Auxiliary decision-making method and system based on optimal transmission theory and data fusion

Abstract

The invention discloses an auxiliary decision-making method based on an optimal transmission theory and data fusion. The method comprises the following steps: acquiring an observation data set and a non-hidden confusion experiment data set and respectively preprocessing the observation data set and the non-hidden confusion experiment data set; training a preset distribution alignment weight-adjusting neural network through the preprocessed observation data set and the non-hidden confusion experiment data set to obtain a trained distribution alignment weight-adjusting neural network; performing causal effect estimation on the user evaluation data set by using the trained distribution alignment weight adjustment neural network to obtain a relation curve of a decision variable and a long-term output result; and auxiliary decision making is carried out through the relation curve of the decision variable and the long-term output result. According to the invention, high-precision long-term causal effect estimation is realized.

Description

Technical Field

[0001] The present invention relates to the field of causal inference, and in particular to an auxiliary decision-making method and system based on optimal transmission theory and data fusion. Background Art

[0002] Causal inference is a popular data analysis technology in the era of big data. Its purpose is to discover the impact of decision variables on target results through the acquired data sets. For example, in real-life scenarios, causal inference technology can discover the relationship curve between the dosage of Class A drugs (decision variables) and the patient's blood pressure value (target result) through data analysis of a large number of patients. However, since the data sets obtained in reality are not generated by randomized controlled trials, the data have deviations caused by hidden confounding factors, so it is not possible to simply use the regression fitting of decision variables to target results to obtain the correct effect relationship curve, which creates a difficult problem in the field of causal inference technology. The most direct way to solve this problem is to obtain training data without hidden confounding factors by designing randomized trials. However, in real-life scenarios, it is extremely costly to obtain a large number of experimental data sets for training, so this approach is difficult to implement.

[0003] Furthermore, when we extend the causal inference problem to short-term and long-term application scenarios, effect estimation will become more difficult. Consider the following application scenario: a food delivery platform wants to estimate the impact of the total amount of coupons issued to new users in the first year on their total consumption on its platform in the first year. Limited by the timeliness of decision-making, the company can only make decisions on the amount of coupons issued based on the short-term consumption data of the newly registered user group on the platform this year and the observation data of the old user group registered last year for a whole year. Since the observation data of old users registered last year have hidden confounding factors, how to effectively use long-term observation data with hidden confounding factors and estimate the causal effect of long-term results for new individuals has become a problem that needs to be solved. Only by making good use of the confounded observation data and estimating the unconfused causal effect can the advantages of causal inference technology be fully utilized.

[0004] The optimal transmission theory was first used to solve the mathematical problem of transporting an object from area A to area B at the minimum cost. Later, with the development of the technology, the optimal transmission technology can not only solve the mathematical problem of the minimum cost of transporting an object, but also solve the data transport problem of aligning the distribution of data sets. This provides theoretical support for the present invention to transport the long-term observation data with obfuscation to the experimental data without hidden obfuscation, thereby eliminating the hidden obfuscation factors of the observation data.

[0005] The outstanding fitting ability of machine learning methods has attracted more and more attention and integration in the technical field. In the present invention, machine learning technology can be applied to the estimation of long-term results and the learning of sample representation.

[0006] In summary, this study uses optimal transmission theory and machine learning methods to solve the causal inference problem in long-term and short-term scenarios, and estimates the causal effect between the decision-maker's decision and the target long-term result. Applying optimal transmission theory and machine learning methods to the causal effect estimation of long-term and short-term scenarios breaks through the inference difficulties of traditional causal inference methods in long-term and short-term scenarios, and further improves the accuracy of effect estimation.

[0007] The Chinese invention application with application number 202010769250.7 discloses "A method for counterfactual prediction of the effect of set-type decisions", and its technical solution is: using a probability density ratio estimation method based on a deep neural network, the probability density ratio of the joint distribution of the latent representation of the decision variable and the confounding variable corresponding to the observed data sample and the joint distribution of the unrelated latent representation of the decision variable and the confounding variable is used as the weight of the data point composed of the latent representation of the decision variable and the confounding variable. The variational sample reweighting method is used to combine the weights of the data points composed of the latent representation of the decision variable and the confounding variable into the weights of the samples in the observed data. The weighted observed data samples are used to train the counterfactual prediction model to make counterfactual predictions on the effects of individuals under the influence of specific decisions. Compared with this technical solution, the present invention combines the optimal transmission theory with data fusion technology to achieve higher accuracy estimation of long-term causal effects. Summary of the invention

[0008] In order to solve the defects of the prior art causal inference, such as the difficulty in estimating the effect of long-term scenarios and the narrow scope of application, the present invention provides an auxiliary decision-making method and system based on optimal transmission theory and data fusion. The technical solution adopted by the present invention is: A first aspect of the present invention provides an auxiliary decision method based on optimal transmission theory and data fusion, the method comprising: Obtain the observation data set and the non-hidden confusion experimental data set and preprocess them respectively; The preset distribution alignment weighted neural network is trained by using the preprocessed observation data set and the non-hidden confusion experimental data set to obtain a trained distribution alignment weighted neural network; The trained distribution alignment weighted neural network is used to estimate the causal effect of the user evaluation dataset, and the relationship curve between the decision variables and the long-term output results is obtained; Auxiliary decision making is performed through the relationship curve between the decision variables and the long-term output results.

[0009] As a preferred solution, the observation data set is expressed as , represents the decision variables considered by the decision maker in the observation data set, represents the sample feature variable in the observation data set, represents the short-term outcome variable in the observational dataset, represents the long-term outcome variable in the observational dataset; The representation of the non-hidden confusion experimental dataset is , represents the decision variable in the non-hidden confusion experimental data set, represents the sample feature variable in the non-hidden confusion experimental data set, Represents the short-term outcome variable in the experimental dataset without hidden confounding.

[0010] As a preferred solution, the method of obtaining the observation data set and the non-hidden confusion experimental data set and preprocessing them separately includes: For the observation dataset In Perform normalization:

[0011]

[0012]

[0013]

[0014] In the formula, Represents variables The mean of Represents variables The variance of is the standardized value; The experimental dataset without hidden confusion In Perform normalization:

[0015]

[0016]

[0017] In the formula, is the standardized value; The preprocessed observation dataset and the non-hidden confusion experimental dataset are and .

[0018] As a preferred solution, the preset distribution alignment weighted neural network includes an optimal transmission weighted module, a balanced representation learning module and a long-term result estimation module; The optimal transmission weighting module is used to calculate a set of weight values ​​based on the optimal transmission theory. ; The weight value Used as a fixed value to participate in the calculation of the integral probability measurement value of the balance characterization learning module; The input of the balanced representation learning module is a subset of the preprocessed observation dataset And a subset of the preprocessed non-hidden obfuscation experimental dataset , the output is the characterization variable ; The balance representation learning module includes a function and function , the functional function Used to fit conditional distribution , the functional function Used for calculation and The integrated probability measure of represents the sample characteristic variable, represents the representation variable obtained by sample learning, represents the weight value provided by the optimal transmission weight adjustment module; represents the computational function for the integral probability measure, represents a multilayer perceptron, subscript Represents the preset parameters, for the balance representation learning module:

[0019] The loss function for the balanced representation learning module is:

[0020] in express The calculated integral probability measure, represents the decision variable of the sample; The input of the long-term outcome estimation module is the representation variable provided by the balance representation learning module and the decision variables of the samples, and the output is the intermediate results and long-term results of the model corresponding to the decision variables; the long-term result estimation module includes a recurrent neural network, an attention mechanism network, and two multi-layer perceptron networks; wherein the recurrent neural network is used to generate intermediate representation variables corresponding to the intermediate process results , , Represents the number of intermediate results; the attention mechanism network is used to fit the distribution ; The two multi-layer perceptron networks are used to fit the conditional distribution and ; Let RNN represent a recurrent neural network, Represents the activation function in the attention mechanism network function. For the long-term result estimation module, there are:

[0021]

[0022]

[0023]

[0024]

[0025]

[0026] in is a learned weight value, is the weight parameter of the attention mechanism network, which is related to the input value; is the estimated value of the long-term outcome of the corresponding decision variable A; The loss function of the long-term result estimation module includes: the error loss function between the short-term intermediate result predicted by the long-term result estimation module and the actual short-term intermediate result And the estimated error loss function of the long-term result predicted by the long-term result estimation module and the true long-term result .

[0027] As a preferred solution, for the optimal transmission weighting module, the learning weight value The method is as follows:

[0028]

[0029] in is a hyperparameter, is the transport cost matrix, which is used to reflect the transport of observation data sets. To the experimental data set The specific expression of the transportation cost is as follows:

[0030] in is the Euclidean distance.

[0031] As a preferred solution, a method for training a preset distribution alignment weighted neural network by using a preprocessed observation data set and a non-hidden confusion experimental data set to obtain a trained distribution alignment weighted neural network includes: The observed data set is evenly divided into training batches and input into the preset distribution alignment weighted neural network for training with the non-hidden confusion experimental data set in turn, and the loss function of the neural network is set , the gradient descent algorithm is used to iteratively update the neural network parameters until convergence, and finally the trained distribution alignment weighted neural network is obtained.

[0032] As a preferred solution, the loss function Includes: the integral probability metric in the balanced representation learning module, the short-term intermediate results predicted by the long-term result estimation module, and the error loss function of the actual short-term intermediate results And the estimated error loss function of the long-term result predicted by the long-term result estimation module and the true long-term result , specifically:

[0033] in control The influence strength of the term in the loss function, Represents the weight value provided by the optimal transmission weight adjustment module; Indicates the intermediate result, Indicates long-term results.

[0034] As a preferred solution, the method of using the trained distribution alignment weighted neural network to estimate the causal effect of the user evaluation data set and obtaining the relationship curve between the decision variables and the long-term output results includes: Get the user's characteristic variables , normalized to obtain ; then Input the balanced representation learning module to obtain the corresponding representation variables , and then characterize the variable Input the long-term result estimation module and the optimal transmission weight adjustment module to obtain the characteristic variables of the long-term result ; Finally, use the calculation formula in the long-term result estimation module Obtain the relationship curve between the decision variables of individual samples and the long-term output results.

[0035] The second aspect of the present invention provides an auxiliary decision system based on optimal transmission theory and data fusion, the system comprising a data preprocessing submodule, a distributed alignment weighted neural network construction submodule, a distributed alignment weighted neural network training submodule, a user hidden confusion removal evaluation module and a user decision evaluation module; The data preprocessing submodule is used to obtain the observation data set and the non-hidden obfuscation experimental data set and preprocess them respectively. The data preprocessing submodule uses a visual interface as a human-computer interaction interface. The user only needs to input the data file, and the system can complete the subsequent data processing process; The distribution alignment weighted neural network construction submodule is used to construct a distribution alignment weighted neural network according to the preprocessed observation data set and the non-hidden confusion experimental data set; The distribution alignment weighted neural network training submodule is used to train the preset distribution alignment weighted neural network through the preprocessed observation data set and the non-hidden confusion experimental data set to obtain a trained distribution alignment weighted neural network; The user hidden confusion removal evaluation module is used to encapsulate the data preprocessing submodule, the distribution alignment weighted neural network construction submodule and the distribution alignment weighted neural network training submodule, and obtain a user decision evaluation module; The user decision evaluation module is used to use the trained distribution alignment weighted neural network to estimate the causal effect of the user evaluation data set, obtain the relationship curve between the decision variables and the long-term output results, and finally assist in decision-making through the relationship curve between the decision variables and the long-term output results.

[0036] The third aspect of the present invention provides a computer device, including a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor, wherein the computer program, when executed by the processor, implements the steps of the aforementioned auxiliary decision-making method based on optimal transmission theory and data fusion.

[0037] Compared with the prior art, the present invention has the following beneficial effects: The present invention improves the training effect in the subsequent neural network by preprocessing the observation data set and the designed experimental data set, and then constructs a distribution alignment weighted neural network according to the data characteristics of the observation data set and the experimental data set. Finally, according to the loss function of the constructed distribution alignment weighted neural network, the model is fitted by gradient descent. The fitted model is used to predict the new observation data sample individuals, and the output decision variable and long-term result change relationship curve provides effective guidance for decision-making. Compared with the traditional causal inference method of long-term scenarios, the accuracy of effect estimation of data using this method is also higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flow chart of an auxiliary decision-making method based on optimal transmission theory and data fusion provided by an embodiment of the present invention; Figure 2 A block diagram of an auxiliary decision-making system based on optimal transmission theory and data fusion provided in an embodiment of the present invention. DETAILED DESCRIPTION The accompanying drawings are only used for illustrative purposes and are not to be construed as limiting the present invention; It should be clear that the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the embodiments of the present application.

[0039] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0040] When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0041] In addition, in the description of this application, unless otherwise specified, "multiple" 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 represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0042] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0043] Example 1 Please refer to Figure 1 This embodiment provides an auxiliary decision-making method based on optimal transmission theory and data fusion, and the method includes: S1: Obtain the observation dataset and the non-hidden confusion experimental dataset and preprocess them respectively; Specifically, the non-hidden obfuscation experimental data set is obtained through a random experiment; In a specific embodiment, the observation data set is represented as , represents the decision variables considered by the decision maker in the observation data set (such as the dosage of medicine given by the doctor to the patient, the amount of coupons issued by the shopping platform to the user, the teaching time arranged by the teacher for the student, etc.), Represents the sample feature variables in the observation data set (such as the patient's blood pressure, fat content index; the user's monthly salary; the student's age and IQ level, etc.), Represents the short-term outcome variable in the observation data set (e.g., blood sugar level of a patient after taking medicine for one month; platform consumption amount of a user for one month; improvement of a student's grades for one month, etc.), Represents the long-term outcome variables in the observational data set (e.g., blood sugar level of a patient after taking medication for one year; amount of platform consumption of a user for one year; improvement of a student's grades for one year, etc.); The representation of the non-hidden confusion experimental dataset is , represents the decision variable in the non-hidden confusion experimental data set, represents the sample feature variable in the non-hidden confusion experimental data set, Represents the short-term outcome variable in the experimental dataset without hidden confounding.

[0044] In a specific embodiment, the method of obtaining an observation data set and a non-hidden obfuscation experimental data set and preprocessing them respectively includes: For the observation dataset In Perform normalization:

[0045]

[0046]

[0047]

[0048] In the formula, Represents variables The mean of Represents variables The variance of is the standardized value; The experimental dataset without hidden confusion In Perform normalization:

[0049]

[0050]

[0051] In the formula, is the standardized value; The preprocessed observation dataset and the non-hidden confusion experimental dataset are and .

[0052] S2: Train the preset distribution alignment weighted neural network through the preprocessed observation data set and the non-hidden confusion experimental data set to obtain a trained distribution alignment weighted neural network; It should be noted that the trained distribution alignment weighted neural network can directly estimate the relationship curve between the decision variables of the new observed individual and the long-term output results. For example, in the application scenario of issuing coupons on a consumer platform, you only need to input the user's individual characteristic information (such as age, gender, income, etc.) to directly estimate the relationship curve between the total amount of coupons issued to the user within a year and the user's total consumption on the platform in a year. This will guide merchants to formulate accurate coupon issuance strategies.

[0053] Specifically, the preset distribution alignment weighted neural network includes an optimal transmission weighted module, a balanced representation learning module and a long-term result estimation module; The optimal transmission weighting module is used to calculate a set of weight values ​​based on the optimal transmission theory. ; The weight value Used as a fixed value to participate in the calculation of the integral probability measurement value of the balance characterization learning module; The input of the balanced representation learning module is a subset of the preprocessed observation dataset And a subset of the preprocessed non-hidden obfuscation experimental dataset , the output is the characterization variable ; The balance representation learning module includes a function and function , the functional function Used to fit conditional distribution , the functional function Used for calculation and The integrated probability measure of represents the sample characteristic variable, represents the representation variable obtained by sample learning, represents the weight value provided by the optimal transmission weight adjustment module; represents the computational function for the integral probability measure, represents a multilayer perceptron, subscript Represents the preset parameters, for the balance representation learning module:

[0054] The loss function for the balanced representation learning module is:

[0055] in express The calculated integral probability measure, represents the decision variable of the sample; The input of the long-term outcome estimation module is the representation variable provided by the balance representation learning module and the decision variables of the samples, and the output is the intermediate results and long-term results of the model corresponding to the decision variables; the long-term result estimation module includes a recurrent neural network, an attention mechanism network, and two multi-layer perceptron networks; wherein the recurrent neural network is used to generate intermediate representation variables corresponding to the intermediate process results , , Represents the number of intermediate results; the attention mechanism network is used to fit the distribution ; The two multi-layer perceptron networks are used to fit the conditional distribution and ; Let RNN represent a recurrent neural network, Represents the activation function in the attention mechanism network function. For the long-term result estimation module, there are:

[0056]

[0057]

[0058]

[0059]

[0060]

[0061] in is a learned weight value, is the weight parameter of the attention mechanism network, which is related to the input value; is the estimated value of the long-term outcome of the corresponding decision variable A; The loss function of the long-term result estimation module includes: the error loss function between the short-term intermediate result predicted by the long-term result estimation module and the actual short-term intermediate result And the estimated error loss function of the long-term result predicted by the long-term result estimation module and the true long-term result .

[0062] In a specific embodiment, for the optimal transmission weighting module, the learning weight value The method is as follows:

[0063]

[0064] in is a hyperparameter, is the transport cost matrix, which is used to reflect the transport of observation data sets. To the experimental data set The specific expression of the transportation cost is as follows:

[0065] in is the Euclidean distance.

[0066] In a specific embodiment, a method for training a preset distribution alignment weighted neural network by using a preprocessed observation data set and a non-hidden confusion experimental data set to obtain a trained distribution alignment weighted neural network includes: The observed data set is evenly divided into training batches and input into the preset distribution alignment weighted neural network for training with the non-hidden confusion experimental data set in turn, and the loss function of the neural network is set , the gradient descent algorithm is used to iteratively update the neural network parameters until convergence, and finally the trained distribution alignment weighted neural network is obtained.

[0067] Specifically, the loss function Includes: the integral probability metric in the balanced representation learning module, the short-term intermediate results predicted by the long-term result estimation module, and the error loss function of the actual short-term intermediate results And the estimated error loss function of the long-term result predicted by the long-term result estimation module and the true long-term result , specifically:

[0068] in control The influence strength of the term in the loss function, Represents the weight value provided by the optimal transmission weight adjustment module; Indicates the intermediate result, Indicates long-term results.

[0069] S3: Use the trained distribution alignment weighted neural network to estimate the causal effect of the user evaluation dataset and obtain the relationship curve between the decision variables and the long-term output results; In a specific embodiment, a method for estimating the causal effect of a user evaluation data set using a trained distribution alignment weighted neural network to obtain a relationship curve between a decision variable and a long-term output result includes: Get the user's characteristic variables , normalized to obtain ; then Input the balanced representation learning module to obtain the corresponding representation variables , and then characterize the variable Input the long-term result estimation module and the optimal transmission weight adjustment module to obtain the characteristic variables of the long-term result ; Finally, use the calculation formula in the long-term result estimation module Obtain the relationship curve between the decision variables of individual samples and the long-term output results.

[0070] S4: Assist decision making through the relationship curve between the decision variables and long-term output results.

[0071] The present invention realizes the effective utilization of a large amount of long-term observation data with confounding factors acquired at low cost in real scenarios through the transport alignment effect of optimal transmission, so that causal effect estimation of long-term scenario decisions can be made at low cost and effectively in real scenarios to guide the execution of long-term decisions.

[0072] The present invention does not restrict the data form of the decision variables, and the decision variables can be discrete variables or continuous variables, breaking through the limitation that traditional causal inference tools can only estimate binary decisions, making the present invention widely applicable.

[0073] Example 2 Please refer to Figure 2 , this embodiment provides an auxiliary decision system based on optimal transmission theory and data fusion, the system includes a data preprocessing submodule, a distributed alignment weighted neural network construction submodule, a distributed alignment weighted neural network training submodule, a user hidden confusion removal evaluation module and a user decision evaluation module; The data preprocessing submodule is used to obtain the observation data set and the non-hidden obfuscation experimental data set and preprocess them respectively. The data preprocessing submodule uses a visual interface as a human-computer interaction interface. The user only needs to input the data file, and the system can complete the subsequent data processing process; The distribution alignment weighted neural network construction submodule is used to construct a distribution alignment weighted neural network according to the preprocessed observation data set and the non-hidden confusion experimental data set; It should be noted that the distribution alignment weighted neural network construction submodule automatically receives the data set to be processed from the data preprocessing subsystem and parses it to construct a suitable distribution alignment weighted neural network; The distribution alignment weighted neural network training submodule is used to train the preset distribution alignment weighted neural network through the preprocessed observation data set and the non-hidden confusion experimental data set to obtain a trained distribution alignment weighted neural network; It should be noted that the distribution alignment weighted neural network training submodule inputs the preprocessed observation data set into batches and the entire preprocessed experimental data set into the distribution alignment weighted neural network for training, and sets the loss function of the distribution alignment weighted neural network, and uses the gradient descent algorithm to update the distribution alignment weighted neural network parameters until convergence, so as to obtain a trained distribution alignment weighted neural network; The user hidden confusion removal evaluation module is used to encapsulate the data preprocessing submodule, the distribution alignment weighted neural network construction submodule and the distribution alignment weighted neural network training submodule, and obtain a user decision evaluation module; Through the above improvements, users do not need to know the contents of the subsystems encapsulated by the system. They only need to input the training data set to the user's hidden confusion removal evaluation system through the visual interactive interface provided by the data preprocessing subsystem to obtain a usable user decision evaluation system, which reduces the difficulty for users to use the system.

[0074] The user decision evaluation module is used to use the trained distribution alignment weighted neural network to estimate the causal effect of the user evaluation data set, obtain the relationship curve between the decision variable and the long-term output result, and finally assist in decision making through the relationship curve between the decision variable and the long-term output result; It should be noted that the user evaluation decision system also provides a visual data input interface. When the user needs to estimate the effect of the decision plan on the decision target individual, he only needs to input the individual's characteristic data (height, weight, age, etc.) through the interactive interface provided by the system. The system will immediately display the impact curve of the decision on the target individual and provide the user with a recommended decision plan based on the expected effect provided by the user. From then on, the entire system framework has completed the function of assisting users in making long-term decision choices.

[0075] Example 3 This embodiment provides a computer device, including a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor. When the computer program is executed by the processor, the steps of an auxiliary decision-making method based on optimal transmission theory and data fusion described in Example 1 are implemented.

[0076] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. An auxiliary decision-making method based on optimal transmission theory and data fusion, characterized in that: The method comprises: Obtain the observation data set and the non-hidden confusion experimental data set and preprocess them respectively; The preset distribution alignment weighted neural network is trained by using the preprocessed observation data set and the non-hidden confusion experimental data set to obtain a trained distribution alignment weighted neural network; The trained distribution alignment weighted neural network is used to estimate the causal effect of the user evaluation dataset, and the relationship curve between the decision variables and the long-term output results is obtained; Auxiliary decision making is performed through the relationship curve between the decision variables and the long-term output results.

2. The auxiliary decision-making method based on optimal transmission theory and data fusion according to claim 1, characterized in that: The observation dataset is represented as , represents the decision variables considered by the decision maker in the observation data set, represents the sample feature variable in the observed data set, represents the short-term outcome variable in the observational dataset, represents the long-term outcome variable in the observational dataset; The representation of the non-hidden confusion experimental dataset is , represents the decision variable in the non-hidden confusion experimental data set, represents the sample feature variable in the non-hidden confusion experimental data set, Represents the short-term outcome variable in the experimental dataset without hidden confounding.

3. The auxiliary decision-making method based on optimal transmission theory and data fusion according to claim 1, characterized in that: Methods for obtaining an observation data set and a non-hidden confusion experimental data set and preprocessing them separately include: For the observation dataset In Perform normalization: In the formula, Represents variables The mean of Represents variables The variance of is the standardized value; The experimental dataset without hidden confusion In Perform normalization: In the formula, is the standardized value; The preprocessed observation dataset and the non-hidden confusion experimental dataset are and .

4. The auxiliary decision-making method based on optimal transmission theory and data fusion according to claim 1, characterized in that: The preset distributed alignment weighted neural network includes an optimal transmission weighted module, a balanced representation learning module and a long-term result estimation module; The optimal transmission weighting module is used to calculate a set of weight values ​​based on the optimal transmission theory. ; The weight value Used as a fixed value to participate in the calculation of the integral probability measurement value of the balance characterization learning module; The input of the balanced representation learning module is a subset of the preprocessed observation dataset And a subset of the preprocessed non-hidden obfuscation experimental dataset , the output is the characterization variable ; The balance representation learning module includes a function and function , the functional function Used to fit conditional distribution , the functional function Used for calculation and The integrated probability measure of represents the sample characteristic variable, represents the representation variable obtained by sample learning, represents the weight value provided by the optimal transmission weight adjustment module; represents the computational function for the integral probability measure, represents a multilayer perceptron, subscript Represents the preset parameters, for the balance representation learning module: The loss function for the balanced representation learning module is: in express The calculated integral probability measure, represents the decision variable of the sample; The input of the long-term outcome estimation module is the representation variable provided by the balance representation learning module and the decision variables of the samples, and the output is the intermediate results and long-term results of the model corresponding to the decision variables; the long-term result estimation module includes a recurrent neural network, an attention mechanism network, and two multi-layer perceptron networks; wherein the recurrent neural network is used to generate intermediate representation variables corresponding to the intermediate process results , , Represents the number of intermediate results; the attention mechanism network is used to fit the distribution ; The two multi-layer perceptron networks are used to fit the conditional distribution and ; Let RNN denote a recurrent neural network, Represents the activation function in the attention mechanism network function. For the long-term result estimation module, there are: in is a learned weight value, is the weight parameter of the attention mechanism network, which is related to the input value; is the estimated value of the long-term outcome of the corresponding decision variable A; The loss function of the long-term result estimation module includes: the error loss function between the short-term intermediate result predicted by the long-term result estimation module and the actual short-term intermediate result And the estimated error loss function of the long-term result predicted by the long-term result estimation module and the true long-term result .

5. The auxiliary decision-making method based on optimal transmission theory and data fusion according to claim 4 is characterized in that: For the optimal transmission weighting module, the learning weight value The method is as follows: in is a hyperparameter, is the transport cost matrix, which is used to reflect the transport of observation data sets. To the experimental data set The specific expression of the transportation cost is as follows: in is the Euclidean distance.

6. The auxiliary decision-making method based on optimal transmission theory and data fusion according to claim 4, characterized in that: The method of training a preset distribution alignment weighted neural network by using the preprocessed observation data set and the non-hidden confusion experimental data set to obtain a trained distribution alignment weighted neural network includes: The observed data set is evenly divided into training batches and input into the preset distribution alignment weighted neural network for training with the non-hidden confusion experimental data set in turn, and the loss function of the neural network is set , the gradient descent algorithm is used to iteratively update the neural network parameters until convergence, and finally the trained distribution alignment weighted neural network is obtained.

7. The auxiliary decision-making method based on optimal transmission theory and data fusion according to claim 6, characterized in that: The loss function Includes: the integral probability metric in the balanced representation learning module, the short-term intermediate results predicted by the long-term result estimation module, and the error loss function of the actual short-term intermediate results And the estimated error loss function of the long-term result predicted by the long-term result estimation module and the true long-term result , specifically: in control The influence strength of the term in the loss function, Represents the weight value provided by the optimal transmission weight adjustment module; Indicates the intermediate result, Indicates long-term results.

8. The auxiliary decision-making method based on optimal transmission theory and data fusion according to claim 4 is characterized in that: Methods for estimating the causal effect of the user evaluation data set using the trained distribution alignment weighted neural network and obtaining the relationship curve between the decision variables and the long-term output results include: Get the user's characteristic variables , normalized to obtain ; then Input the balanced representation learning module to obtain the corresponding representation variables , and then characterize the variable Input the long-term result estimation module and the optimal transmission weight adjustment module to obtain the characteristic variables of the long-term result ; Finally, use the calculation formula in the long-term result estimation module Obtain the relationship curve between the decision variables of individual samples and the long-term output results.

9. An auxiliary decision-making system based on optimal transmission theory and data fusion, characterized in that: The system includes a data preprocessing submodule, a distribution alignment weighted neural network construction submodule, a distribution alignment weighted neural network training submodule, a user hidden confusion removal evaluation module and a user decision evaluation module; The data preprocessing submodule is used to obtain the observation data set and the non-hidden confusion experimental data set and perform preprocessing respectively; The distribution alignment weighted neural network construction submodule is used to construct a distribution alignment weighted neural network according to the preprocessed observation data set and the non-hidden confusion experimental data set; The distribution alignment weighted neural network training submodule is used to train the preset distribution alignment weighted neural network through the preprocessed observation data set and the non-hidden confusion experimental data set to obtain a trained distribution alignment weighted neural network; The user hidden confusion removal evaluation module is used to encapsulate the data preprocessing submodule, the distribution alignment weighted neural network construction submodule and the distribution alignment weighted neural network training submodule, and obtain a user decision evaluation module; The user decision evaluation module is used to use the trained distribution alignment weighted neural network to estimate the causal effect of the user evaluation data set, obtain the relationship curve between the decision variables and the long-term output results, and finally assist in decision-making through the relationship curve between the decision variables and the long-term output results.

10. A computer device, characterized in that: It includes a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor, and when the computer program is executed by the processor, it implements the steps of an auxiliary decision-making method based on optimal transmission theory and data fusion as described in any one of claims 1 to 8.

Images