding actual value reduction information corresponding to the second initial confounding variable vector, respectively as the second initial adjustment variable vector, the second initial instrumental variable vector, the third actual value reduction vector, and the third actual value reduction information; inputting the second initial confounding variable vector, the third actual value reduction vector, and the second initial adjustment variable vector into the third initial value payment method probability generation model to generate the third initial value payment method probability information; and inputting the second initial adjustment variable vector into the fourth initial value payment method probability generation model to generate the fourth initial value payment. Method probability information; inputting the aforementioned second initial instrumental variable vector into the second initial value reduction information generation model to generate second initial value reduction information; generating third loss information for the aforementioned third initial value payment method probability information and corresponding actual value payment method information; generating fourth loss information for the aforementioned fourth initial value payment method probability information and corresponding actual value payment method information; generating fifth loss information for the aforementioned second initial value reduction information and the aforementioned third actual value reduction information; in response to determining that the aforementioned third loss information is less than a third value, the aforementioned fourth loss information is less than a fourth value, and the aforementioned fifth loss information is less than a fifth value, the aforementioned third initial value payment method probability generation model, the aforementioned fourth initial value payment method probability generation model, the aforementioned second initial value reduction information generation model, and the aforementioned initial feature data vector representation model are respectively determined as the third value payment method probability generation model, the fourth value payment method probability generation model, the second value reduction information generation model, and the feature data vector representation model.

[0013] Optionally, the above-mentioned target value payment method probability generation model is validated through the following steps: obtaining a user information set; for each value reduction value among the multiple value reduction values ​​included in the value reduction information, performing the following information generation steps: determining the value distribution value corresponding to each user information in the above-mentioned user information set; selecting a subset of user information whose corresponding value distribution value is the target value from the above-mentioned user information set as the first user information subset; selecting a subset of user information whose corresponding value distribution value is the above-mentioned value reduction value from the above-mentioned user information set as the second user information subset; merging the above-mentioned first user information subset and the above-mentioned second user information subset to obtain a merged user information set; for each merged user information in the above-mentioned merged user information set, using the above-mentioned target value payment method probability generation model, determining the prediction probability of the first target value payment method of the above-mentioned merged user information under the above-mentioned value reduction value. The system generates probability information for predicting the second target value payment method under the target value, based on the rate information and the aforementioned fused user information. It then groups each fused user information in the aforementioned fused user information set according to the information difference between the first target value payment method prediction probability information and the second target value payment method prediction probability information corresponding to each fused user information, obtaining at least one user information group. For each user information group in the aforementioned at least one user information group, it determines the predicted average gain value corresponding to the aforementioned user information group based on the information difference set corresponding to the aforementioned user information group. It then determines the actual average gain value for each user information group, obtaining at least one actual average gain value. Finally, it compares the obtained at least one predicted average gain value with the aforementioned at least one actual average gain value to obtain comparison information. Based on the obtained comparison information, it generates model validation results for the aforementioned target value payment method probability generation model.

[0014] Secondly, some embodiments of this disclosure provide a model training apparatus, comprising: a first acquisition unit configured to acquire a training dataset, wherein the training data in the training dataset includes: variable data, actual value reduction information, and actual value payment method information; a first generation unit configured to input the variable dataset into an initial feature data vector representation model included in an initial target value payment method probability generation model, to output an initial variable vector set; and a first training unit configured to train each initial discriminant model in the initial discriminant model set included in the initial target value payment method probability generation model according to the initial variable vector set, the actual value reduction information set, and the actual value payment method information set, to obtain a discriminant model set, wherein the discriminant model set includes: a first value reduction information generation model. The system comprises: a first value expenditure probability generation model and a second value expenditure probability generation model; and a second training unit configured to fix the model parameters of the first value reduction information generation model, the first value expenditure probability generation model, and the second value expenditure probability generation model, and to train the target value expenditure probability generation model by using the initial generative model set and the initial feature data vector representation model, based on the initial variable vector set, the actual value reduction information set, and the actual value expenditure information set, to obtain the target value expenditure probability generation model. The generative model set in the target value expenditure probability generation model includes: a third value expenditure probability generation model, a fourth value expenditure probability generation model, and a second value reduction information generation model.

[0015] Optionally, the aforementioned variable data includes: instrumental variable data, adjustment variable data, and confounding variable data; and the first generation unit can be configured to: select an initial adjustment variable vector from the aforementioned initial adjustment variable vector set as a first initial adjustment variable vector, and perform the following first training step: input the aforementioned first initial adjustment variable vector into the first initial value reduction information generation model included in the aforementioned initial target value expenditure mode probability generation model to generate first initial value reduction information; determine the actual value reduction information corresponding to the aforementioned first initial adjustment variable vector as the first actual value reduction information; determine the actual value reduction information corresponding to the aforementioned first initial value reduction information and the aforementioned first actual value reduction information. The first loss information of the actual value reduction information; in response to determining that the first loss information is greater than a first value, the first initial value reduction information generation model is determined as the first value reduction information generation model; in response to determining that the first loss information is less than or equal to the first value, the model parameters of the first initial value reduction information generation model are updated according to the first loss information to obtain the updated first value reduction information generation model, and a first initial adjustment variable vector is reselected from the initial adjustment variable vector set, and the updated first value reduction information generation model is determined as the first initial value reduction information generation model to continue to execute the first training step.

[0016] Optionally, the first training unit can be configured to: select an initial confusion variable vector from the aforementioned initial confusion variable vector set as a first initial confusion variable vector, and perform the following second training steps: determine the actual value reduction information corresponding to the aforementioned first initial confusion variable vector as second actual value reduction information, and determine the initial instrumental variable vector corresponding to the aforementioned first initial confusion variable vector as a first initial instrumental variable vector; generate a first actual value reduction vector for the aforementioned second actual value reduction information; input the aforementioned first initial confusion variable vector and the aforementioned first actual value reduction vector into a first initial value payment method probability generation model to generate first initial value payment method probability information; input the aforementioned first initial instrumental variable vector, the aforementioned first initial confusion variable vector, and the aforementioned first actual value reduction vector into a second initial value payment method probability generation model to generate second initial value payment method probability information; generate probability information for the aforementioned first initial value payment method and the aforementioned second initial value payment method. The second loss information of probability information; in response to determining that the second loss information is greater than the second value, the first initial value expenditure method probability generation model is determined as the first value expenditure method probability generation model, and the second initial value expenditure method probability generation model is determined as the second value expenditure method probability generation model; in response to determining that the second loss information is less than or equal to the second value, the model parameters of the first initial value expenditure method probability generation model and the second initial value expenditure method probability generation model are updated according to the second loss information to obtain the updated first value expenditure method probability generation model and the updated first value expenditure method probability generation model, and the first initial confusion variable vector is reselected from the set of initial confusion variable vectors, the updated first value expenditure method probability generation model is used as the first initial value expenditure method probability generation model, and the updated second value expenditure method probability generation model is used as the second initial value expenditure method probability generation model, so as to continue to execute the second training step.

[0017] Optionally, the second training unit can be configured to: select an initial confounding variable vector from the aforementioned initial confounding variable vector set as the second initial confounding variable vector, and perform the following third training step: determine the initial adjustment variable vector, the initial instrumental variable vector, and the corresponding actual value reduction information corresponding to the aforementioned second initial confounding variable vector, and use them as the second initial adjustment variable vector, the second initial instrumental variable vector, and the third actual value reduction information, respectively; input the aforementioned second initial confounding variable vector and the aforementioned second initial adjustment variable vector into the third initial value payment method probability generation model to generate the third initial value payment method probability information; input the aforementioned second initial adjustment variable vector into the fourth initial value payment method probability generation model to generate the fourth initial value payment method probability information; input the aforementioned second initial instrumental variable vector into the second initial value reduction information generation model to generate the second initial value... Eliminate information; generate third loss information for the probability information of the third initial value payment method and the corresponding actual value payment method information; generate fourth loss information for the probability information of the fourth initial value payment method and the corresponding actual value payment method information; generate fifth loss information for the second initial value reduction information and the third actual value reduction information; in response to determining that the third loss information is less than a third value, the fourth loss information is less than a fourth value, and the fifth loss information is less than a fifth value, the probability generation model of the third initial value payment method, the probability generation model of the fourth initial value payment method, the generation model of the second initial value reduction information, and the initial feature data vector representation model are respectively determined as the probability generation model of the third value payment method, the probability generation model of the fourth value payment method, the generation model of the second value reduction information, and the feature data vector representation model.

[0018] Thirdly, some embodiments of this disclosure provide a method for generating prediction probability information, including: acquiring variable data and value reduction information for a target user, including multiple value reduction values; for each of the multiple value reduction values, inputting the variable data and the value reduction value into a pre-trained target value payment method probability generation model to generate target value payment method prediction probability information for the value reduction value, wherein the target value payment method probability generation model is generated based on the model training method of the first aspect of this disclosure; selecting value reduction values ​​from the multiple value reduction values ​​whose target value payment method prediction probability information satisfies preset information conditions as target value reduction values; and pushing the reduction information corresponding to the target value reduction value to the user terminal corresponding to the target user.

[0019] Optionally, for each of the plurality of value reduction values, inputting the variable data and the value reduction value into a pre-trained target value payment method probability generation model to generate target value payment method prediction probability information for the value reduction value includes: for each of the plurality of value reduction values, inputting the variable data and the value reduction value into a third value payment method probability generation model included in the target value payment method probability generation model to generate target value payment method prediction probability information for the value reduction value.

[0020] Fourthly, some embodiments of this disclosure provide a predictive probability information generation apparatus, comprising: a second acquisition unit configured to acquire variable data and value reduction information for a target user, including a plurality of value reduction values; a second generation unit configured to input the variable data and the value reduction value into a pre-trained target value payment method probability generation model for each of the plurality of value reduction values, to generate target value payment method predictive probability information for the value reduction value, wherein the target value payment method probability generation model is generated based on the model training method of the first aspect of this disclosure; a filtering unit configured to filter out value reduction values ​​from the plurality of value reduction values ​​whose corresponding target value payment method predictive probability information satisfies preset information conditions, as target value reduction values; and a push unit configured to push the reduction information corresponding to the target value reduction value to the user terminal corresponding to the target user.

[0021] Optionally, the second generation unit can be configured to: for each of the above multiple value reduction values, input the above variable data and the above value reduction value into the third value payment method probability generation model included in the target value payment method probability generation model, so as to generate target value payment method prediction probability information for the above value reduction value.

[0022] Fifthly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any of the implementations of the first and third aspects.

[0023] Sixthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method as described in any of the implementations of the first and third aspects.

[0024] In a seventh aspect, some embodiments of this disclosure provide a computer program product, including a computer program that, when executed by a processor, implements the methods described in any of the implementations of the first and third aspects above.

[0025] The above embodiments of this disclosure have the following beneficial effects: By utilizing the trained target value payment method probability generation model through the model training methods of some embodiments of this disclosure, target value payment method prediction probability information can be accurately generated. Specifically, the reason for the inaccuracy of the related target value payment method prediction probability information lies in the inaccuracy of the individual causal effect representation, leading to inaccurate generated target value payment method prediction probability information. Based on this, the model training method of some embodiments of this disclosure first obtains a training dataset, wherein the training data in the training dataset includes: variable data, actual value reduction information, and actual value payment method information. Here, the obtained variable data, real-time value reduction information, and real-time value payment method information are used as input data and data labels for subsequent model training. Then, the variable dataset is input into the initial feature data vector representation model included in the initial target value payment method probability generation model to output an initial variable vector set. Here, the initial feature data vector representation model can convert the variable data in the variable dataset into vector form, so that it can be subsequently input into the initial target value payment method probability generation model. Next, based on the aforementioned initial variable vector set, actual value reduction information set, and actual value payment method information set, the initial discriminant models in the initial discriminant model set included in the initial target value payment method probability generation model are trained to obtain a discriminant model set. This discriminant model set includes: a first value reduction information generation model, a first value payment method probability generation model, and a second value payment method probability generation model. Here, by first training the discriminant models in the initial discriminant model set, an accurate discriminant model set is trained to ensure the subsequent auxiliary training models serving as the initial generative model set, thus aiding in the training of the initial generative model set. Furthermore, the first value reduction information generation model, the first value payment method probability generation model, and the second value payment method probability generation model can be discriminant models set up based on variable causal inference, accurately representing individual causal effects. Therefore, by setting up a first value reduction information generation model, a first value payment method probability generation model, and a second value payment method probability generation model as discriminators to assist in the training of subsequent third value payment method probability generation models, fourth value payment method probability generation models, and second value reduction information generation models, the accuracy of model training can be greatly improved.Finally, the model parameters of the first value reduction information generation model, the first value expenditure method probability generation model, and the second value expenditure method probability generation model are fixed. Based on the initial variable vector set, the actual value reduction information set, and the actual value expenditure method information set, the initial generative model set and the initial feature data vector representation model included in the trained target value expenditure method probability generation model are trained to obtain the target value expenditure method probability generation model. The generative model set in the target value expenditure method probability generation model includes: the third value expenditure method probability generation model, the fourth value expenditure method probability generation model, and the second value reduction information generation model. Here, the third value expenditure method probability generation model, the fourth value expenditure method probability generation model, and the second value reduction information generation model are also generative models based on causal inference, which can accurately generate corresponding individual causal effect information. With the assistance of model training using the discriminant model set, the target value expenditure method probability generation model can be accurately obtained. In summary, by first training the discriminative model set and then training the generative model set, we can fully characterize individual causal effects, enabling us to accurately generate prediction probability information of target value expenditure methods using the trained target value expenditure method probability generation model. Attached Figure Description

[0026] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0027] Figures 1-2 This is a schematic diagram illustrating an application scenario of a model training method according to some embodiments of the present disclosure;

[0028] Figure 3 This is a flowchart of some embodiments of the model training method according to this disclosure;

[0029] Figure 4 This is a flowchart of some embodiments of the method for generating prediction probability information according to this disclosure;

[0030] Figure 5 These are flowcharts of other embodiments of the method for generating prediction probability information according to this disclosure;

[0031] Figure 6 This is a schematic diagram of the structure of some embodiments of the model training apparatus according to the present disclosure;

[0032] Figure 7 These are schematic diagrams illustrating the structure of some embodiments of the predictive probability information generation apparatus according to this disclosure;

[0033] Figure 8 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0034] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0035] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0036] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0037] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0038] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0039] Before performing any of the operations involving the collection, storage, or use of user information (such as information on the reduction of actual value and information on the method of payment of actual value) disclosed in this disclosure, the relevant organizations or individuals shall fulfill their obligations, including conducting information security impact assessments, informing the information subjects, and obtaining prior authorization and consent from the information subjects.

[0040] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0041] Figures 1-2 This is a schematic diagram illustrating an application scenario of a model training method according to some embodiments of the present disclosure.

[0042] exist Figures 1-2In the application scenario, firstly, the electronic device 101 can acquire a training dataset 102. The training data in the training dataset 102 includes: variable data, actual value reduction information, and actual value payment method information. Then, the electronic device 101 can input the variable dataset 103 into the initial feature data vector representation model 107 included in the initial target value payment method probability generation model 106 to output an initial variable vector set 108. Next, the electronic device 101 can train each initial discriminant model in the initial discriminant model set 109 included in the initial target value payment method probability generation model 106 based on the initial variable vector set 108, the actual value reduction information set 104, and the actual value payment method information set 105, to obtain a discriminant model set 112. The discriminant model set 112 includes: a first value reduction information generation model 1121, a first value payment method probability generation model 1122, and a second value payment method probability generation model 1123. In this application scenario, the aforementioned initial target value expenditure probability generation model 106 further includes an initial generative model set 110. The initial generative model set 110 may include a third initial value expenditure probability generation model 1101, a fourth initial value expenditure probability generation model 1102, and a second initial value reduction information generation model 1103. Finally, the electronic device 101 can fix the model parameters of the aforementioned first value reduction information generation model 1121, the aforementioned first value expenditure probability generation model 1122, and the aforementioned second value expenditure probability generation model 1123, and, based on the aforementioned initial variable vector set 108, the aforementioned actual value reduction information set 104, and the aforementioned actual value expenditure information set 105, train the initial generative model set 110 included in the trained target value expenditure probability generation model 111 and the aforementioned initial feature data vector representation model 107 to obtain the target value expenditure probability generation model 113. The generative model set 115 in the aforementioned target value payment method probability generation model 113 includes: a third value payment method probability generation model 1151, a fourth value payment method probability generation model 1152, and a second value reduction information generation model 1153. In this application scenario, the target value payment method probability generation model 113 also includes: a feature data vector representation model 114.

[0043] It should be noted that the aforementioned electronic device 101 can be either hardware or software. When the electronic device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the electronic device is software, it can be installed in the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0044] It should be understood that Figures 1-2 The number of electronic devices shown is merely illustrative. Any number of electronic devices can be used depending on the implementation requirements.

[0045] Continue to refer to Figure 3 The diagram illustrates a flow 300 of some embodiments of a model training method according to the present disclosure. This model training method includes the following steps:

[0046] Step 301: Obtain the training dataset.

[0047] In some embodiments, the execution entity of the above model training method (e.g. Figure 1 The electronic device 101 shown can acquire the training dataset via a wired or wireless connection. The training data in the training dataset includes: variable data, actual value depreciation information, and actual value payment method information. The training data in the training dataset can be used for subsequent model training. Variable data can be user-related feature data. Specifically, feature data can be feature information corresponding to user feature variables. For example, in the e-commerce field, variable data can include, but is not limited to, at least one of the following: user historical payment preference information, user shopping information, and user identification information. For example, user historical payment preference information can be user preference information for using a target payment method to make value payments. For example, the target payment method can be the use of a target prepayment payment method. Actual value depreciation information can be the value depreciation transformation information of the corresponding item. In the e-commerce scenario, actual value depreciation information can be the coupon actually used, or it can be a payment instant discount coupon. Actual value payment method information can be the payment method information for actually making value payments. Specifically, payment method information can be the identification information of the payment method. The payment method can be various prepayment methods. Each training dataset includes variable data, actual value depreciation information, and actual value payment method information, and there is a data correspondence relationship between them.

[0048] Step 302: Input the variable dataset into the initial feature data vector representation model included in the initial target value payment method probability generation model to output the initial variable vector set.

[0049] In some embodiments, the aforementioned executing entity can input the variable dataset into the initial feature data vector representation model included in the initial target value payment method probability generation model to output an initial variable vector set. The initial target value payment method probability generation model can be a target value payment method probability generation model whose training is not yet complete. The target value payment method probability generation model can be a model that generates target value payment method prediction probability information. The target value payment method prediction probability information can be probability information assuming the user uses the target value payment method. The value payment method can be a payment method for making value payments. The target value payment method can be a pre-determined value payment method. For example, the target value payment method can be a payment method using a target application. The target value payment method prediction probability information is a value between 0 and 1. The higher the target value payment method prediction probability information, the more likely the corresponding user is to use the target value payment method as the payment method. The initial feature data vector representation model can be a feature data vector representation model whose training is not yet complete. The feature data vector representation model can be a model that converts feature data into corresponding vectors to represent the semantic content of the feature data. Specifically, the feature data can be user feature data. There is a one-to-one correspondence between the initial variable vectors in the initial variable vector set and the variable data in the variable dataset. The variable vectors can represent the semantic content of the corresponding variable data. In practice, the probability generation model for the initial target value expenditure method can be a multi-layered cascaded temporal neural network model. The initial feature data vector representation model can be a BERT encoding model.

[0050] In some optional implementations of certain embodiments, the aforementioned variable data includes: instrumental variable data, adjustment variable data, and confounding variable data. Instrumental variable data can be the content of variables corresponding to instrumental variables. Adjustment variable data can be the content of variables corresponding to adjustment variables. Confounding variable data can be the content of variables corresponding to confounding variables. Instrumental variables can be feature variables related to intervention information and user characteristics. Instrumental variables can be determining factors of intervention information. Intervention information can be information about intervention methods. Specifically, intervention information can be information about issued coupons or payment discount coupons. Adjustment variable data can be feature variables related to the method of paying the target value and user characteristics. Adjustment variables can be determining factors of the method of paying the target value. Confounding variable data are feature variables related to the method of paying the target value and intervention information, and user characteristics. Confounding variables can be determining factors of both the method of paying the target value and intervention information.

[0051] Optionally, the above-mentioned inputting the variable dataset into the initial feature data vector representation model included in the initial target value expenditure mode probability generation model to output an initial variable vector set may include the following steps:

[0052] The first step involves the execution entity inputting the instrumental variable dataset, adjustment variable dataset, and confounding variable dataset into the initial feature data vector representation model to generate initial instrumental variable vector sets, initial adjustment variable vector sets, and initial confounding variable vector sets. The initial instrumental variable vectors represent the semantic content of the instrumental variable data. The initial confounding variable vectors represent the semantic content of the confounding variable data. The initial adjustment variable vectors represent the semantic content of the adjustment variable data.

[0053] The second step is for the aforementioned executing entity to perform vector set fusion of the aforementioned initial instrumental variable vector set, the aforementioned initial adjustment variable vector set, and the aforementioned initial confusion variable vector set to obtain the initial variable vector set.

[0054] As an example, the aforementioned execution entity can perform a one-to-one vector fusion of the initial instrumental variable vector set, the aforementioned initial adjustment variable vector set, and the aforementioned initial obfuscation variable vector set to generate a fused vector, which serves as the initial variable vector, thus obtaining the initial variable vector set.

[0055] Optionally, the initial feature data vector representation model includes: an instrumental variable data vector representation model for instrumental variable data, an adjustment variable data vector representation model for adjustment variable data, and a confounding variable data vector representation model for confounding variable data. In practice, the instrumental variable data vector representation model can be a BERT pre-trained model. The adjustment variable data vector representation model can be a word embedding model. The confounding variable data vector representation model can be a Transformer encoding model.

[0056] The aforementioned executing entity can input the instrumental variable dataset, adjustment variable dataset, and confounding variable dataset into the aforementioned initial feature data vector representation model to generate the initial instrumental variable vector set, the initial adjustment variable vector set, and the initial confounding variable vector set, including the following steps:

[0057] The first step is to input each instrumental variable data in the instrumental variable dataset into the instrumental variable data vector representation model to generate an initial instrumental variable vector, thus obtaining the initial instrumental variable vector set.

[0058] The second step is to input each adjusted variable data in the adjusted variable dataset into the adjusted variable data vector representation model to generate an initial adjusted variable vector, thus obtaining the initial adjusted variable vector set.

[0059] The third step is to input each of the confounding variable data in the above confounding variable dataset into the confounding variable data vector representation model to generate an initial confounding variable vector, thus obtaining the initial confounding variable vector set.

[0060] Step 303: Based on the above initial variable vector set, actual value reduction information set, and actual value payment method information set, train each initial discriminant model in the initial discriminant model set included in the above initial target value payment method probability generation model to obtain the discriminant model set.

[0061] In some embodiments, the aforementioned executing entity can train each initial discriminant model in the initial discriminant model set included in the initial target value payment method probability generation model based on the aforementioned initial variable vector set, actual value reduction information set, and actual value payment method information set, to obtain a discriminant model set. The initial discriminant models in the initial discriminant model set may be discriminant models whose training has not yet been completed. The aforementioned discriminant model set includes: a first value reduction information generation model, a first value payment method probability generation model, and a second value payment method probability generation model. The first value reduction information generation model can be a model that generates value reduction information based on the initial variable vector set. In practice, the first value reduction information generation model can be a recurrent neural network model. The first value payment method probability generation model can be a model that generates probability information about using the target value payment method based on the first initial variable vector and the corresponding actual value reduction information. In practice, the first value payment method probability generation model can be a multi-layered serially connected recurrent neural network model. The second value payment method probability generation model can be a model that generates probability information about using the target value payment method based on the second initial variable vector and the corresponding actual value reduction information. In practice, the probabilistic generation model for the second value payment method can be a multi-layered, serially connected recurrent neural network model. The vector content of the first initial variable vector corresponding to the probabilistic generation model for the first value payment method differs from that of the probabilistic generation model for the second value payment method. Correspondingly, the network structure of the first value payment method probabilistic generation model can also differ from that of the second value payment method probabilistic generation model. Specifically, the number of network layers differs between the two.

[0062] In some optional implementations of certain embodiments, the process of training each initial discriminant model in the initial discriminant model set included in the initial target value expenditure method probability generation model based on the initial variable vector set, the actual value reduction information set, and the actual value expenditure method information set to obtain a discriminant model set may include the following steps:

[0063] The first step is to select an initial adjustment variable vector from the above set of initial adjustment variable vectors as the first initial adjustment variable vector, and then perform the following first training step:

[0064] Sub-step 1: The aforementioned executing entity can input the first initial adjustment variable vector into the first initial value reduction information generation model included in the initial target value expenditure method probability generation model to generate the first initial value reduction information. The first initial value reduction information generation model can be a first value reduction information generation model that has not yet finished training. The first value reduction information generation model can be a model that predicts value reduction information based on adjustment variable data.

[0065] In sub-step 2, the aforementioned executing entity can determine the actual value reduction information corresponding to the aforementioned initial adjustment variable vector, as the first actual value reduction information.

[0066] Sub-step 3: The executing entity can determine the first loss information for the first initial value reduction information and the first actual value reduction information. The first loss information can characterize the value difference information between the first initial value reduction information and the first actual value reduction information.

[0067] As an example, the aforementioned executing entity can utilize the cross-entropy loss function to generate first loss information for the aforementioned first initial value reduction information and the aforementioned first actual value reduction information.

[0068] Sub-step 4: In response to determining that the first loss information is greater than the first value, the executing entity can determine the first initial value reduction information generation model as the first value reduction information generation model. The first value can be a pre-set value.

[0069] The second step involves, in response to determining that the first loss information is less than or equal to the first value, the executing entity can train the model parameters of the first initial value reduction information generation model based on the first loss information to obtain the updated first value reduction information generation model, and reselect the first initial adjustment variable vector from the set of initial adjustment variable vectors to determine the updated first value reduction information generation model as the first initial value reduction information generation model, so as to continue to execute the first training step.

[0070] In some optional implementations of certain embodiments, the process of training each initial discriminant model in the initial discriminant model set included in the initial target value expenditure method probability generation model based on the initial variable vector set, the actual value reduction information set, and the actual value expenditure method information set to obtain a discriminant model set may include the following steps:

[0071] The first step is for the execution entity to select an initial obfuscation variable vector from the aforementioned initial obfuscation variable vector set as the first initial obfuscation variable vector, and then perform the following second training step:

[0072] Sub-step 1: The aforementioned executing entity can determine the actual value reduction information corresponding to the first initial obfuscation variable vector as the second actual value reduction information, and determine the initial instrumental variable vector corresponding to the first initial obfuscation variable vector as the first initial instrumental variable vector.

[0073] Sub-step 2: The aforementioned executing entity can generate a first actual value reduction vector for the aforementioned second actual value reduction information.

[0074] As an example, the aforementioned executing entity can use the initial feature data vector representation model to generate a first actual value reduction vector for the aforementioned second actual value reduction information.

[0075] Sub-step 3 involves inputting the aforementioned first initial confusion variable vector and the aforementioned first actual value reduction vector into the first initial value expenditure method probability generation model to generate the first initial value expenditure method probability information. The first initial value expenditure method probability generation model can be a model that has not yet completed its training. The first value expenditure method probability generation model can be a model that generates predicted probability information for the value expenditure method based on the confusion variable data and value reduction information.

[0076] Sub-step 4 involves inputting the first initial instrumental variable vector, the first initial confounding variable vector, and the first actual value reduction vector into the second initial value expenditure mode probability generation model to generate second initial value expenditure mode probability information. The second initial value expenditure mode probability generation model can be a model that has not yet completed training. The second value expenditure mode probability generation model can be a model that generates predicted probability information for value expenditure modes based on instrumental variable data, confounding variable data, and value reduction information.

[0077] Sub-step 5 generates second loss information based on the probability information of the first initial value payment method and the probability information of the second initial value payment method. The second loss information characterizes the difference in probability information between the first and second initial value payment method probabilities.

[0078] As an example, the aforementioned executing entity can utilize the cross-entropy loss function to generate second loss information for the probability information of the first initial value payment method and the probability information of the second initial value payment method.

[0079] Sub-step 6: In response to determining that the second loss information is greater than the second value, the first initial value payment method probability generation model is determined as the first value payment method probability generation model, and the second initial value payment method probability generation model is determined as the second value payment method probability generation model. The second value can be a pre-defined numerical value.

[0080] The second step involves, in response to determining that the second loss information is less than or equal to the second value, training the model parameters of the first initial value payment method probability generation model and the second initial value payment method probability generation model based on the second loss information, to obtain the updated first value payment method probability generation model and the updated first value payment method probability generation model. Additionally, a new first initial confusion variable vector is selected from the initial confusion variable vector set, and the updated first value payment method probability generation model is used as the first initial value payment method probability generation model, and the updated second value payment method probability generation model is used as the second initial value payment method probability generation model, to continue executing the second training step.

[0081] Step 304: Fix the model parameters of the first value reduction information generation model, the first value payment method probability generation model, and the second value payment method probability generation model. Based on the initial variable vector set, the actual value reduction information set, and the actual value payment method information set, train the initial generative model set and the initial feature data vector representation model included in the trained target value payment method probability generation model to obtain the target value payment method probability generation model.

[0082] In some embodiments, the executing entity can fix the model parameters of the first value reduction information generation model, the first value payment method probability generation model, and the second value payment method probability generation model, and train the target value payment method probability generation model by using the initial generative model set and the initial feature data vector representation model, based on the initial variable vector set, the actual value reduction information set, and the actual value payment method information set, to obtain the target value payment method probability generation model. The generative model set in the target value payment method probability generation model includes: a third value payment method probability generation model, a fourth value payment method probability generation model, and a second value reduction information generation model. Fixing the model parameters means that the model parameters of the first value reduction information generation model, the first value payment method probability generation model, and the second value payment method probability generation model do not change during the subsequent training of the target value payment method probability generation model. The generative model set in the target value payment method probability generation model includes: a third value payment method probability generation model, a fourth value payment method probability generation model, and a second value reduction information generation model. The third value payment method probability generation model can be a model that generates probability information for using the target value payment method based on the third initial variable vector and the corresponding actual value reduction information. In practice, the third value payment method probability generation model can be a multi-layered, serially connected recurrent neural network model. The vector content of the first initial variable vector corresponding to the first value payment method probability generation model, the second value payment method probability generation model, and the third value payment method probability generation model are different. Correspondingly, the network structure of the first value payment method probability generation model can also be different from the network structures of the second and third value payment method probability generation models. Specifically, the number of network layers for the three models is different. The fourth value payment method probability generation model can be a model that generates probability information for using the target value payment method based on the fourth initial variable vector. Specifically, the fourth initial variable vector is different from the first, second, and third initial variable vectors. The network structures of the first, second, third, and fourth value payment method probability generation models can all be different.

[0083] In some optional implementations of certain embodiments, the above-mentioned training of the target value expenditure mode probability generation model, including the initial generative model set and the initial feature data vector representation model, based on the initial variable vector set, the actual value reduction information set, and the actual value expenditure mode information set, to obtain the target value expenditure mode probability generation model, may include the following steps:

[0084] The first step is to select an initial confusion variable vector from the above initial confusion variable vector set as the second initial confusion variable vector, and then perform the following third training step:

[0085] Sub-step 1: Determine the initial adjustment variable vector, initial instrumental variable vector, corresponding actual value reduction vector, and corresponding actual value reduction information corresponding to the above-mentioned second initial confusion variable vector, and use them as the second initial adjustment variable vector, the second initial instrumental variable vector, the third actual value reduction vector, and the third actual value reduction information, respectively.

[0086] Sub-step 2 involves inputting the aforementioned second initial confounding variable vector, the aforementioned third actual value reduction vector, and the aforementioned second initial adjustment variable vector into the third initial value expenditure method probability generation model to generate the third initial value expenditure method probability information. The third initial value expenditure method probability generation model can be a model that has not yet completed training. The third value expenditure method probability generation model can be a model that generates value expenditure method prediction probability information based on confounding variable data, value reduction data, and adjustment variable data.

[0087] Sub-step 3 involves inputting the aforementioned second initial adjustment variable vector into the fourth initial value expenditure method probability generation model to generate the fourth initial value expenditure method probability information. This fourth initial value expenditure method probability generation model can be a model that has not yet completed its training. The fourth value expenditure method probability generation model can be a model that generates value expenditure method prediction probability information based on the adjustment variable data.

[0088] Sub-step 4 involves inputting the aforementioned second initial instrumental variable vector into the second initial value reduction information generation model to generate second initial value reduction information. This second initial value reduction information generation model can be a model that has not yet completed its training. The second value reduction information generation model can be a model that generates value reduction information based on instrumental variable data.

[0089] Sub-step 5 generates third loss information based on the probability information of the third initial value payment method and the corresponding actual value payment method information. The third loss information characterizes the information difference between the probability information of the third initial value payment method and the corresponding actual value payment method information. For specific implementation details, please refer to the generation of second loss information.

[0090] Sub-step 6 generates fourth loss information based on the probability information of the fourth initial value payment method and the corresponding actual value payment method information. The fourth loss information characterizes the information difference between the probability information of the fourth initial value payment method and the corresponding actual value payment method information. For specific implementation details, please refer to the generation of second loss information.

[0091] Sub-step 7 generates fifth loss information corresponding to the second initial value reduction information and the third actual value reduction information. The fifth loss information characterizes the information difference between the second initial value reduction information and the third actual value reduction information. For specific implementation details, please refer to the generation of the second loss information.

[0092] Sub-step 8, in response to determining that the third loss information is less than the third value, the fourth loss information is less than the fourth value, and the fifth loss information is less than the fifth value, determines the third initial value expenditure method probability generation model, the fourth initial value expenditure method probability generation model, the second initial value reduction information generation model, and the initial feature data vector representation model as the third value expenditure method probability generation model, the fourth value expenditure method probability generation model, the second value reduction information generation model, and the feature data vector representation model, respectively. The third, fourth, and fifth values ​​can be pre-set values.

[0093] Optionally, the steps also include:

[0094] In the first step, in response to determining that at least one of the following exists: a third loss information greater than or equal to a third value, a fourth loss information greater than or equal to a fourth value, and a fifth loss information greater than or equal to a fifth value, the model parameters of the third initial value expenditure method probability generation model, the fourth initial value expenditure method probability generation model, the second initial value reduction information generation model, and the initial feature data vector representation model are updated based on the third, fourth, and fifth loss information. This results in updated third initial value expenditure method probability generation model, updated fourth initial value expenditure method probability generation model, updated second initial value reduction information generation model, and updated initial feature data vector representation model, which are respectively used as the third initial value expenditure method probability generation model, the fourth initial value expenditure method probability generation model, the second initial value reduction information generation model, and the initial feature data vector representation model. Additionally, a second initial confusion variable vector is reselected to continue the third training step.

[0095] In some optional implementations of certain embodiments, the above-mentioned target value expenditure probability generation model is validated through the following steps:

[0096] The first step is to obtain a user information set. This user information can be identity information representing a user's identity. In practice, user information can be user identification information. The user information in the user information set can be used to subsequently verify the model effectiveness of the probability generation model for the target value payment method.

[0097] The second step involves performing the following information generation steps for each of the multiple value reduction values ​​included in the value reduction information:

[0098] Sub-step 1: Determine the value distribution value corresponding to each user information in the aforementioned user information set. Here, value reduction information can be the value reduction transformation information of the corresponding item. For e-commerce scenarios, value reduction information can be a payment discount coupon. The value reduction value can be the degree of reduction transformation of the value information. Each value reduction value among multiple value reduction values ​​is different. Specifically, for the value reduction information being a payment discount coupon, the corresponding value reductions could be {"0 yuan", "1 yuan", "3 yuan", "5 yuan", "10 yuan"}. The value distribution value can be the discount value corresponding to the payment discount coupon actually issued to the user information.

[0099] Sub-step 2 involves selecting a subset of user information whose corresponding value distribution value is the target value from the aforementioned user information set, and using this subset as the first user information subset. The target value can be "0".

[0100] Sub-step 3: Select a subset of user information whose corresponding value distribution value is the aforementioned value reduction value from the above user information set, and use it as the second subset of user information.

[0101] Sub-step 4: Merge the first subset of user information and the second subset of user information to obtain a merged user information set.

[0102] Sub-step 5: For each piece of information in the aforementioned integrated user information set, using the aforementioned target value payment method probability generation model, determine the first target value payment method prediction probability information and the second target value payment method prediction probability information for the aforementioned integrated user information under the aforementioned value reduction value. For example, the value reduction value is "3". The first target value payment method prediction probability information for the integrated user information under the aforementioned value reduction value can be the probability information of the user using the target value payment method when the instant discount coupon corresponds to a discount value of 3 yuan. The second target value payment method prediction probability information for the integrated user information under the aforementioned target value can be the probability information of the user using the target value payment method when the instant discount coupon corresponds to a discount value of 0 yuan.

[0103] Sub-step 6: Based on the information difference between the first target value payment method prediction probability information and the second target value payment method prediction probability information corresponding to each fused user information, group the fused user information in the above fused user information set to obtain at least one user information group.

[0104] As an example, firstly, for each piece of merged user information, the corresponding information difference is determined. Then, the number of user information groups is determined. Finally, based on the magnitude of the information difference, each piece of merged user information is grouped to obtain at least one user information group.

[0105] Sub-step 7: For each user information group in the above at least one user information group, determine the predicted average gain value corresponding to the user information group based on the information difference set corresponding to the user information group.

[0106] As an example, the average information difference set corresponding to the user information group is averaged to obtain the average information difference, which is used as the prediction average gain value corresponding to the above user information group.

[0107] Sub-step 8 involves determining the actual average gain value for each user information group, resulting in at least one actual average gain value. The actual average gain value characterizes the impact of payment reduction information (where the value issued is the same as the value reduced) on each user information within the user information group. Specifically, the actual average gain value can be the degree to which payment reduction information (where the value issued is the same as the value reduced) influences a user's choice of a target value payment method.

[0108] As an example, for each user information group, firstly, user information whose corresponding value distribution value is the target value is identified as the first user information subgroup. Then, user information whose corresponding value distribution value is the aforementioned value reduction value is identified as the second user information subgroup. Finally, the proportion of users in the second user information subgroup is divided by the proportion of users in the first user information subgroup to obtain the actual average gain value.

[0109] Sub-step 9 involves comparing the obtained at least one predicted average gain value with the at least one actual average gain value to obtain comparison information. There is a one-to-one correspondence between the predicted average gain value and the actual average gain value among the at least one actual average gain value. The comparison information characterizes the difference between the at least one actual average gain value and the at least one predicted average gain value.

[0110] The third step is to generate model validation results for the probability generation model of the above-mentioned target value payment method based on the obtained comparison information.

[0111] As an example, the aforementioned execution entity can generate accuracy information based on multiple comparison pieces of information. Then, based on the accuracy information, it generates a model validation result indicating whether the probability generation model for the target value expenditure method has passed validation.

[0112] The above embodiments of this disclosure have the following beneficial effects: By utilizing the trained target value payment method probability generation model through the model training methods of some embodiments of this disclosure, target value payment method prediction probability information can be accurately generated. Specifically, the reason for the inaccuracy of the related target value payment method prediction probability information lies in the inaccuracy of the individual causal effect representation, leading to inaccurate generated target value payment method prediction probability information. Based on this, the model training method of some embodiments of this disclosure first obtains a training dataset, wherein the training data in the training dataset includes: variable data, actual value reduction information, and actual value payment method information. Here, the obtained variable data, real-time value reduction information, and real-time value payment method information are used as input data and data labels for subsequent model training. Then, the variable dataset is input into the initial feature data vector representation model included in the initial target value payment method probability generation model to output an initial variable vector set. Here, the initial feature data vector representation model can convert the variable data in the variable dataset into vector form, so that it can be subsequently input into the initial target value payment method probability generation model. Next, based on the aforementioned initial variable vector set, actual value reduction information set, and actual value payment method information set, the initial discriminant models in the initial discriminant model set included in the initial target value payment method probability generation model are trained to obtain a discriminant model set. This discriminant model set includes: a first value reduction information generation model, a first value payment method probability generation model, and a second value payment method probability generation model. Here, by first training the discriminant models in the initial discriminant model set, an accurate discriminant model set is trained to ensure the subsequent auxiliary training models serving as the initial generative model set, thus aiding in the training of the initial generative model set. Furthermore, the first value reduction information generation model, the first value payment method probability generation model, and the second value payment method probability generation model can be discriminant models set up based on variable causal inference, accurately representing individual causal effects. Therefore, by setting up a first value reduction information generation model, a first value payment method probability generation model, and a second value payment method probability generation model as discriminators to assist in the training of subsequent third value payment method probability generation models, fourth value payment method probability generation models, and second value reduction information generation models, the accuracy of model training can be greatly improved.Finally, the model parameters of the first value reduction information generation model, the first value expenditure method probability generation model, and the second value expenditure method probability generation model are fixed. Based on the initial variable vector set, the actual value reduction information set, and the actual value expenditure method information set, the initial generative model set and the initial feature data vector representation model included in the trained target value expenditure method probability generation model are trained to obtain the target value expenditure method probability generation model. The generative model set in the target value expenditure method probability generation model includes: the third value expenditure method probability generation model, the fourth value expenditure method probability generation model, and the second value reduction information generation model. Here, the third value expenditure method probability generation model, the fourth value expenditure method probability generation model, and the second value reduction information generation model are also generative models based on causal inference, which can accurately generate corresponding individual causal effect information. With the assistance of model training using the discriminant model set, the target value expenditure method probability generation model can be accurately obtained. In summary, by first training the discriminative model set and then training the generative model set, we can fully characterize individual causal effects, enabling us to accurately generate prediction probability information of target value expenditure methods using the trained target value expenditure method probability generation model.

[0113] Continue to refer to Figure 4 The diagram illustrates a flow 400 of some embodiments of a prediction probability information generation method according to the present disclosure. This prediction probability information generation method includes the following steps:

[0114] Step 401: Obtain variable data and value reduction information for the target user, including multiple value reduction values.

[0115] In some embodiments, the entity executing the above-described method for generating probability prediction information (e.g., an electronic device) can acquire variable data and value reduction information for the target user, including multiple value reduction values, via wired or wireless means. Specifically, the variable data and multiple value reduction values ​​will not be further explained.

[0116] Step 402: For each of the above multiple value reduction values, input the above variable data and the above value reduction value into the pre-trained target value payment method probability generation model to generate target value payment method prediction probability information for the above value reduction value.

[0117] In some embodiments, the executing entity may input the variable data and the value reduction value into a pre-trained target value payment method probability generation model for each of the plurality of value reduction values, to generate target value payment method prediction probability information for the value reduction value. The target value payment method probability generation model is generated based on a model training method. The target value payment method prediction probability information can be the probability information of a target user choosing the target value payment method to make value payment, assuming the instant discount value corresponding to the instant discount coupon is a value reduction value.

[0118] Step 403: Select the value reduction value from the above multiple value reduction values ​​that corresponds to the target value payment method prediction probability information that meets the preset information conditions, and use it as the target value reduction value.

[0119] In some embodiments, the executing entity may select the value reduction value from the plurality of value reduction values ​​that corresponds to a target value payment method prediction probability information that meets a preset information condition, and use it as the target value reduction value. The preset information condition may be the largest value among the plurality of value reduction values.

[0120] Step 404: Push the reduction information corresponding to the above target value reduction value to the user terminal corresponding to the above target user.

[0121] In some embodiments, the executing entity may push the reduction information corresponding to the target value reduction to the user terminal corresponding to the target user. The reduction information corresponding to the target value reduction may be a payment discount coupon with a payment reduction of the target value reduction.

[0122] The above embodiments of this disclosure have the following beneficial effects: the prediction probability information method of some embodiments of this disclosure can push accurate reduction information to the target user.

[0123] Further reference Figure 5 The diagram illustrates a flow 500 of another embodiment of the prediction probability information generation method according to the present disclosure. This prediction probability information generation method includes the following steps:

[0124] Step 501: Obtain variable data and value reduction information for the target user, including multiple value reduction values.

[0125] In some embodiments, an implementing entity (e.g., an electronic device) may acquire variable data and value reduction information for a target user, including multiple value reduction values, via wired or wireless means.

[0126] Step 502: For each of the above multiple value reduction values, input the above variable data and the above value reduction value into the third value payment method probability generation model included in the target value payment method probability generation model, so as to generate target value payment method prediction probability information for the above value reduction value.

[0127] In some embodiments, the execution entity may input the variable data and the value reduction value into the third value payment method probability generation model included in the target value payment method probability generation model for each of the plurality of value reduction values, so as to generate target value payment method prediction probability information for the value reduction value.

[0128] Step 503: Select the value reduction value from the above multiple value reduction values ​​that corresponds to the target value payment method prediction probability information that meets the preset information conditions, and use it as the target value reduction value.

[0129] Step 504: Push the reduction information corresponding to the above-mentioned target value reduction value to the user terminal corresponding to the above-mentioned target user.

[0130] In some embodiments, the specific implementation of steps 501, 503-504 and their resulting technical effects can be found in [reference needed]. Figure 4 Steps 401 and 403-404 in the corresponding embodiments will not be repeated here.

[0131] from Figure 5 It can be seen from this that, with Figure 4 Compared to the description of some corresponding embodiments, Figure 5 In some corresponding embodiments, the process 500 of the prediction probability information generation method utilizes a third value payment method probability generation model to accurately generate target value payment method prediction probability information for the value reduction value.

[0132] Further reference Figure 6 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a model training apparatus, which are similar to... Figure 3 Corresponding to the method embodiments shown, this model training device can be specifically applied to various electronic devices.

[0133] like Figure 6As shown, a model training device 600 includes: a first acquisition unit 601, a first generation unit 602, a first training unit 603, and a second training unit 604. The first acquisition unit 601 is configured to acquire a training dataset, wherein the training data in the training dataset includes: variable data, actual value reduction information, and actual value payment method information; the first generation unit 602 is configured to input the variable dataset into an initial feature data vector representation model included in the initial target value payment method probability generation model to output an initial variable vector set; the first training unit 603 is configured to train each initial discriminant model in the initial discriminant model set included in the initial target value payment method probability generation model based on the initial variable vector set, the actual value reduction information set, and the actual value payment method information set, to obtain a discriminant model set, wherein the discriminant model set includes: a first value reduction information generation model, a first value payment method probability generation model, and a second training model. The first value depreciation information generation model and the second value payment method probability generation model are configured to fix the model parameters of the first value depreciation information generation model, the first value payment method probability generation model, and the second value payment method probability generation model, and to train the target value payment method probability generation model by using the initial generative model set and the initial feature data vector representation model, based on the initial variable vector set, the actual value depreciation information set, and the actual value payment method information set, to obtain the target value payment method probability generation model. The generative model set in the target value payment method probability generation model includes: the third value payment method probability generation model, the fourth value payment method probability generation model, and the second value depreciation information generation model.

[0134] In some optional implementations of some embodiments, the variable data in the above-mentioned variable dataset includes: instrumental variable data, adjustment variable data, and obfuscation variable data; and the first generation unit 602 can be further configured to: input the instrumental variable dataset, adjustment variable dataset, and obfuscation variable dataset into the above-mentioned initial feature data vector representation model to generate an initial instrumental variable vector set, an initial adjustment variable vector set, and an initial obfuscation variable vector set; and perform vector set fusion on the above-mentioned initial instrumental variable vector set, the above-mentioned initial adjustment variable vector set, and the above-mentioned initial obfuscation variable vector set to obtain an initial variable vector set.

[0135] In some optional implementations of certain embodiments, the first training unit 603 may be further configured to: select an initial adjustment variable vector from the aforementioned initial adjustment variable vector set as a first initial adjustment variable vector, and perform the following first training steps: input the aforementioned first initial adjustment variable vector into the first initial value reduction information generation model included in the aforementioned initial target value expenditure mode probability generation model to generate first initial value reduction information; determine the actual value reduction information corresponding to the aforementioned initial adjustment variable vector as the first actual value reduction information; and determine the first initial value reduction information and the aforementioned first actual value reduction information. The first loss information is determined; in response to determining that the first loss information is greater than a first value, the first initial value reduction information generation model is determined as the first value reduction information generation model; in response to determining that the first loss information is less than or equal to the first value, the model parameters of the first initial value reduction information generation model are updated according to the first loss information to obtain the updated first value reduction information generation model, and a first initial adjustment variable vector is reselected from the initial adjustment variable vector set, and the updated first value reduction information generation model is determined as the first initial value reduction information generation model to continue to execute the first training step.

[0136] In some optional implementations of certain embodiments, the first training unit 603 may be further configured to: select an initial confusion variable vector from the aforementioned initial confusion variable vector set as a first initial confusion variable vector, and perform the following second training steps: determine the actual value reduction information corresponding to the aforementioned first initial confusion variable vector as second actual value reduction information, and determine the initial instrumental variable vector corresponding to the aforementioned first initial confusion variable vector as a first initial instrumental variable vector; generate a first actual value reduction vector for the aforementioned second actual value reduction information; input the aforementioned first initial confusion variable vector and the aforementioned first actual value reduction vector into a first initial value payment method probability generation model to generate first initial value payment method probability information; input the aforementioned first initial instrumental variable vector, the aforementioned first initial confusion variable vector, and the aforementioned first actual value reduction vector into a second initial value payment method probability generation model to generate second initial value payment method probability information; generate a first actual value reduction vector for the aforementioned first initial value payment method probability information and the aforementioned... The second loss information of the probability information of the second initial value payment method; in response to determining that the second loss information is greater than the second value, the first initial value payment method probability generation model is determined as the first value payment method probability generation model, and the second initial value payment method probability generation model is determined as the second value payment method probability generation model; in response to determining that the second loss information is less than or equal to the second value, the model parameters of the first initial value payment method probability generation model and the second initial value payment method probability generation model are trained according to the second loss information to obtain the updated first value payment method probability generation model and the updated first value payment method probability generation model, and the first initial confusion variable vector is reselected from the set of initial confusion variable vectors, the updated first value payment method probability generation model is used as the first initial value payment method probability generation model, and the updated second value payment method probability generation model is used as the second initial value payment method probability generation model, so as to continue to execute the second training step.

[0137] In some optional implementations of certain embodiments, the second training unit 604 may be further configured to: select an initial confusion variable vector from the aforementioned initial confusion variable vector set as a second initial confusion variable vector, and perform the following third training step: determine the initial adjustment variable vector, initial instrumental variable vector, corresponding actual value reduction vector, and corresponding actual value reduction information corresponding to the aforementioned second initial confusion variable vector, respectively as the second initial adjustment variable vector, the second initial instrumental variable vector, the third actual value reduction vector, and the third actual value reduction information; input the aforementioned second initial confusion variable vector, the aforementioned third actual value reduction vector, and the aforementioned second initial adjustment variable vector into the third initial value payment method probability generation model to generate third initial value payment method probability information; input the aforementioned second initial adjustment variable vector into the fourth initial value payment method probability generation model to generate fourth initial value payment method probability information; and input the aforementioned second initial instrumental variable vector into the third initial value payment method probability generation model. The input is fed into the second initial value reduction information generation model to generate second initial value reduction information; third loss information is generated for the probability information of the third initial value payment method and the corresponding actual value payment method information; fourth loss information is generated for the probability information of the fourth initial value payment method and the corresponding actual value payment method information; fifth loss information is generated for the second initial value reduction information and the third actual value reduction information; in response to determining that the third loss information is less than the third value, the fourth loss information is less than the fourth value, and the fifth loss information is less than the fifth value, the probability generation model of the third initial value payment method, the probability generation model of the fourth initial value payment method, the generation model of the second initial value reduction information, and the initial feature data vector representation model are respectively determined as the probability generation model of the third value payment method, the probability generation model of the fourth value payment method, the generation model of the second value reduction information, and the feature data vector representation model.

[0138] It is understandable that the units described in the model training device 600 are related to the reference... Figure 3 The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the model training device 600 and the units contained therein, and will not be repeated here.

[0139] Further reference Figure 7 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a predictive probability information generation device, which are similar to... Figure 4 Corresponding to the method embodiments shown, this predictive probability information generation device can be specifically applied to various electronic devices.

[0140] like Figure 7As shown, a predictive probability information generation device 700 includes: a second acquisition unit 701, a second generation unit 702, a filtering unit 703, and a push unit 704. The second acquisition unit 701 is configured to acquire variable data and value reduction information for a target user, including multiple value reduction values. The second generation unit 702 is configured to input the variable data and the value reduction value into a pre-trained target value payment method probability generation model for each of the multiple value reduction values ​​to generate target value payment method predictive probability information for the value reduction value. The target value payment method probability generation model is generated based on a model training method. The filtering unit 703 is configured to filter out value reduction values ​​from the multiple value reduction values ​​that satisfy preset information conditions for the target value payment method predictive probability information, and use these as target value reduction values. The push unit 704 is configured to push the reduction information corresponding to the target value reduction value to the user terminal corresponding to the target user.

[0141] In some optional implementations of some embodiments, the second generation unit 702 may be further configured to: for each of the plurality of value reduction values, input the variable data and the value reduction value into the third value payment method probability generation model included in the target value payment method probability generation model, so as to generate target value payment method prediction probability information for the value reduction value.

[0142] It is understandable that the units described in the prediction probability information generation device 700 and the reference Figure 4 The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the prediction probability information generation device 700 and the units contained therein, and will not be repeated here.

[0143] The following is for reference. Figure 8 It illustrates electronic devices suitable for implementing some embodiments of this disclosure (e.g., Figure 1 A schematic diagram of the structure of electronic device 101)800 in the middle. Figure 8 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0144] like Figure 8As shown, the electronic device 800 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory 802 or a program loaded from a storage device 808 into a random access memory 803. The random access memory 803 also stores various programs and data required for the operation of the electronic device 800. The processing unit 801, the read-only memory 802, and the random access memory 803 are interconnected via a bus 804. An input / output interface 805 is also connected to the bus 804.

[0145] Typically, the following devices can be connected to the input / output interface 805: input devices 806 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 807 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 808 including, for example, magnetic tape, hard disk, etc.; and communication devices 809. Communication device 809 allows electronic device 800 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 8 An electronic device 800 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 8 Each box shown can represent a device or multiple devices as needed.

[0146] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 809, or installed from a storage device 808, or installed from a read-only memory 802. When the computer program is executed by the processing device 801, it performs the functions defined above in the methods of some embodiments of this disclosure.

[0147] It should be noted that, in some embodiments of this disclosure, the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0148] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0149] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: acquire a training dataset, wherein the training data in the training dataset includes: variable data, actual value reduction information, and actual value payment method information;

[0150] The variable dataset is input into the initial feature data vector representation model included in the initial target value expenditure method probability generation model to output the initial variable vector set;

[0151] Based on the aforementioned initial variable vector set, actual value reduction information set, and actual value payment method information set, the initial discriminant models in the initial discriminant model set included in the aforementioned initial target value payment method probability generation model are trained to obtain a discriminant model set. This discriminant model set includes: a first value reduction information generation model, a first value payment method probability generation model, and a second value payment method probability generation model. With the model parameters of the first value reduction information generation model, the first value payment method probability generation model, and the second value payment method probability generation model fixed, the initial generative model set included in the trained target value payment method probability generation model and the aforementioned initial feature data vector representation model are trained based on the aforementioned initial variable vector set, the actual value reduction information set, and the actual value payment method information set to obtain a target value payment method probability generation model. This generative model set in the target value payment method probability generation model includes: a third value payment method probability generation model, a fourth value payment method probability generation model, and a second value reduction information generation model. The process involves acquiring multiple value reduction values, including variable data and value reduction information for the target user; for each of these multiple value reduction values, inputting the variable data and the value reduction value into a pre-trained target value payment method probability generation model to generate target value payment method prediction probability information for the value reduction value, wherein the target value payment method probability generation model is generated based on a model training method; selecting value reduction values ​​from the multiple value reduction values ​​whose target value payment method prediction probability information meets preset information conditions, and using these as target value reduction values; and pushing the reduction information corresponding to the target value reduction values ​​to the user terminal corresponding to the target user.

[0152] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0153] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0154] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including a first acquisition unit, a first generation unit, a first training unit, and a second training unit. The names of these units do not necessarily limit the specific unit; for example, the first acquisition unit may also be described as a "unit for acquiring training datasets."

[0155] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0156] Some embodiments of this disclosure also provide a computer program product, including a computer program that, when executed by a processor, implements any of the model training methods or prediction probability information generation methods described above.

[0157] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A model training method, comprising: obtaining a training data set, wherein the training data in the training data set comprises variable data, actual value reduction information, and actual value payment method information; inputting a variable data set into an initial feature data vector representation model included in an initial target value payment method probability generation model to output an initial variable vector set; performing model training on each of an initial discriminant model set included in the initial target value payment method probability generation model according to the initial variable vector set, an actual value reduction information set, and an actual value payment method information set, to obtain a discriminant model set, wherein the discriminant model set comprises a first value reduction information generation model, a first value payment method probability generation model, and a second value payment method probability generation model; fixing model parameters of the first value reduction information generation model, the first value payment method probability generation model, and the second value payment method probability generation model, and performing model training on an initial generative model set included in a trained target value payment method probability generation model and the initial feature data vector representation model according to the initial variable vector set, the actual value reduction information set, and the actual value payment method information set, to obtain a target value payment method probability generation model, wherein the generative model set in the target value payment method probability generation model comprises a third value payment method probability generation model, a fourth value payment method probability generation model, and a second value reduction information generation model.

2. The method of claim 1, wherein, the variable data in the variable data set comprises tool variable data, adjustment variable data, and confusion variable data, the tool variable data is a decision variable factor of intervention information, the adjustment variable data is a characteristic variable associated with a target value payment method and related to user characteristics, the confusion variable data is a characteristic variable associated with a target value payment method and intervention information and related to user characteristics, and the intervention information is issued coupon information or payment reduction coupon information; and the inputting of the variable data set into the initial feature data vector representation model included in the initial target value payment method probability generation model to output the initial variable vector set comprises: inputting a tool variable data set, an adjustment variable data set, and a confusion variable data set into the initial feature data vector representation model to generate an initial tool variable vector set, an initial adjustment variable vector set, and an initial confusion variable vector set; performing vector set fusion on the initial tool variable vector set, the initial adjustment variable vector set, and the initial confusion variable vector set to obtain an initial variable vector set.

3. The method of claim 2, wherein, the performing of model training on each of the initial discriminant model set included in the initial target value payment method probability generation model according to the initial variable vector set, the actual value reduction information set, and the actual value payment method information set to obtain the discriminant model set comprises: selecting an initial adjustment variable vector from the initial adjustment variable vector set as a first initial adjustment variable vector, and performing the following first training step: inputting the first initial adjustment variable vector into a first initial value reduction information generation model included in the initial target value payment mode probability generation model to generate first initial value reduction information; determining actual value reduction information corresponding to the first initial adjustment variable vector as first actual value reduction information; determining first loss information for the first initial value reduction information and the first actual value reduction information; in response to determining that the first loss information is greater than a first value, determining the first initial value reduction information generation model as a first value reduction information generation model; in response to determining that the first loss information is less than or equal to the first value, performing model parameter updating on the first initial value reduction information generation model according to the first loss information to obtain an updated first value reduction information generation model, and reselecting a first initial adjustment variable vector from the initial adjustment variable vector set, and determining the updated first value reduction information generation model as the first initial value reduction information generation model to continue to perform the first training step.

4. The method of claim 2, wherein, The model training on each initial discrimination model in the initial discrimination model set included in the initial target value payment mode probability generation model according to the initial variable vector set, the actual value reduction information set and the actual value payment mode information set to obtain a discrimination model set includes: selecting an initial confusion variable vector from the initial confusion variable vector set as a first initial confusion variable vector, and performing the following second training step: determining actual value reduction information corresponding to the first initial confusion variable vector as second actual value reduction information, and determining an initial instrumental variable vector corresponding to the first initial confusion variable vector as a first initial instrumental variable vector; generating a first actual value reduction vector for the second actual value reduction information; inputting the first initial confusion variable vector and the first actual value reduction vector into a first initial value payment mode probability generation model to generate first initial value payment mode probability information; inputting the first initial instrumental variable vector, the first initial confusion variable vector and the first actual value reduction vector into a second initial value payment mode probability generation model to generate second initial value payment mode probability information; generating second loss information for the first initial value payment mode probability information and the second initial value payment mode probability information; in response to determining that the second loss information is greater than a second value, determining the first initial value payment mode probability generation model as a first value payment mode probability generation model, and determining the second initial value payment mode probability generation model as a second value payment mode probability generation model; In response to determining that the second loss information is less than or equal to the second numerical value, performing model parameter updating on the first initial value payment mode probability generation model and the second initial value payment mode probability generation model according to the second loss information to obtain an updated first value payment mode probability generation model and an updated second value payment mode probability generation model, and reselecting a first initial confusion variable vector from the initial confusion variable vector set, taking the updated first value payment mode probability generation model as the first initial value payment mode probability generation model, taking the updated second value payment mode probability generation model as the second initial value payment mode probability generation model, and continuing to perform the second training step.

5. The method of claim 2, wherein, The model training on the initial generation model set included in the trained target value payment mode probability generation model and the initial feature data vector representation model according to the initial variable vector set, the actual value reduction information set and the actual value payment mode information set to obtain the target value payment mode probability generation model includes: selecting an initial confusion variable vector from the initial confusion variable vector set as a second initial confusion variable vector, and performing the following third training step: determining the initial adjustment variable vector, the initial tool variable vector corresponding to the second initial confusion variable vector, the corresponding actual value reduction vector and the corresponding actual value reduction information as the second initial adjustment variable vector, the second initial tool variable vector, the third actual value reduction vector and the third actual value reduction information respectively; inputting the second initial confusion variable vector, the third actual value reduction vector and the second initial adjustment variable vector into a third initial value payment mode probability generation model to generate third initial value payment mode probability information; inputting the second initial adjustment variable vector into a fourth initial value payment mode probability generation model to generate fourth initial value payment mode probability information; inputting the second initial tool variable vector into a second initial value reduction information generation model to generate second initial value reduction information; generating third loss information for the third initial value payment mode probability information and the corresponding actual value payment mode information; generating fourth loss information for the fourth initial value payment mode probability information and the corresponding actual value payment mode information; generating fifth loss information for the second initial value reduction information and the third actual value reduction information; in response to determining that the third loss information is less than a third numerical value, the fourth loss information is less than a fourth numerical value, and the fifth loss information is less than a fifth numerical value, determining the third initial value payment mode probability generation model, the fourth initial value payment mode probability generation model, the second initial value reduction information generation model and the initial feature data vector representation model as a third value payment mode probability generation model, a fourth value payment mode probability generation model, a second value reduction information generation model and a feature data vector representation model respectively.

6. The method of claim 1, wherein, The target value payment mode probability generation model performs model verification through the following steps: obtain a user information set; for each value reduction value included in the value reduction information, perform the following information generation steps: determine the value distribution value corresponding to each user information in the user information set; filter a user information subset corresponding to the target value distribution value from the user information set as a first user information subset; filter a user information subset corresponding to the value reduction value from the user information set as a second user information subset; fuse the first user information subset and the second user information subset to obtain a fused user information set; for each fused user information in the fused user information set, use the target value payment mode probability generation model to determine the first target value payment mode prediction probability information of the fused user information under the value reduction value and the second target value payment mode prediction probability information of the fused user information under the target value; group each fused user information in the fused user information set according to the information difference between the first target value payment mode prediction probability information and the second target value payment mode prediction probability information corresponding to each fused user information, to obtain at least one user information group; for each user information group in the at least one user information group, determine the predicted average gain value corresponding to the user information group according to the information difference set corresponding to the user information group; determine the actual average gain value for each user information group to obtain at least one actual average gain value; compare the obtained at least one predicted average gain value and the at least one actual average gain value to obtain comparison information; generate a model verification result for the target value payment mode probability generation model according to the obtained multiple comparison information.

7. A prediction probability information generation method, comprising: obtaining variable data for a target user and a plurality of value reduction values included in value reduction information; for each value reduction value in the plurality of value reduction values, inputting the variable data and the value reduction value into a pre-trained target value payment mode probability generation model to generate target value payment mode prediction probability information for the value reduction value, wherein the target value payment mode probability generation model is generated based on the method of any one of claims 1-6; filtering a value reduction value corresponding to target value payment mode prediction probability information that satisfies a preset information condition from the plurality of value reduction values as a target value reduction value; pushing the reduction information corresponding to the target value reduction value to a user terminal corresponding to the target user.

8. The method of claim 7, wherein, The for each value reduction value in the plurality of value reduction values, inputting the variable data and the value reduction value into a pre-trained target value payment mode probability generation model to generate target value payment mode prediction probability information for the value reduction value, comprises: For each of the plurality of value reduction numerical values, inputting the variable data and the value reduction numerical value into a third value payment mode probability generation model included in the target value payment mode probability generation model to generate target value payment mode prediction probability information for the value reduction numerical value.

9. A model training apparatus, comprising: a first obtaining unit configured to obtain a training data set, wherein training data in the training data set comprises variable data, actual value reduction information, and actual value payment mode information; a first generating unit configured to input a variable data set into an initial feature data vector representation model included in an initial target value payment mode probability generation model to output an initial variable vector set; a first training unit configured to perform model training on each of an initial discriminant model set included in the initial target value payment mode probability generation model according to the initial variable vector set, an actual value reduction information set, and an actual value payment mode information set, to obtain a discriminant model set, wherein the discriminant model set comprises a first value reduction information generation model, a first value payment mode probability generation model, and a second value payment mode probability generation model; a second training unit configured to fix model parameters of the first value reduction information generation model, the first value payment mode probability generation model, and the second value payment mode probability generation model, and perform model training on an initial generative model set included in a trained target value payment mode probability generation model and the initial feature data vector representation model according to the initial variable vector set, the actual value reduction information set, and the actual value payment mode information set, to obtain the target value payment mode probability generation model, wherein the generative model set in the target value payment mode probability generation model comprises a third value payment mode probability generation model, a fourth value payment mode probability generation model, and a second value reduction information generation model.

10. A prediction probability information generation apparatus, comprising: a second obtaining unit configured to obtain variable data and a plurality of value reduction numerical values included in value reduction information for a target user; a second generating unit configured to, for each of the plurality of value reduction numerical values, input the variable data and the value reduction numerical value into a pre-trained target value payment mode probability generation model to generate target value payment mode prediction probability information for the value reduction numerical value, wherein the target value payment mode probability generation model is generated based on the method of any one of claims 1-6; a screening unit configured to screen, from the plurality of value reduction numerical values, a value reduction numerical value corresponding to target value payment mode prediction probability information satisfying a preset information condition as a target value reduction numerical value; a pushing unit configured to push value reduction information corresponding to the target value reduction numerical value to a user terminal corresponding to the target user.

11. An electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement a method as claimed in any of claims 1-8.

12. A computer readable medium having stored thereon a computer program, wherein, The computer program, which when executed by a processor implements a method as claimed in any of claims 1-8.

13. A computer program product comprising a computer program which, when executed by a processor, implements a method according to any of claims 1-8.

Images