Data processing method and related equipment

Abstract

The embodiment of the invention discloses a data processing method, in the method, first intervention data of an intervention variable in a decision service can be utilized to perform causal effect estimation on the decision service based on an anti-fact thought, and anti-fact intervention data obtained based on the anti-fact thought can be a continuous variable. The variable can be a discrete variable, so that the causal effect estimation can be carried out on various types of intervention variables in various decision service scenes. Besides, in the method, the causal effect model can perform causal effect estimation on the input intervention variable in a group corresponding to the input confusion variable so as to obtain an output result about the result variable. Thus, the anti-fact intervention data and the first confusion data are input into the causal effect model, the first estimation result can be obtained through the causal effect model, and for a group described by the first confusion data, accurate causal effect estimation is obtained according to the anti-fact intervention data.

Description

Technical Field

[0001] The present application relates to the field of business decision-making technology, and in particular to a data processing method and related equipment. Background Art

[0002] In decision-making business scenarios such as price decision-making businesses and various recommendation businesses, causal effect estimation is often required to estimate the impact of changes in intervention variables such as price on outcome variables such as sales.

[0003] Currently, the most commonly used method for estimating causal effects is randomized controlled experiments. However, in many business scenarios, randomized controlled experiments are often not possible, and only real-world observational data are available. For example, in pricing, it is often difficult to directly experiment with the impact of different prices on sales.

[0004] Therefore, there is an urgent need for a method to efficiently and accurately estimate causal effects in decision-making operations when randomized controlled experiments are not possible and only real-world observational data are available. Summary of the Invention

[0005] The present invention provides a data processing method that can efficiently and accurately estimate causal effects for decision-making services when randomized controlled experiments are not possible and only real-world observational data are available. The present invention also provides corresponding devices, equipment, computer-readable storage media, and computer program products.

[0006] A first aspect of the present application provides a data processing method, in which a first data set can be obtained, the first data set including first intervention data of an intervention variable and first confounding data of a confounding variable, wherein the intervention variable is a first decision business parameter in a decision business, and the confounding variable is a second decision business parameter in the decision business; counterfactual intervention data corresponding to the first intervention data is obtained; and a first estimation result is obtained through a causal effect model based on the counterfactual intervention data and the first confounding data, wherein the causal effect model is trained based on the confounding data corresponding to the confounding variable, intervention data corresponding to the intervention variable that is not affected by the confounding variable, and result data corresponding to the result variable that is not affected by the confounding variable, and the first estimation result is used to describe the causal effect estimation result of the result variable in the group corresponding to the first confounding data when the intervention variable is the counterfactual intervention data.

[0007] It can be seen that in the first aspect, the observation data of the intervention variables in the decision-making business (first intervention data) can be used to estimate the causal effect of the decision-making business based on the counterfactual idea. The counterfactual intervention data obtained based on the counterfactual idea can be a continuous variable or a discrete variable. Therefore, the causal effect of various types of intervention variables can be estimated in various decision-making business scenarios.

[0008] Furthermore, in the first aspect, during the training process of the causal effect model, confounding variables are reasonably controlled to avoid interference of the confounding variables on the outcome variable and the intervention variable, respectively. Thus, the confounded data of the input confounding variables can be converted into a description of a population with specific characteristics, so that the causal effect model obtained after training can estimate the causal effect of the input intervention variable in the population corresponding to the input confounding variable and obtain an output result regarding the outcome variable. Therefore, the causal effect model can be used to estimate the causal effect of the input intervention variable in the population corresponding to the confounded data of the input confounding variable and obtain an output result regarding the outcome variable.

[0009] Specifically, the counterfactual intervention data and the first confounding data are input into the causal effect model. Based on the output of the causal effect model, a first estimation result is generated. This first estimation result is used to describe the estimated causal effect of the outcome variable on the population corresponding to the first confounding data, when the intervention variable is the counterfactual intervention data. This fully accounts for the heterogeneity of the causal effect for different values ​​of the confounding variable, and for the population described by the first confounding data, a relatively accurate causal effect estimate is obtained based on the counterfactual intervention data. In other words, this causal effect model can be used to perform targeted causal effect estimation for the populations corresponding to different confounding data of the confounding variable, thereby obtaining accurate causal effect estimation results.

[0010] That is to say, through this causal effect model, in a variety of application scenarios, causal effect estimation can be carried out specifically for groups corresponding to different confounding data of confounding variables to obtain accurate causal effect estimation results.

[0011] For example, in a mobile phone pricing decision-making business, changes in mobile phone prices may have different impacts on male and female groups. The first confounded data can be for either male or female. This allows the causal effect model to evaluate the estimated sales volume of mobile phones in the male or female group when the phone price is the assumed price, thereby providing a more accurate causal effect estimate based on the characteristics of different groups.

[0012] In a possible implementation of the first aspect, a first estimation result is obtained through a causal effect model based on counterfactual intervention data and first obfuscated data, including: inputting the counterfactual intervention data and the first obfuscated data into the causal effect model, obtaining a second estimation result output by the causal effect model, and using the second estimation result as the first estimation result.

[0013] In this possible implementation, the confounding variables are reasonably controlled, and the confounded data of the input confounding variables are converted into a description of a group with specific characteristics, so that the causal effect model can estimate the causal effect of the input counterfactual intervention data in the group corresponding to the input first confounded data to obtain an output result about the outcome variable, and this output result can be used as the first estimation result.

[0014] In a possible implementation of the first aspect, the first data set also includes first result data of the result variable, and the result variable is the third decision business parameter in the decision business. The method also includes: inputting the first intervention data and the first confusion data into the causal effect model to obtain a third estimation result output by the causal effect model; obtaining a first estimation result through the causal effect model based on the counterfactual intervention data and the first confusion data, including: inputting the counterfactual intervention data and the first confusion data into the causal effect model to obtain a second estimation result output by the causal effect model; obtaining the first estimation result based on the difference between the second estimation result and the third estimation result, and the first result data.

[0015] In this possible implementation, the third estimation result can be an output result regarding the outcome variable obtained by estimating the causal effect of the first intervention data in the population corresponding to the first confounded data. The second estimation result can be an output result regarding the outcome variable obtained by estimating the causal effect of the counterfactual intervention data corresponding to the first intervention data in the population corresponding to the first confounded data. The difference between the second and third estimation results can be considered the estimated impact of the intervention variable on the outcome variable when the first intervention data is converted into the counterfactual intervention data in the population corresponding to the first confounded data.

[0016] In a possible implementation of the first aspect, a first estimation result is obtained through a causal effect model based on counterfactual intervention data and first confounded data, including: inputting the counterfactual intervention data and the first confounded data into the causal effect model to obtain a second estimation result output by the causal effect model; inputting the first confounded data into the first prediction model to obtain a fourth estimation result output by the first prediction model, where the fourth estimation result is result data of the outcome variable predicted by the first prediction model based on the first confounded data of the confounded variable; and obtaining the first estimation result based on the sum of the second estimation result and the fourth estimation result.

[0017] In this possible implementation, considering that the confounding variables are reasonably controlled during the training process of the causal effect model, the interference of the confounding variables on the outcome variable and the intervention variable is avoided, and thus the confounded data of the input confounding variables can be converted into a description of a group with specific characteristics. Therefore, the causal effect model obtained after training may have difficulty in evaluating the impact of the confounding variables on the outcome variable. In other words, the second estimated result output by the causal effect model mainly contains the result data affected by the counterfactual intervention data, but lacks the result data affected by the first confounded data. Therefore, the first confounded data can be input into the first prediction model to obtain the fourth estimated result output by the first prediction model. The fourth estimated result can be considered as the result data affected by the first confounded data.

[0018] In this way, the sum of the second estimation result and the fourth estimation result can be considered as the complete result data obtained through causal effect estimation under the situation described by the counterfactual intervention data and the first confounding data. Therefore, the first estimation result can be obtained based on the sum of the second estimation result and the fourth estimation result. For example, the sum of the second estimation result and the fourth estimation result can be used as the first estimation result.

[0019] In a possible implementation of the first aspect, before obtaining a first estimation result through a causal effect model based on the counterfactual intervention data and the first confounding data, the method further includes: obtaining a second data set, the second data set including second intervention data of the intervention variable, second confounding data of the confounding variable, and second result data of the result variable; determining third result data that is not affected by the second confounding data from the second result data; determining third intervention data that is not affected by the second confounding data from the second intervention data; and training the causal effect model to be trained using the second confounding data and the third intervention data as inputs and the third result data as labels to obtain a causal effect model.

[0020] In this possible implementation, the second confounded data and the third intervention data are used as inputs, and the third result data is used as a label. When the causal effect model to be trained is trained, since the third result data is not affected by the second confounded data, and the third intervention data is not affected by the second confounded data, the interference of the confounding variable on the intervention variable and the result variable is eliminated.

[0021] In other words, this possible implementation method rationally controls confounding variables, converting the confounded data of the input confounding variables into descriptions of populations with specific characteristics. This allows the trained causal effect model to estimate the causal effect of the input intervention variable within the population corresponding to the input confounding variable, thereby generating output results related to the outcome variable. This possible implementation method fully accounts for the heterogeneity of causal effects, and through this causal effect model, it is possible to evaluate the causal effect estimates of the intervention variable within populations described by different confounding data.

[0022] In a possible implementation of the first aspect, determining third result data that is not affected by the second obfuscated data from the second result data includes: inputting the second obfuscated data into a first prediction model to obtain fourth result data output by the first prediction model, where the fourth result data is result data in the second result data that is affected by the second obfuscated data, and the first prediction model is used to predict result data of the result variable based on the obfuscated data of the confounding variable; and determining a difference between the second result data and the fourth result data as the third result data.

[0023] In this possible implementation, the difference between the second result data and the fourth result data can be determined as the result data in the second result data that cannot be predicted by the second obfuscated data, that is, the third result data that is not affected by the second obfuscated data. Thus, in this possible implementation, the interference of the confounding variable on the result variable can be eliminated by calculating the residual.

[0024] In a possible implementation manner of the first aspect, the method further includes: using the second obfuscated data as input data and the second result data as a label to train the first prediction model to obtain the first prediction model.

[0025] In this possible implementation, the difference between the output of the first prediction model and the corresponding second result data can be evaluated by a loss function, so that the first prediction model obtained after training can infer all the result data in the second result data that are affected by the second obfuscation data (or explained by the second obfuscation data) as the fourth result data based on the input second obfuscation data.

[0026] In a possible implementation of the first aspect, third intervention data that is not affected by the second confounding data is determined from the second intervention data, including: inputting the second confounding data into a second prediction model to obtain fourth intervention data output by the second prediction model, where the fourth intervention data is intervention data in the second intervention data that is affected by the second confounding data, and the second prediction model is used to predict the intervention data of the intervention variable based on the confounding data of the confounding variable; and determining the difference between the second intervention data and the fourth intervention data as the third intervention data.

[0027] In this possible implementation, the difference between the second intervention data and the fourth intervention data may be determined as intervention data in the second intervention data that cannot be predicted by the second obfuscated data, that is, determined as third intervention data that is not affected by the second obfuscated data.

[0028] It can be seen that in this possible implementation method, the interference of confounding variables on intervention variables can be eliminated by calculating residuals.

[0029] In a possible implementation manner of the first aspect, the method further includes: using the second obfuscated data as input data and the second intervention data as labels to train the second prediction model to obtain the second prediction model.

[0030] Traditional causal effect estimation methods (such as inverse probability weighting method, matching method and dual machine learning) all use parameterized estimation methods, and the resulting causal effect estimates are linear.

[0031] In this possible implementation method, the causal effect can be implicitly described by the causal effect model, that is, the causal effect is estimated using a non-parametric method. It can be seen that the causal effect model can be used to estimate nonlinear causal effects and obtain more accurate causal effect estimation results.

[0032] A second aspect of the present application provides a data processing device that has the functionality to implement the method of the first aspect or any possible implementation of the first aspect. This functionality can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the aforementioned functionality, such as an acquisition module and a processing module.

[0033] A third aspect of the present application provides a computing device, which includes a processor, a memory, and computer-executable instructions stored in the memory and executable by the processor. When the computer-executable instructions are executed by the processor, the processor executes the method as described in the first aspect or any possible implementation of the first aspect.

[0034] In a fourth aspect, the present application provides a computing device cluster, which includes at least one computing device, and the at least one computing device includes a processor and a memory. The memory of at least one computing device stores computer execution instructions that can be run on the processor. When the computer execution instructions are executed by the processor, the processor executes the method as described in the first aspect or any possible implementation of the first aspect.

[0035] In a fifth aspect, the present application provides a computer-readable storage medium storing one or more computer-executable instructions. When the computer-executable instructions are executed by a processor, the processor executes a method as described in the first aspect or any possible implementation of the first aspect.

[0036] In a sixth aspect, the present application provides a computer program product that stores one or more computer-executable instructions. When the computer-executable instructions are executed by a processor, the processor executes a method as described in the first aspect or any possible implementation of the first aspect.

[0037] In a seventh aspect, the present application provides a chip system, which includes a processor configured to support the processor in implementing the functions involved in the first aspect or any possible implementation of the first aspect. In one possible design, the chip system may further include a memory configured to store necessary program instructions and data. The chip system may be composed of a chip or may include a chip and other discrete devices.

[0038] Among them, the technical effects brought about by the second to seventh aspects or any possible implementation methods thereof can refer to the technical effects brought about by the first aspect or the relevant possible implementation methods of the first aspect, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is an exemplary schematic diagram of the system architecture provided by the embodiment of the present application;

[0040] Figure 2 This is an exemplary schematic diagram of the data processing stage provided in an embodiment of the present application;

[0041] Figure 3 This is a schematic diagram of an embodiment of a data processing method provided in an embodiment of the present application;

[0042] Figure 4a This is an exemplary schematic diagram of the correlation between the intervention variable, confounding variable and outcome variable provided in the embodiment of the present application;

[0043] Figure 4b is an exemplary schematic diagram of the correlation between the third intervention variable, the second confounding variable, and the third outcome variable provided in the embodiments of the present application;

[0044] Figure 5 This is a schematic diagram of an embodiment of a data processing method provided in an embodiment of the present application;

[0045] Figure 6 1 is a schematic diagram of an embodiment of a data processing device provided in an embodiment of the present application;

[0046] Figure 7is a structural diagram of a computing device provided in an embodiment of the present application;

[0047] Figure 8 This is a schematic diagram of the structure of a computing device cluster provided in an embodiment of the present application;

[0048] Figure 9 This is a structural diagram of a computing device cluster provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] The following describes the embodiments of the present application in conjunction with the accompanying drawings. The terms used in the implementation methods of the present application are only used to explain the specific embodiments of the present application and are not intended to limit the present application.

[0050] Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0051] In this application, "at least one" means one or more, and "more" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable where appropriate. This is merely a way of distinguishing objects with the same properties when describing them in the embodiments of this application. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a list of elements is not necessarily limited to those elements but may include other elements not expressly listed or inherent to such process, method, product, or apparatus.

[0052] Below, some terms involved in the embodiments of this application are first introduced.

[0053] 1. Causation

[0054] Causation refers to the interaction between one event (i.e., the "cause") and a second event (i.e., the "result").

[0055] 2. Causal effect

[0056] The causal effect is the quantity that describes the effect of a change in the treatment variable on the outcome variable. In other words, it describes the change in the outcome variable caused by changing the treatment variable, while controlling for all other variables. Confounders are variables that may affect the outcome.

[0057] Understandably, causal effects study the causal relationship between an intervening variable and an outcome variable. However, in real-world business scenarios, confounding variables can affect both the intervening and outcome variables, leading to correlation and confounding in the causal relationship. This correlation and confounding often compromises the accuracy of the causal relationship.

[0058] 3. Causal Effect Model

[0059] In actual business scenarios, the causal effect model can estimate the causal effect based on the model.

[0060] 4. Counterfactual Reasoning

[0061] Counterfactual reasoning can be understood as reasoning about probabilistic answers to questions such as "What if...", where the conditional clause "What if..." can be an intervening variable in a causal effect.

[0062] For example, in a price decision scenario, the intervention variable is the price of a mobile phone, the confounding variables are the configuration of the mobile phone, the user's gender, the user's income, etc., and the outcome variable is sales volume.

[0063] The current price of a mobile phone is 3,000 yuan. When setting a preferential price for a mobile phone, we hope to estimate the impact of the preferential price separately, that is, "If the price is adjusted to 2,000 yuan while keeping other external environmental factors unchanged, what impact will it have on sales?" An equivalent statement is: "If the price is adjusted to 2,000 yuan while excluding the interference of other relevant factors, what impact will it have?"

[0064] In this causal effect estimation problem, the actual data (or observed data) of the intervention variable is 3,000 yuan, and the counterfactual data of the intervention variable is that the price of the mobile phone is assumed to be adjusted to 2,000 yuan. Then, in the counterfactual reasoning, the reasoning is to infer how much the sales volume of the mobile phone will be if the price of the mobile phone is adjusted to 2,000 yuan, thereby providing a reference for the price decision of the mobile phone in the subsequent sales process.

[0065] 5. Neural Networks

[0066] A neural network can be composed of neural units, which can be represented by x s(i.e. input data) and intercept 1 as input operation unit, the output of the operation unit can be:

[0067]

[0068] Where, s = 1, 2, ... n, n is a natural number greater than 1, W s is x s The weight parameter of the neural unit, b is the bias of the neural unit. f is the activation function of the neural unit, which is used to introduce nonlinear characteristics into the neural network to convert the input signal in the neural unit into the output signal. The output signal of the activation function can be used as the input of the next convolutional layer, and the activation function can be a sigmoid function. A neural network is a network formed by connecting multiple single neural units mentioned above, that is, the output of one neural unit can be the input of another neural unit. The input of each neural unit can be connected to the local receptive field of the previous layer to extract the features of the local receptive field. The local receptive field can be an area composed of several neural units.

[0069] 6. Loss Function

[0070] During neural network training, because we want the output of the neural network to be as close as possible to the desired predicted value, we can compare the current network's predicted value with the desired target value and then update the weight vector of each layer of the neural network based on the difference between the two. (Of course, there is usually an initialization process before the first update, which is to pre-configure the parameters for each layer in the deep neural network.) For example, if the network's prediction value is too high, the weight vector is adjusted to make it predict a lower value. This adjustment is continued until the neural network can predict the desired target value or a value very close to the desired target value. Therefore, it is necessary to predefine "how to compare the difference between the predicted value and the target value." This is the loss function (or objective function), which is an important equation used to measure the difference between the predicted value and the target value. For example, the loss function output value (loss) indicates a greater difference, so neural network training becomes a process of minimizing this loss as much as possible.

[0071] 7. Backpropagation algorithm

[0072] Neural networks can use the back propagation (BP) algorithm to correct the size of the parameters in the initial super-resolution model during training, reducing the reconstruction error loss of the super-resolution model. Specifically, the forward propagation of the input signal to the output generates an error loss. This error loss is then backpropagated to update the parameters of the initial super-resolution model, thereby converging the error loss. The BP algorithm is a backward propagation movement dominated by the error loss, aiming to obtain the optimal super-resolution model parameters, such as the weight matrix.

[0073] In decision-making business scenarios such as price decision-making businesses and various recommendation businesses, causal effect estimation is often required to estimate the impact of changes in intervention variables such as price on outcome variables such as sales.

[0074] Currently, the most commonly used method for estimating causal effects is randomized controlled experiments. However, in many business scenarios, randomized controlled experiments are often not possible, and only real-world observational data are available. For example, in pricing, it is often difficult to directly experiment with the impact of different prices on sales.

[0075] When randomized controlled trials are unavailable and only real-world observational data are available, one current method for estimating causal effects is the inverse probability weighting method. However, this method is typically used to estimate causal effects for binary variables and is difficult to apply to continuous variables. Therefore, it is difficult to apply to a variety of business scenarios and cannot meet the needs of diverse business scenarios.

[0076] Another approach to estimating causal effects is the matching method. This method involves pruning observations to achieve better balance between the control and treatment groups. However, this method results in a loss of sample size; unmatched samples are discarded, thus failing to fully utilize the information in the observations.

[0077] Another data processing method is dual machine learning. Unlike traditional machine learning, dual machine learning is based on methods such as linear regression. Therefore, dual machine learning uses linear models and cannot estimate nonlinear causal effects.

[0078] Based on this, an embodiment of the present application provides a data processing method that can utilize pre-collected business data and other observation data, combined with counterfactual thinking to efficiently and accurately estimate the causal effects in decision-making business, and can meet the needs of various business scenarios such as continuous variables. In addition, in some scenarios, nonlinear causal effects can be estimated through nonlinear causal effect models.

[0079] The following is an exemplary introduction to the data processing method of the embodiment of the present application.

[0080] The data processing method of the embodiment of the present application can be applied to a computing device cluster, which may include one or more computing devices.

[0081] The type of any computing device is not limited herein. For example, any computing device may be a terminal device, or may be a server, server cluster, container, or virtual machine. When a computing device cluster includes multiple computing devices, the types of the different computing devices may be the same or different.

[0082] In the embodiments of the present application, the system architecture of the computing device cluster can have various situations, which are not limited here. For example, the computing device cluster can include one or more servers located in the cloud, or the computing device cluster can also be a local device interacting with a server in the cloud. For another example, the computing device cluster can include a server in the cloud and a local device.

[0083] See also Figure 1 , Figure 1 A schematic diagram of an exemplary system architecture 100 provided in an embodiment of the present application. Figure 1 As shown, in the system architecture 100, the execution device 110 can be implemented by one or more servers. For example, the execution device 110 can be located on a cloud platform. Optionally, the execution device 110 cooperates with other computing devices, such as data storage, routers, load balancers, and other devices; the execution device 110 can be deployed on a single physical site or distributed across multiple physical sites. The execution device 110 can use data in the data storage system 120 (e.g., observation data such as the first data set and the second data set of the decision-making business), or call the program code in the data storage system 120 to implement the data processing method provided in the embodiment of the present application.

[0084] Users can operate their respective user devices (such as local device 101 and local device 102) to interact with execution device 110. Each local device can represent any computing device, such as a personal computer, a computer workstation, a smart phone, a tablet computer, a laptop computer, and a smart car.

[0085] Each user's local device can interact with the execution device 110 through a communication network of any communication mechanism / communication standard. The communication network can be a wide area network, a local area network, a point-to-point connection, etc., or any combination thereof.

[0086] In one implementation, the execution device 110 is used to train a machine learning model to be trained (e.g., one or more of the causal effect model to be trained, the first prediction model to be trained, and the second prediction model to be trained in the embodiment of the present application) to obtain a trained model (e.g., one or more of the causal effect model, the first prediction model, and the second prediction model). Moreover, in the process where the local device 101 and the local device 102 need to use the model to perform reasoning, the execution device 110 processes the business data provided by the user based on the trained model (e.g., one or more of the causal effect model, the first prediction model, and the second prediction model) to perform causal effect estimation, and then returns the corresponding processing results to the local device 101 and the local device 102. In this example, the execution device can be used as a computing device cluster.

[0087] In another implementation, the execution device 110 is used to train a machine learning model to be trained (for example, one or more of the causal effect model to be trained, the first prediction model to be trained, and the second prediction model to be trained in the embodiment of the present application) to obtain a trained model, and send the obtained trained model to the local device 101 and the local device 102. In this way, the local device 101 and the local device 102 can deploy the trained model locally (for example, obtain one or more of the causal effect model, the first prediction model, and the second prediction model), so as to process the business data based on the trained model to perform causal effect estimation and obtain corresponding processing results. In this example, the execution device and the local device can be regarded as different computing devices in a computing device cluster.

[0088] In another implementation, one or more aspects of the execution device 110 can be implemented by each local device. For example, the local device 101 can provide business data or one or more trained models (such as a first prediction model and / or a second prediction model) to the execution device 110, so that the execution device executes the data processing method provided in the embodiment of the present application; or, the local device 101 can also execute the data processing method provided in the embodiment of the present application.

[0089] In general, Figure 1 In the example shown, the data processing method provided in the embodiment of the present application can be applied to the execution device 110, local device 101 and / or local device 102 mentioned above, that is, the collection of one or more of the execution device 110, local device 101 and local device 102 can be a computing device cluster. It should be noted that Figure 1 This is merely an exemplary description of the system architecture of the embodiment of the present application and is not intended to be limiting.

[0090] The data processing method can be applied to decision-making business.

[0091] For example, the decision-making business can be a sales decision-making business, such as a price decision-making business, or a recommendation business, such as a push recommendation business for videos, advertisements, links, official accounts, etc. In different decision-making businesses, the intervention variables, confounding variables, and outcome variables in the data processing method each correspond to different business parameters. Specifically, the intervention variable can be the first decision-making business parameter in the decision-making business, the confounding variable can be the second decision-making business parameter in the decision-making business, and the outcome variable can be the third decision-making business parameter in the decision-making business.

[0092] For example, in a mobile phone pricing decision-making business, the first decision-making business parameter is the phone's price, the second decision-making business parameter is the phone's processor configuration, and the third decision-making business parameter is the phone's sales volume. Therefore, in the causal effect method applied to this pricing decision-making business, the intervening variable is the phone's price, the confounding variable is the phone's processor configuration, and the outcome variable is the phone's sales volume.

[0093] For example, in a social media app's video recommendation service, the first decision-making parameter is the product type shown in the video, the second is the user's gender, and the third is the probability of the user purchasing the corresponding product through the link in the video. Therefore, in the causal effect method applied to this video recommendation service, the intervening variable is the product type, the confounding variable is the user's gender, and the outcome variable is the purchase probability.

[0094] In other scenarios, the decision-making service may also be other types of services, and the corresponding intervening variables, confounding variables, and outcome variables may also be other service parameters. Furthermore, the number of intervening variables, confounding variables, and other variables may also be one or more, and is not limited here. For example, in the aforementioned mobile phone price decision service, the second decision-making service parameter may include the processor configuration and the memory configuration of the mobile phone. Accordingly, the confounding variables may include the processor configuration and the memory configuration of the mobile phone.

[0095] It can be seen that in the embodiment of the present application, the first decision-making service parameter, the second decision-making service parameter and the third decision-making service parameter are different from each other and are determined based on the specific type of the decision-making service and the relevant observation data of the decision-making service.

[0096] Based on the above computing device cluster, such as Figure 2 As shown, the data processing method may include one or more of the following processing stages:

[0097] Train causal effect models and estimate causal effects based on causal effect models.

[0098] The following is an exemplary introduction to each processing stage.

[0099] 1. Train the causal effect model.

[0100] In the embodiment of the present application, subsequent causal effect estimation can be performed through a causal effect model.

[0101] The causal effect model is a machine learning model. For example, the causal effect model can be a neural network or a decision tree. In this way, the causal effect can be implicitly described by the causal effect model.

[0102] The causal effect model is used to estimate the causal effect of the intervention variable under the condition described by the confounding variable to obtain the output results about the outcome variable.

[0103] The causal effect model can be obtained by training based on the confounding data corresponding to the confounding variable, the intervention data corresponding to the intervention variable that are not affected by the confounding variable, and the result data corresponding to the result variable that are not affected by the confounding variable.

[0104] Among them, the confounded data corresponding to the confounding variables used to train the causal effect model, the intervention data corresponding to the intervention variables that are not affected by the confounding variables, and the result data corresponding to the result variables that are not affected by the confounding variables can be extracted from the business data through machine learning models or linear models such as regression or difference models.

[0105] In the embodiment of the present application, the method for obtaining the causal effect model is not limited here. For example, the causal effect model can be obtained and provided by a third party through training based on business data, or it can be obtained through training by the above-mentioned computing device cluster.

[0106] The following uses the training scenario of the causal effect model in the above-mentioned computing device cluster as an example to exemplify the training process of the causal effect model.

[0107] like Figure 3 As shown, in some embodiments, the method includes steps 301-304.

[0108] Step 301: Acquire a second data set.

[0109] The second data set includes second intervention data for the intervention variable, second confounding data for the confounding variable, and second outcome data for the outcome variable.

[0110] In an embodiment of the present application, one or more second data sets can be obtained from pre-collected relevant business data of the decision-making business (which can also be considered as observation data of the decision-making business). Wherein, each second data set contains the same variable type, but the data corresponding to the variable is different.

[0111] The following describes the content of a second data set and the training process of a causal effect model to be trained based on the second data set, using a second data set as an example. In actual application scenarios, multiple second data sets can be used to train the causal effect model based on the content of the second data set and the corresponding application.

[0112] The second data set may include second intervention data of the intervention variable, second confounding data of the confounding variable, and second result data of the result variable.

[0113] The second data set can be considered to be derived from business data. For example, in a mobile phone pricing decision-making business, the prices of existing mobile phones can be collected in advance to serve as the second intervening data for the intervening variable in the second data set. The confounding variable is determined to be gender, and the second confounding data is female. The outcome variable can be the sales volume of the released mobile phones. In this case, the second data set reflects the sales volume of the released mobile phones among females.

[0114] The source of the second data set can be various, and is not limited in the embodiments of the present application. For example, the second data set can be pre-collected by the user based on the user's related business and stored in the computing device cluster; or it can be obtained from a third party.

[0115] Step 302: Determine third result data from the second result data that is not affected by the second obfuscated data.

[0116] In the embodiment of the present application, third result data that is not affected by the second obfuscated data can be extracted from the second result data. In other words, the third result data can be considered as result data that is not interpreted by the second obfuscated data.

[0117] Generally speaking, the third result data can be affected only by the second intervention data, or in other words, the third result data can be explained by the second intervention data. In other words, the third result data can be considered as the result data obtained by inferring the causal effect through the second intervention data. In this way, the causal effect between the second intervention data and the third result data can be obtained.

[0118] In some examples, the third result data can be extracted from the second result data based on the second obfuscated data using a machine learning model.

[0119] For example, in some embodiments, step 302 includes:

[0120] Inputting the second obfuscated data into the first prediction model to obtain fourth result data output by the first prediction model, where the fourth result data is result data in the second result data that is affected by the second obfuscated data, and the first prediction model is used to predict the result data of the outcome variable based on the obfuscated data of the obfuscated variable;

[0121] A difference between the second result data and the fourth result data is determined as third result data.

[0122] In an embodiment of the present application, the first prediction model may be a machine learning model, for example, a neural network model or a decision tree such as a random forest.

[0123] The first prediction model is used to predict the result data of the result variable based on the confusion data of the confusion variable. In other words, when the input of the first prediction model is the second confusion data, the first prediction model can output the result data obtained based on the second confusion data, that is, obtain the fourth result data affected by the second confusion data.

[0124] In this way, the difference between the second result data and the fourth result data can be determined as the result data in the second result data that cannot be predicted by the second obfuscated data, that is, determined as the third result data not affected by the second obfuscated data.

[0125] It can be seen that in the embodiment of the present application, the interference of confounding variables on the outcome variables can be eliminated by calculating the residual method.

[0126] The first prediction model may be obtained by training on other devices and deployed to a computing device cluster that executes the embodiments of the present application, or may be obtained by training on the computing device cluster.

[0127] In one example, the first prediction model to be trained can be trained through multiple iterative processes based on one or more second data sets, and the first prediction model can be obtained after the training is completed.

[0128] Specifically, in some embodiments, the method further includes:

[0129] The second confused data is used as input data, and the second result data is used as a label to train the first prediction model to obtain a first prediction model.

[0130] In an embodiment of the present application, after completing the training of the first prediction model to be trained, the corresponding trained model can be used as the first prediction model and deployed to a computing device cluster.

[0131] During the training process, the first prediction model to be trained can be trained through backpropagation based on the second obfuscated data and second result data in one or more second data sets until the number of iterations reaches a specified threshold, or until the first prediction model to be trained converges to a desired state. If there are multiple second data sets, the second obfuscated data from each of the multiple second data sets can be batch-inputted into the first prediction model to be trained; alternatively, the second obfuscated data from different second data sets can be separately input into the first prediction model to be trained.

[0132] During the training process, the difference between the output of the first prediction model and the corresponding second result data can be evaluated by a loss function, so that the first prediction model obtained after training can infer all the result data in the second result data that are affected by the second confusion data (or explained by the second confusion data) as the fourth result data based on the input second confusion data.

[0133] The type of the loss function is not limited here. For example, the loss function can be cross entropy loss, etc.

[0134] Step 303: Determine third intervention data from the second intervention data that is not affected by the second obfuscated data.

[0135] In the embodiment of the present application, third intervention data that is not affected by the second obfuscated data can be extracted from the second intervention data. That is, the third intervention data can be considered as intervention data that is not interpreted by the second obfuscated data.

[0136] In some examples, third intervention data can be inferred from the second intervention data based on the second obfuscated data using a machine learning model.

[0137] In some embodiments, step 303 includes:

[0138] inputting the second confounded data into the second prediction model to obtain fourth intervention data output by the second prediction model, where the fourth intervention data is intervention data in the second intervention data that is affected by the second confounded data, and the second prediction model is used to predict the intervention data of the intervention variable based on the confounded data of the confounding variable;

[0139] A difference between the second intervention data and the fourth intervention data is determined as third intervention data.

[0140] In the embodiment of the present application, the second prediction model can be a machine learning model, for example, a neural network model or a decision tree such as a random forest. The structure of the second prediction model can be the same as or different from the structure of the first prediction model.

[0141] The second prediction model is used to predict the intervention data of the intervention variable based on the confounding data of the confounding variable. In other words, when the input of the second prediction model is the second confounding data, the second prediction model can output the intervention data obtained based on the second confounding data, that is, obtain the fourth intervention data affected by the second confounding data.

[0142] In the embodiment of the present application, the second obfuscated data may be input into the second prediction model to obtain fourth intervention data output by the second prediction model.

[0143] In this way, the difference between the second intervention data and the fourth intervention data can be determined as the intervention data in the second intervention data that cannot be predicted by the second obfuscated data, that is, determined as the third intervention data not affected by the second obfuscated data.

[0144] It can be seen that in the embodiment of the present application, the interference of confounding variables on intervention variables can be eliminated by calculating the residual method.

[0145] The second prediction model may be obtained by training on other devices and deployed to a computing device cluster that executes the embodiments of the present application, or may be obtained by training on the computing device cluster.

[0146] In one example, the second prediction model to be trained may be trained using one or more second data sets, and after the training is completed, the second prediction model is obtained.

[0147] Specifically, in some embodiments, the method further includes:

[0148] The second confused data is used as input data, and the second intervention data is used as a label to train the second prediction model to obtain a second prediction model.

[0149] In the embodiment of the present application, after completing the training of the second prediction model to be trained, the corresponding trained model can be used as the second prediction model and deployed in the computing device cluster.

[0150] During the training process, the second prediction model to be trained can be trained by back propagation based on the second confusion data and second intervention data in one or more second data sets until the number of iterations reaches a specified threshold, or until the second prediction model to be trained converges to the desired state.

[0151] In which, the difference between the output of the second prediction model and the corresponding second intervention data can be evaluated by a loss function, so that the second prediction model obtained after training can infer all intervention data in the second intervention data that are affected by the second confusion data (or explained by the second confusion data) as the fourth intervention data based on the input second confusion data.

[0152] The type of the loss function is not limited here. For example, the loss function can be cross entropy loss, etc.

[0153] Step 304 : Using the second confused data and the third intervention data as input and the third result data as a label, the causal effect model to be trained is trained to obtain a causal effect model.

[0154] In an embodiment of the present application, after obtaining the third intervention data and the third result data, it can be considered that the interference of the confounding variable on the intervention variable and the result variable, respectively, has been eliminated, thereby facilitating the causal effect model to learn the influence of the intervention variable on the result variable through the third intervention data and the third result data while eliminating the interference caused by the confounding variable.

[0155] For example, based on the influence between various business parameters in the business data and related decision-making business, the intervention variable T, the confounding variable X and the result variable Y can be determined from the various business parameters, and the relationship between the various variables can be determined.

[0156] like Figure 4a As shown, it is the correlation between the various variables involved in the current business data.

[0157] Where is the intervention variable, which is the object of study that requires intervention, such as price in a pricing decision-making business; X is the confounding variable, which affects both the intervention variable T and the outcome variable Y. Therefore, in actual business scenarios, confounding variables will interfere with the estimation of causal effects, making it impossible to quantitatively estimate the impact of changes in the intervention variable on the outcome variable of interest.

[0158] like Figure 4b As shown, it is the correlation relationship between the second obfuscated data, the third intervention data and the third result data in the embodiment of the present application.

[0159] It can be seen that in the embodiment of the present application, the third result data is not affected by the second obfuscated data, and the third intervention data is not affected by the second obfuscated data, thereby eliminating the interference of the obfuscating variable on the intervention variable and the result variable respectively.

[0160] In this way, in the embodiment of the present application, the second confounding data and the third intervention data are used as input, and the third result data is used as a label, so that the causal effect model can learn to estimate the causal effect of the intervention variable under the conditions described by the confounding variable and obtain the output result about the result variable. That is to say, in the embodiment of the present application, the corresponding causal effect estimate can be given for the group with specific characteristics described by the confounding variable.

[0161] Among them, the difference between the output of the causal effect model to be trained during the training process and the corresponding third result data can be evaluated by a loss function, so that the causal effect model obtained after training can infer the result data affected by the input intervention data based on the input intervention data in the population described by the characteristics of the input confounding data to obtain a causal effect estimation result, and the inference process of the causal effect estimation result reasonably controls the confounding variables.

[0162] Traditional causal effect estimation methods (such as the inverse probability weighting method, matching method, and dual machine learning) all use parameterized estimation methods, and the resulting causal effect estimates are linear. In the embodiments of the present application, the causal effect model can be used to implicitly describe the causal effect, that is, a non-parametric method is used to estimate the causal effect. It can be seen that the embodiments of the present application can use the causal effect model to estimate nonlinear causal effects and obtain more accurate causal effect estimation results.

[0163] In addition, in an embodiment of the present application, the second confounded data and the third intervention data are used as inputs, and the third result data is used as a label. When the causal effect model to be trained is trained, since the third result data is not affected by the second confounded data, and the third intervention data is not affected by the second confounded data, the interference of the confounding variable on the intervention variable and the result variable is eliminated.

[0164] That is, in the embodiments of the present application, confounding variables are properly controlled, and the confounded data of the input confounding variables are converted into descriptions of populations with specific characteristics, so that the causal effect model obtained after training can estimate the causal effect of the input intervention variable in the population corresponding to the input confounding variable and obtain an output result regarding the outcome variable. It can be seen that in the embodiments of the present application, the heterogeneity of causal effects is fully considered. Through this causal effect model, the causal effect estimation results of the intervention variable in the population described by different confounding data can be evaluated.

[0165] For example, in the mobile phone decision-making business, the impact of mobile phone prices on mobile phone sales may be different among different gender groups.

[0166] Then, in this example, the confounding variable can be gender. In this way, the causal effect model can be used to evaluate the causal effect based on the selling price of mobile phones, targeting the characteristics of groups of different genders, to obtain the causal effect estimation results on mobile phone sales. Specifically, in the input data of the causal effect model, the confounding data corresponding to the confounding variable can indicate females, and the intervention data corresponding to the intervention variable can indicate the assumed selling price of mobile phones. Then the output result of the causal effect model can indicate the estimated sales volume of mobile phones in the female group when the selling price of mobile phones is the assumed selling price of mobile phones. Similarly, the causal effect model can be used to evaluate the estimated sales volume of mobile phones in the male group when the selling price of mobile phones is the assumed selling price of mobile phones, so as to provide correspondingly more accurate causal effect estimation results based on the characteristics of different groups.

[0167] After the causal effect model is obtained through training, the causal effect model can be deployed to a computing device cluster.

[0168] 2. Estimating causal effects based on causal effect models

[0169] Specifically, if Figure 5 As shown, in some embodiments, the method includes steps 501-503.

[0170] Step 501: Acquire a first data set.

[0171] The first data set includes first intervention data of an intervention variable and first confounding data of a confounding variable, wherein the intervention variable is a first decision-making business parameter and the confounding variable is a second decision-making business parameter.

[0172] The first intervention data in the first data set may be the same as or similar to the second intervention data, and reference may be made to the relevant description of the second intervention data. The first obfuscated data may be the same as or similar to the second obfuscated data, and reference may be made to the relevant description of the second obfuscated data.

[0173] In an embodiment of the present application, the first data set can be obtained from pre-collected business data. The first data set can be obtained from one or more second data sets used to train the causal effect model. For example, the first intervention data can be a certain second intervention data, and the first obfuscated data can be the second obfuscated data corresponding to the second intervention data. Alternatively, the first data set can be a data set different from the one or more second data sets.

[0174] Step 502: Obtain counterfactual intervention data corresponding to the first intervention data.

[0175] The counterfactual intervention data may be input by the user into the computing device cluster, for example, a hypothetical value or option of a first decision-making service parameter that the user is interested in. For example, in a mobile phone price decision service, the counterfactual intervention data may be a hypothetical mobile phone price that is different from the current price of the mobile phone and that the user is interested in.

[0176] Step 503: Obtain a first estimation result through a causal effect model based on the counterfactual intervention data and the first confounding data.

[0177] The causal effect model is obtained by training based on the confounded data corresponding to the confounding variable, the intervention data corresponding to the intervention variable that are not affected by the confounding variable, and the result data corresponding to the result variable that are not affected by the confounding variable.

[0178] It can be seen that during the training process of the causal effect model, the confounding variables are reasonably controlled to avoid interference of the confounding variables on the outcome variable and the intervention variable. Therefore, the confounded data of the input confounding variable can be converted into a description of a population with specific characteristics, so that the causal effect model obtained after training can estimate the causal effect of the input intervention variable in the population corresponding to the input confounding variable and obtain an output result regarding the outcome variable. Therefore, the causal effect model can be used to estimate the causal effect of the input intervention variable in the population corresponding to the confounded data of the input confounding variable and obtain an output result regarding the outcome variable. The first estimation result is used to describe the causal effect estimation result on the outcome variable in the population corresponding to the first confounding data when the intervention variable is counterfactual intervention data.

[0179] In the embodiment of the present application, the group corresponding to the first obfuscated data may also be referred to as the group described by the first obfuscated data.

[0180] The group can be a user group. For example, in a video recommendation task, the group can be a user group among users of the video application that meets the characteristics described by the first obfuscation data; or, the group can also be a collection of commodities. For example, in the price decision-making business of a mobile phone, the first obfuscation data is used to describe the hardware parameters of the mobile phone. The group described by the first obfuscation data is the collection of mobile phones whose hardware parameters are the first obfuscation data.

[0181] The first estimation result may have various specific forms.

[0182] For example, in some examples, the first estimation result may be in the form of result data.

[0183] In another example, the first estimation result may be in the form of a ratio, for example, the first estimation result may be Y1' / T1', or may be Y1' / (T1', X1).

[0184] The numerator can be the result data Y1', which is the result data obtained by estimating the causal effect in the population corresponding to the first confounded data when the intervention variable is the counterfactual intervention data. The denominator contains the counterfactual intervention data T1' of the first intervention data and may also include the first confounded data X1.

[0185] In this way, through the information of the numerator and denominator, relevant personnel can efficiently and quickly identify the specific circumstances in the decision-making business scenario (i.e., the intervention data and confounding data involved), as well as the estimated results obtained through causal effect estimation (i.e., the result data obtained through causal effect estimation).

[0186] In an embodiment of the present application, the observation data of the intervention variables in the decision-making business (first intervention data) can be used to estimate the causal effect of the decision-making business based on the counterfactual idea. The counterfactual intervention data obtained based on the counterfactual idea can be a continuous variable or a discrete variable. Therefore, causal effect estimation can be performed on various types of intervention variables in various decision-making business scenarios.

[0187] Furthermore, in an embodiment of the present application, during the training process of the causal effect model, the confounding variables are reasonably controlled to avoid the interference of the confounding variables on the outcome variable and the intervention variable, respectively. Therefore, the confounded data of the input confounding variable can be converted into a description of a population with specific characteristics, so that the causal effect model obtained after training can estimate the causal effect of the input intervention variable in the population corresponding to the input confounding variable and obtain an output result about the outcome variable. Therefore, the causal effect model can be used to estimate the causal effect of the intervention data of the input intervention variable in the population corresponding to the confounded data of the input confounding variable and obtain an output result about the outcome variable. The first estimation result is used to describe the causal effect estimation result of the outcome variable in the population corresponding to the first confounding data when the intervention variable is counterfactual intervention data.

[0188] Specifically, the counterfactual intervention data and the first confounding data are input into the causal effect model. Based on the output of the causal effect model, a first estimation result is generated. This first estimation result is used to describe the estimated causal effect of the outcome variable on the population corresponding to the first confounding data, when the intervention variable is the counterfactual intervention data. This fully accounts for the heterogeneity of the causal effect for different values ​​of the confounding variable, and for the population described by the first confounding data, a relatively accurate causal effect estimate is obtained based on the counterfactual intervention data. In other words, this causal effect model can be used to perform targeted causal effect estimation for the populations corresponding to different confounding data of the confounding variable, thereby obtaining accurate causal effect estimation results.

[0189] For example, in a mobile phone pricing decision-making business, changes in mobile phone prices may have different impacts on male and female groups. The first confounded data can be for either male or female. This allows the causal effect model to evaluate the estimated sales volume of mobile phones in the male or female group when the phone price is the assumed price, thereby providing a more accurate causal effect estimate based on the characteristics of different groups.

[0190] Among them, according to the counterfactual intervention data and the first confounding data, there may be one or more specific ways to obtain the first estimation result through the causal effect model. The following is an illustrative introduction to various possible situations.

[0191] 1) Take the output of the causal effect model as the first estimation result.

[0192] Specifically, in some embodiments, the above step 503 includes:

[0193] The counterfactual intervention data and the first confounded data are input into the causal effect model to obtain a second estimation result output by the causal effect model, and the second estimation result is used as the first estimation result.

[0194] In an embodiment of the present application, the confounding variables are reasonably controlled, and the confounded data of the input confounding variables are converted into a description of a group with specific characteristics, so that the causal effect model can estimate the causal effect of the input counterfactual intervention data in the group corresponding to the input first confounded data to obtain an output result about the outcome variable, and the output result can be used as the first estimation result.

[0195] 2) Obtaining a first estimation result according to an output result of the causal effect model and first result data of the result variable included in the first data set.

[0196] Specifically, in some embodiments, the first data set further includes first result data of a result variable, where the result variable is a third decision-making business parameter;

[0197] The method further includes:

[0198] inputting the first intervention data and the first confounded data into the causal effect model to obtain a third estimation result output by the causal effect model;

[0199] The above step 503 includes:

[0200] Inputting the counterfactual intervention data and the first confounded data into the causal effect model to obtain a second estimation result output by the causal effect model;

[0201] A first estimation result is obtained according to a difference between the second estimation result and the third estimation result, and the first result data.

[0202] In an embodiment of the present application, the step of inputting the first intervention data and the first obfuscated data into the causal effect model to obtain the third estimated result output by the causal effect model may be performed before or after the step of inputting the counterfactual intervention data and the first obfuscated data into the causal effect model to obtain the second estimated result output by the causal effect model. In one example, the step of inputting the first intervention data and the first obfuscated data into the causal effect model to obtain the third estimated result output by the causal effect model may be performed in advance. During the application process, the user may input and obtain multiple counterfactual intervention data corresponding to the first intervention data. However, in the process of obtaining the first estimated result corresponding to each counterfactual intervention data according to an embodiment of the present application, there is no need to repeatedly execute the step of inputting the first intervention data and the first obfuscated data into the causal effect model to obtain the third estimated result output by the causal effect model, thereby improving data processing efficiency and reducing resource consumption.

[0203] The third estimation result may be an output result on the outcome variable obtained by performing causal effect estimation on the first intervention data in the group corresponding to the first confounded data.

[0204] The second estimation result may be an output result on the outcome variable obtained by performing causal effect estimation on the counterfactual intervention data corresponding to the first intervention data in the group corresponding to the first confounded data.

[0205] It can be seen that the difference between the second estimation result and the third estimation result can be considered as the estimated impact of the intervention variable on the outcome variable when the intervention variable is converted from the first intervention data to the counterfactual intervention data in the group corresponding to the first confounded data.

[0206] For example, in a mobile phone pricing decision-making business, the first confounding data is female, the first intervention data is the current price of the phone at 3,000 yuan, and the counterfactual intervention data is a hypothetical price of 2,800 yuan. Therefore, the third estimate is the causal effect of the current price of 3,000 yuan on the sales volume of the phone among women, while the second estimate is the causal effect of the phone's hypothetical price of 2,800 yuan on the sales volume of the phone among women.

[0207] In this way, the difference can be added to the first estimation result to obtain the estimated result of the causal effect of the change of the intervention variable on the outcome variable in the group corresponding to the first confounded data when the intervention variable is counterfactual intervention data, which is also the first estimation result.

[0208] For example, the third estimate is the causal effect of the current price of a phone, 3,000 yuan, on sales among women; the second estimate is the causal effect of the phone's price, assuming it is 2,800 yuan, on sales among women. The difference between the second and third estimates describes the impact of the price change on sales among women when the price of the phone is adjusted from 3,000 yuan to 2,800 yuan. Adding this difference to the first estimate yields the first estimate, which is the estimated sales among women when the price is 2,800 yuan.

[0209] 3) Obtaining a first estimation result based on the output result of the causal effect model and the output result of the first prediction model.

[0210] Specifically, in some embodiments, the above step 503 includes:

[0211] Inputting the counterfactual intervention data and the first confounded data into the causal effect model to obtain a second estimation result output by the causal effect model;

[0212] inputting the first confounded data into a first prediction model to obtain a fourth estimation result output by the first prediction model, the first prediction model being used to predict result data of the outcome variable based on the confounded data of the confounding variable;

[0213] A first estimation result is obtained according to the sum of the second estimation result and the fourth estimation result.

[0214] In the embodiment of the present application, considering that the confounding variables are reasonably controlled during the training process of the causal effect model, the interference of the confounding variables on the outcome variables and the intervention variables is avoided, and thus the confounded data of the input confounding variables can be converted into a description of a group with specific characteristics, therefore, the causal effect model obtained after training may be difficult to evaluate the impact of the confounding variables on the outcome variables. In other words, the second estimation result output by the causal effect model mainly contains the result data affected by the counterfactual intervention data, but lacks the result data affected by the first confounding data. Therefore, the first confounding data can be input into the first prediction model to obtain the fourth estimation result output by the first prediction model, and the fourth estimation result can be considered as the result data affected by the first confounding data.

[0215] In this way, the sum of the second estimation result and the fourth estimation result can be considered as the complete result data obtained through causal effect estimation under the situation described by the counterfactual intervention data and the first confounding data.

[0216] Therefore, the first estimation result can be obtained according to the sum of the second estimation result and the fourth estimation result.

[0217] For example, if the first estimation result is presented as a ratio of Y1' / (T1', X1), the numerator Y1' can be the sum of the second estimation result and the fourth estimation result, and the denominator contains the counterfactual intervention data T1' of the first intervention data and the first confused data X1.

[0218] Furthermore, in some cases, after obtaining the first estimation result, it can be compared with the current business situation to facilitate decision-making on the business.

[0219] For example, the first data set also includes first result data of a result variable, where the result variable is a third decision-making business parameter;

[0220] After obtaining the first estimation result, it also includes:

[0221] The difference between the first estimation result and the first result data is calculated to obtain a fifth estimation result. The fifth estimation result is used to describe the impact of the intervention variable on the outcome variable when the intervention variable is converted from the first intervention data to the counterfactual intervention data in the group corresponding to the first confounded data. That is to say, it describes the change of the outcome variable estimated when the intervention variable is converted from the first intervention data to the counterfactual intervention data in the group corresponding to the first confounded data.

[0222] The specific form of the fifth estimation result is not limited.

[0223] In an example, the fifth estimation result may be result data estimated when the intervention variable is converted from the first intervention data to the counterfactual intervention data in the group corresponding to the first confounded data.

[0224] In another example, the fifth estimation result can be (Y1'-Y1) / (T1'-X1), where the numerator is the difference between the second estimation result Y1' and the first result data Y1, and the denominator is the difference between the counterfactual intervention data T1' of the first intervention data and the first confused data X1.

[0225] The following describes an exemplary process of training the causal effect model and estimating the causal effect through the causal effect model through an example.

[0226] This exemplary process may include the following steps:

[0227] 1) Acquire multiple second data sets from the observation data of the decision-making business, each second data set including relevant data corresponding to the intervention variable T, the confusion variable X, and the result variable Y, which are respectively second intervention data, second confusion data, and second result data.

[0228] 2) The first prediction model to be trained is trained using the second confused data as input data and the second result data as labels, and a first prediction model My is obtained after the training is completed.

[0229] 3) Using the second obfuscated data as input data and the second intervention data as labels, the second prediction model to be trained is trained, and a second prediction model MT is obtained after the training is completed.

[0230] 4) Input the second confused data as a feature into the first prediction model My to obtain a predicted value Ypredict of the result variable as the fourth result data.

[0231] 5) The second confused data is input as a feature into the second prediction model MT to obtain a predicted value Tpredict of the intervention variable as the fourth intervention data.

[0232] 6) Subtract the second result data from Ypredict to obtain the residual Y', which is the third result data in the second result data that is not affected by the confounding variable, eliminating the interference of the confounding variable.

[0233] 7) Subtract the second intervention data from Tpredict to obtain the residual T', which is the third intervention data in the second intervention data that is not affected by the confounding variable, eliminating the interference of the confounding variable.

[0234] 8) Using the second intervention data and the residual T' as input features and the residual Y' as a label to train the causal effect model to be trained, and obtaining the causal effect model τ after the training is completed.

[0235] The causal effect model τ can describe the causal effect implicitly, that is, non-parametrically. Specifically, it can estimate the causal effect of the input intervention data on the outcome variable under the conditions described by the input intervention data (such as the second intervention data) (that is, in the population described by the input intervention data).

[0236] 9) In an actual decision-making business, a first data set related to the decision-making business may be obtained. The first data set includes first intervention data T1, first confusion data X1, and first result data Y1.

[0237] 10) Obtain counterfactual intervention data T1′ of the first intervention data.

[0238] The counterfactual intervention data T1 ′ and the first confused data X1 are input into the causal effect model τ to obtain a second estimation result Y1 ′ output by the causal effect model τ.

[0239] 11) Input the first confused data X1 into the first prediction model My to obtain a fourth estimation result Ypredict1 output by the first prediction model.

[0240] 12) The sum of the second estimation result Y1′ and the fourth estimation result Ypredict1 is the causal effect estimation result on the outcome variable in the population corresponding to the first confounded data X1 when the intervention variable is the counterfactual intervention data T1′.

[0241] 13) Describe the fifth estimation result using the ratio (Y1'-Y1) / (T1'-T1), where the numerator is the difference between the second estimation result Y1' and the first outcome data Y1, and the denominator is the difference between the counterfactual intervention data T1' of the first intervention data and the first obfuscated data X1. Thus, using the information in the numerator and denominator of this ratio, relevant personnel can efficiently and quickly identify the specific circumstances of the decision-making business scenario (i.e., the intervention data and obfuscated data involved), as well as the estimation result obtained through causal effect estimation (i.e., the outcome data obtained through causal effect estimation), and intuitively understand the impact of changes in the intervention variable on the outcome variable.

[0242] The data processing method provided in the embodiments of the present application has been described above from multiple aspects. The data processing device provided in the embodiments of the present application will be described below with reference to the accompanying drawings.

[0243] like Figure 6 As shown, an embodiment of the present application provides a data processing device 60, which includes:

[0244] The acquisition module 601 is used to:

[0245] Acquire a first data set, the first data set including first intervention data of an intervention variable and first confounding data of a confounding variable, wherein the intervention variable is a first decision-making business parameter in the decision-making business, and the confounding variable is a second decision-making business parameter in the decision-making business;

[0246] Obtaining counterfactual intervention data corresponding to the first intervention data;

[0247] The processing module 602 is configured to:

[0248] According to the counterfactual intervention data and the first confounding data, a first estimation result is obtained through a causal effect model, wherein the causal effect model is trained based on the confounding data corresponding to the confounding variable, the intervention data corresponding to the intervention variable that are not affected by the confounding variable, and the result data corresponding to the result variable that are not affected by the confounding variable. The first estimation result is used to describe the causal effect estimation result on the outcome variable in the group corresponding to the first confounding data when the intervention variable is the counterfactual intervention data.

[0249] Optionally, the processing module 602 is used to: input the counterfactual intervention data and the first confounding data into the causal effect model, obtain a second estimation result output by the causal effect model, and use the second estimation result as the first estimation result.

[0250] Optionally, the first data set further includes first result data of a result variable, where the result variable is a third decision-making service parameter in the decision-making service. The processing module 602 is configured to:

[0251] inputting the first intervention data and the first confounded data into the causal effect model to obtain a third estimation result output by the causal effect model;

[0252] Inputting the counterfactual intervention data and the first confounded data into the causal effect model to obtain a second estimation result output by the causal effect model;

[0253] A first estimation result is obtained according to a difference between the second estimation result and the third estimation result, and the first result data.

[0254] Optionally, the processing module 602 is configured to:

[0255] Inputting the counterfactual intervention data and the first confounded data into the causal effect model to obtain a second estimation result output by the causal effect model;

[0256] Inputting the first confounded data into the first prediction model to obtain a fourth estimation result output by the first prediction model, where the fourth estimation result is result data of the outcome variable predicted by the first prediction model based on the first confounded data of the confounding variable;

[0257] A first estimation result is obtained according to the sum of the second estimation result and the fourth estimation result.

[0258] Optionally, the acquisition module 601 is configured to: acquire a second data set, the second data set including second intervention data of the intervention variable, second confounding data of the confounding variable, and second result data of the result variable;

[0259] The processing module 602 is used to:

[0260] Determining, from the second result data, third result data that is not affected by the second obfuscated data;

[0261] determining, from the second intervention data, third intervention data that is not affected by the second confounding data;

[0262] The second confused data and the third intervention data are taken as input, and the third result data is taken as a label, and the causal effect model to be trained is trained to obtain a causal effect model.

[0263] Optionally, the processing module 602 is configured to:

[0264] Inputting the second obfuscated data into the first prediction model to obtain fourth result data output by the first prediction model, where the fourth result data is result data in the second result data that is affected by the second obfuscated data, and the first prediction model is used to predict the result data of the outcome variable based on the obfuscated data of the obfuscated variable;

[0265] A difference between the second result data and the fourth result data is determined as third result data.

[0266] Optionally, the processing module 602 is configured to: use the second obfuscated data as input data and the second result data as labels to train the first prediction model to be trained, so as to obtain the first prediction model.

[0267] Optionally, the processing module 602 is configured to:

[0268] inputting the second confounded data into the second prediction model to obtain fourth intervention data output by the second prediction model, where the fourth intervention data is intervention data in the second intervention data that is affected by the second confounded data, and the second prediction model is used to predict the intervention data of the intervention variable based on the confounded data of the confounding variable;

[0269] A difference between the second intervention data and the fourth intervention data is determined as third intervention data.

[0270] Optionally, the processing module 602 is configured to: use the second obfuscated data as input data and the second intervention data as labels to train the second prediction model to be trained, so as to obtain the second prediction model.

[0271] Optionally, the causal effect model is a machine learning model.

[0272] The acquisition module and the processing module can be implemented by software or hardware. For example, the implementation of the acquisition module will be described below using the acquisition module as an example. Similarly, the implementation of the processing module can refer to the implementation of the acquisition module.

[0273] As an example of a software functional unit, the module acquisition module may include code running on a computing instance. The computing instance may include at least one of a physical host (computing device), a virtual machine, and a container. Furthermore, the computing instance may be one or more. For example, the acquisition module may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code may be distributed in the same region or in different regions. Furthermore, the multiple hosts / virtual machines / containers used to run the code may be distributed in the same availability zone (AZ) or in different AZs, each AZ including one data center or multiple geographically close data centers. Typically, a region may include multiple AZs.

[0274] Similarly, multiple hosts / virtual machines / containers running the code can be distributed within the same virtual private cloud (VPC) or across multiple VPCs. Typically, a VPC is set up within a region. Cross-region communication between two VPCs within the same region, or between VPCs in different regions, requires a communication gateway within each VPC to interconnect the VPCs.

[0275] As an example of a hardware functional unit, the acquisition module may include at least one computing device, such as a server. Alternatively, the acquisition module may be implemented using a central processing unit (CPU), an application-specific integrated circuit (ASIC), or a programmable logic device (PLD). The PLD may be a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), a data processing unit (DPU), a neural network processing unit (NPU), a system on chip (SoC), an offload card, an accelerator card, or any combination thereof.

[0276] The multiple computing devices included in the acquisition module can be distributed in the same region or in different regions. The multiple computing devices included in the acquisition module can be distributed in the same AZ or in different AZs. Similarly, the multiple computing devices included in the acquisition module can be distributed in the same VPC or in multiple VPCs. The multiple computing devices can be any combination of servers, ASICs, PLDs, CPLDs, FPGAs, GALs, DPUs, NPUs, SoCs, offload cards, accelerator cards, and other computing devices.

[0277] It should be noted that, in other embodiments, the acquisition module can be used to execute any step in the data processing method, and the processing module can be used to execute any step in the data processing method. The steps that the acquisition module and the processing module are responsible for implementing can be specified as needed. The full functions of the data processing device are realized by respectively implementing different steps in the data processing method through the acquisition module and the processing module.

[0278] The present application also provides a computing device 70. Figure 7 As shown, computing device 70 includes a bus 702, a processor 704, a memory 706, and a communication interface 708. Processor 704, memory 706, and communication interface 708 communicate with each other via bus 702. Computing device 70 can be a server or a terminal device. It should be understood that this application does not limit the number of processors and memories in computing device 70.

[0279] The bus 702 may be a peripheral component interconnect Express (PCIe) bus, an extended industry standard architecture (EISA) bus, a unified bus (UBus or UB), a compute express link (CXL), a cache coherent interconnect for accelerators (CCIX), etc. The unified bus is also known as the Lingqu bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 The bus 704 may include a path for transmitting information between various components of the computing device 70 (eg, memory 706, processor 704, communication interface 708).

[0280] The processor 704 may include any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP) or a digital signal processor (DSP), an ASIC, an FPGA, a CPLD, an NPU, a SoC, an offload card, an accelerator card, and the like.

[0281] Memory 706 may include volatile memory, such as random access memory (RAM). Processor 704 may also include non-volatile memory, such as read-only memory (ROM), flash memory, a hard disk drive (HDD), or a solid state drive (SSD). In addition, memory 706 may also be implemented using storage class memory (SCM), phase change memory (PCM), or other types of storage media.

[0282] It is worth noting that the same type of storage medium can be configured in the same computing device to implement the function of memory 706, or two or more types of storage media can be configured to implement the function of memory 706. This application does not limit this.

[0283] The memory 706 stores executable program codes, and the processor 704 executes the executable program codes to respectively implement the functions of the aforementioned acquisition module and processing module, thereby implementing the data processing method. That is, the memory 706 stores instructions for executing the data processing method.

[0284] The communication interface 703 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computing device 70 and other devices or a communication network.

[0285] Embodiments of the present application also provide a computing device cluster. The computing device cluster includes at least one computing device. The computing device can be a server, such as a central server, an edge server, or a local server in a local data center. In some embodiments, the computing device can also be a terminal device such as a desktop computer, a laptop computer, or a smartphone.

[0286] like Figure 8As shown, the computing device cluster includes at least one computing device 70. The memory 706 in one or more computing devices 70 in the computing device cluster may store the same instructions for executing the data processing method.

[0287] In some possible implementations, the memory 706 of one or more computing devices 70 in the computing device cluster may also store some instructions for executing the data processing method. In other words, the combination of one or more computing devices 70 can jointly execute the instructions for executing the data processing method.

[0288] It should be noted that the memory 706 in different computing devices 70 in the computing device cluster can store different instructions, each for executing a portion of the functions of the data processing apparatus. In other words, the instructions stored in the memory 706 in different computing devices 70 can implement the functions of one or more modules in the acquisition module and the processing module.

[0289] In some possible implementations, one or more computing devices in a computing device cluster may be connected via a network, which may be a wide area network or a local area network. Figure 9 A possible implementation is shown. Figure 9 As shown, two computing devices 70A and 70B are connected via a network. Specifically, the connection to the network is achieved through a communication interface in each computing device. In this possible implementation, the memory 706 in computing device 70A stores instructions for executing the functions of the acquisition module. Simultaneously, the memory 706 in computing device 70B stores instructions for executing the functions of the processing module.

[0290] Figure 9 The connection method between the computing device clusters shown can be based on the consideration that the data processing method provided in this application requires a large amount of observation data for decision-making business (for example, including a first data set and a second data set, etc.), so it is considered to entrust the functions implemented by the processing module to the computing device 70B for execution.

[0291] It should be understood that Figure 9 The functionality of the computing device 70A shown in FIG. 7 may also be implemented by multiple computing devices 70. Similarly, the functionality of the computing device 70B may also be implemented by multiple computing devices 70.

[0292] The present application embodiment also provides another computing device cluster. The connection relationship between the computing devices in the computing device cluster can be similarly referred to as Figure 8 and Figure 9 The connection mode of the computing device cluster is different in that the memory 706 of one or more computing devices 70 in the computing device cluster may store the same instructions for executing the data processing method.

[0293] In some possible implementations, the memory 706 of one or more computing devices 70 in the computing device cluster may also store some instructions for executing the data processing method. In other words, the combination of one or more computing devices 70 can jointly execute the instructions for executing the data processing method.

[0294] The present application also provides a computer program product comprising instructions. The computer program product may be software or a program product comprising instructions that can be run on a computing device or stored in any available medium. When the computer program product is run on at least one computing device, the at least one computing device executes the data processing method.

[0295] The present application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center that contains one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive). The computer-readable storage medium includes instructions that instruct the computing device to execute the data processing method.

[0296] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the protection scope of the technical solutions of the embodiments of the present application.

Claims

1. A data processing method, characterized in that: The method comprises: Acquire a first data set, the first data set including first intervention data of an intervention variable and first confounding data of a confounding variable, wherein the intervention variable is a first decision-making service parameter in a decision-making service, and the confounding variable is a second decision-making service parameter in the decision-making service; Obtaining counterfactual intervention data corresponding to the first intervention data; Based on the counterfactual intervention data and the first confounding data, a first estimation result is obtained through a causal effect model, wherein the causal effect model is trained based on the confounding data corresponding to the confounding variable, the intervention data corresponding to the intervention variable that are not affected by the confounding variable, and the result data corresponding to the result variable that are not affected by the confounding variable, and the first estimation result is used to describe the causal effect estimation result on the result variable in the population corresponding to the first confounding data when the intervention variable is the counterfactual intervention data.

2. The method according to claim 1, characterized in that Obtaining a first estimation result based on the counterfactual intervention data and the first confounding data through the causal effect model includes: The counterfactual intervention data and the first confounding data are input into the causal effect model to obtain a second estimation result output by the causal effect model, and the second estimation result is used as the first estimation result.

3. The method according to claim 1, characterized in that The first data set further includes first result data of a result variable, where the result variable is a third decision-making service parameter in the decision-making service. The method further includes: Inputting the first intervention data and the first confounded data into the causal effect model to obtain a third estimation result output by the causal effect model; Obtaining a first estimation result based on the counterfactual intervention data and the first confounding data through the causal effect model includes: Inputting the counterfactual intervention data and the first confounding data into the causal effect model to obtain a second estimation result output by the causal effect model; The first estimation result is obtained according to the difference between the second estimation result and the third estimation result, and the first result data.

4. The method according to claim 1, wherein Obtaining a first estimation result based on the counterfactual intervention data and the first confounding data through the causal effect model includes: Inputting the counterfactual intervention data and the first confounding data into the causal effect model to obtain a second estimation result output by the causal effect model; Inputting the first confounded data into a first prediction model to obtain a fourth estimation result output by the first prediction model, where the fourth estimation result is result data of the outcome variable predicted by the first prediction model based on the first confounded data of the confounding variable; The first estimation result is obtained according to the sum of the second estimation result and the fourth estimation result.

5. The method according to any one of claims 1 to 4, characterized in that Before obtaining a first estimation result through the causal effect model according to the counterfactual intervention data and the first confounding data, the method further includes: Acquire a second data set, the second data set including second intervention data of the intervention variable, second confounding data of the confounding variable, and second result data of the result variable; determining, from the second result data, third result data that is not affected by the second obfuscated data; determining, from the second intervening data, third intervening data that is not affected by the second obfuscated data; The second confounded data and the third intervention data are used as inputs, and the third result data is used as a label to train the causal effect model to obtain the causal effect model.

6. The method according to claim 5, characterized in that Determining third result data that is not affected by the second obfuscated data from the second result data includes: Inputting the second confounded data into a first prediction model to obtain fourth result data output by the first prediction model, wherein the fourth result data is result data in the second result data affected by the second confounded data, and the first prediction model is used to predict the result data of the outcome variable based on the confounded data of the confounding variable; A difference between the second result data and the fourth result data is determined as the third result data.

7. The method according to claim 6, characterized in that The method further comprises: The second confused data is used as input data, and the second result data is used as a label to train the first prediction model to obtain the first prediction model.

8. The method according to claim 5, characterized in that The determining, from the second intervention data, third intervention data that is not affected by the second obfuscated data comprises: inputting the second confounding data into a second prediction model to obtain fourth intervention data output by the second prediction model, wherein the fourth intervention data is intervention data in the second intervention data that is affected by the second confounding data, and the second prediction model is used to predict the intervention data of the intervention variable based on the confounding data of the confounding variable; A difference between the second intervention data and the fourth intervention data is determined as the third intervention data.

9. The method according to claim 8, characterized in that The method further comprises: The second confused data is used as input data, and the second intervention data is used as a label to train the second prediction model to obtain the second prediction model.

10. The method according to any one of claims 1 to 9, characterized in that The causal effect model is a machine learning model.

11. A data processing device, characterized in that: include: Get modules for: Acquire a first data set, the first data set including first intervention data of an intervention variable and first confounding data of a confounding variable, wherein the intervention variable is a first decision-making service parameter in a decision-making service, and the confounding variable is a second decision-making service parameter in the decision-making service; Obtaining counterfactual intervention data corresponding to the first intervention data; Processing module for: Based on the counterfactual intervention data and the first confounding data, a first estimation result is obtained through a causal effect model, wherein the causal effect model is trained based on the confounding data corresponding to the confounding variable, the intervention data corresponding to the intervention variable that are not affected by the confounding variable, and the result data corresponding to the result variable that are not affected by the confounding variable, and the first estimation result is used to describe the causal effect estimation result on the result variable in the population corresponding to the first confounding data when the intervention variable is the counterfactual intervention data.

12. The device according to claim 11, characterized in that The processing module is used to: input the counterfactual intervention data and the first confusion data into the causal effect model, obtain a second estimation result output by the causal effect model, and use the second estimation result as the first estimation result.

13. The device according to claim 11, characterized in that The first data set further includes first result data of a result variable, where the result variable is a third decision-making service parameter in the decision-making service; The processing module is used for: Inputting the first intervention data and the first confounded data into the causal effect model to obtain a third estimation result output by the causal effect model; Inputting the counterfactual intervention data and the first confounding data into the causal effect model to obtain a second estimation result output by the causal effect model; The first estimation result is obtained according to the difference between the second estimation result and the third estimation result, and the first result data.

14. The device according to claim 11, characterized in that The processing module is used for: Inputting the counterfactual intervention data and the first confounding data into the causal effect model to obtain a second estimation result output by the causal effect model; Inputting the first confounded data into a first prediction model to obtain a fourth estimation result output by the first prediction model, where the fourth estimation result is result data of the outcome variable predicted by the first prediction model based on the first confounded data of the confounding variable; The first estimation result is obtained according to the sum of the second estimation result and the fourth estimation result.

15. The device according to any one of claims 11 to 14, characterized in that The acquisition module is used to: acquire a second data set, wherein the second data set includes second intervention data of the intervention variable, second confounding data of the confounding variable, and second result data of the result variable; The processing module is used for: determining, from the second result data, third result data that is not affected by the second obfuscated data; determining, from the second intervening data, third intervening data that is not affected by the second obfuscated data; The second confounded data and the third intervention data are used as inputs, and the third result data is used as a label to train the causal effect model to obtain the causal effect model.

16. The device according to claim 15, characterized in that The processing module is used for: Inputting the second confounded data into a first prediction model to obtain fourth result data output by the first prediction model, wherein the fourth result data is result data in the second result data affected by the second confounded data, and the first prediction model is used to predict the result data of the outcome variable based on the confounded data of the confounding variable; A difference between the second result data and the fourth result data is determined as the third result data.

17. The device according to claim 16, characterized in that The processing module is configured to: use the second obfuscated data as input data and the second result data as labels to train the first prediction model to be trained, so as to obtain the first prediction model.

18. The device according to claim 15, characterized in that The processing module is used for: inputting the second confounding data into a second prediction model to obtain fourth intervention data output by the second prediction model, wherein the fourth intervention data is intervention data in the second intervention data that is affected by the second confounding data, and the second prediction model is used to predict the intervention data of the intervention variable based on the confounding data of the confounding variable; A difference between the second intervention data and the fourth intervention data is determined as the third intervention data.

19. The device according to claim 18, characterized in that The processing module is configured to: use the second obfuscated data as input data and the second intervention data as labels to train the second prediction model to be trained, so as to obtain the second prediction model.

20. A computing device, characterized in that The computing device includes a processor and a memory; The processor is configured to execute instructions stored in the memory, so that the computing device performs the method according to any one of claims 1 to 10.

21. A computing device cluster, characterized in that: comprising at least one computing device, the at least one computing device comprising a processor and a memory; The processor is configured to execute instructions stored in the memory, so that the computing device cluster executes the method according to any one of claims 1 to 10.

22. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

23. A computer program product comprising instructions, characterized in that When the instructions are executed by a processor, the method according to any one of claims 1 to 10 is implemented.

Images