Gain model construction method and device based on credit data, equipment and storage medium
By constructing a gain model of credit data, using the interference set and initial gain model to build a deviation model, the selection deviation problem of the gain model is solved, and the unbiased target gain model is achieved, which improves prediction accuracy and flexibility.
Patent Information
- Application Number
- CN202510479048.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-16
- Publication Date
- 2025-08-01
AI Technical Summary
The prior art has the problem of selection bias when training gain models, and the existing methods may introduce confusion bias, so it is impossible to effectively utilize experimental data containing the unbiased effect of intervention.
By acquiring credit data, constructing the first model, determining the interference set and result set, using the initial gain model to predict the degree of interference, constructing a deviation model to make up for the selection bias, and forming a target gain model.
A target gain model without selection bias is realized, which accurately reflects the causal relationship between the intervention term and the target result, and improves prediction accuracy and flexibility.
Smart Images
Figure CN120410708A_ABST
Abstract
Description
Technical Field
[0001] Embodiments of the present invention relate to the field of artificial intelligence technology, and particularly to a method, device, equipment and storage medium for constructing a gain model based on credit data. Background Art
[0002] In the field of artificial intelligence technology, when training a gain model, an experimental group and a control group are usually selected from a large amount of data to train the gain model. The gain model trained based on this method has the problem of selection bias. To solve this problem, the prior art collects observation data with the same distribution as the control group data from the experimental group and adds the observation data to the control group to improve the prediction accuracy of the model for data without intervention. However, this method cannot solve the problem of selection bias in the trained gain model, but may introduce confounding bias. In addition, a baseline model can be constructed through observation data, and the residual of the baseline model on random data can be learned using experimental data to construct a residual model. The gain model is obtained by combining the baseline model and the residual model. This method does not effectively utilize the experimental data that truly contains the unbiased effect of intervention, and the trained gain model still has the problem of selection bias. Therefore, in the prior art, the problem of selection bias in the gain model cannot be solved when training the gain model. Summary of the Invention
[0003] The purpose of the present invention is to provide at least a method, device, equipment and storage medium for constructing a gain model based on credit data, which can at least solve the problem of selection bias in the gain model and at least achieve the removal of the selection bias existing in the gain model.
[0004] To solve the above technical problems, at least one embodiment of the present application provides a method for constructing a gain model based on credit data, including: inputting a first observation data set obtained into a first model to obtain a first result set corresponding to the first observation data set, where the first observation data set consists of actually collected credit data, and the first model is used to predict the result value of a target result based on the observation data, and the target result includes the credit product repeat purchase ratio and the credit product purchase rate; determining a second result set based on the first model, the first observation data set, and an obtained interference set, where the interference set includes multiple first interference values of a target interference item; determining a first gain set based on the first result set and the second result set, where the first gain in the first gain set corresponds one-to-one with the first interference value in the interference set; obtaining a second gain set corresponding to the interference set based on the interference set and an initial gain model, where the initial gain model is used to predict the influence degree of the interference degree of the target interference item on the target result; constructing a deviation model based on the interference set, a second observation data set for constructing the first model, the first gain set, and the second gain set; constructing a target gain model based on the deviation model and the initial gain model.
[0005] At least one embodiment of the present application further provides a device for constructing a gain model based on credit data, including: a obtaining module, configured to input a first observation data set obtained into a first model to obtain a first result set corresponding to the first observation data set, where the first observation data set consists of actually collected credit data, and the first model is used to predict the result value of a target result based on the observation data, and the target result includes the credit product repeat purchase ratio and the credit product purchase rate; a determining module, configured to determine a second result set based on the first model, the first observation data set, and an obtained interference set, where the interference set includes multiple first interference values of a target interference item; the determining module is further configured to determine a first gain set based on the first result set and the second result set, where the first gain in the first gain set corresponds one-to-one with the first interference value in the interference set; the obtaining module is further configured to obtain a second gain set corresponding to the interference set based on the interference set and an initial gain model, where the initial gain model is used to predict the influence degree of the interference degree of the target interference item on the target result; a constructing module, configured to construct a deviation model based on the interference set, a second observation data set for constructing the first model, the first gain set, and the second gain set; the constructing module is further configured to construct a target gain model based on the deviation model and the initial gain model.
[0006] At least one embodiment of the present application further provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the above-mentioned method for constructing a gain model based on credit data.
[0007] At least one embodiment of the present application further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, it implements the above-mentioned method for constructing a gain model based on credit data.
[0008] The method for constructing a gain model based on credit data provided by the embodiment of the present application inputs the obtained first observation data set into a first model to obtain a first result set corresponding to the first observation data set, wherein the first observation data set is composed of actually collected credit data, and the first model is used to predict the result value of the target result based on the observation data, and the target result includes the repeated purchase ratio of credit products and the purchase rate of credit products; determining a second result set based on the first model, the first observation data set, and the obtained interference set, wherein the interference set includes multiple first interference values of the target interference item; determining a first gain set based on the first result set and the second result set, wherein the first gain in the first gain set corresponds one-to-one to the first interference value in the interference set; obtaining a second gain set corresponding to the interference set based on the interference set and the initial gain model, wherein the initial gain model is used to predict the influence degree of the interference degree of the target interference item on the target result; constructing a deviation model based on the interference set, the second observation data set for constructing the first model, the first gain set, and the second gain set; constructing a target gain model based on the deviation model and the initial gain model. By constructing the first model with observation data to predict the second result of the observation data with added interference values and the first result of the observation data, the actual influence degree of adding the interference value on the target result is determined, and the initial gain model is used to predict the predicted influence degree of the interference value on the target result. The selection bias existing in the initial gain model can be determined through the actual influence degree and the predicted influence degree, so that a deviation model can be constructed based on the determined deviation, and the deviation model is used to compensate for the selection bias existing in the initial gain model to obtain a target gain model without selection bias. It is realized to alleviate the selection bias existing in the initial gain model by using observation data.
[0009] In some alternative embodiments, determining the second result set based on the first model, the first observation data set, and the obtained interference set includes: for each first observation data in the first observation data set, randomly generating a first interference value, and further generating a plurality of first interference values, and forming the interference set based on the plurality of first interference values; for each first observation data in the first observation data set, adding the first interference value corresponding to the first observation data in the interference set to the first observation data to obtain the data to be predicted, and using the first model to predict the data to be predicted to obtain a second result, and further obtaining a plurality of second results, and using the plurality of second results as the second result set. Constructing the first model through the observation data avoids the problem of selection bias in the first model, and makes the second result predicted by the first model conform to the actual result.
[0010] In some alternative embodiments, constructing the bias model based on the interference set, the second observation data set for constructing the first model, the first gain set, and the second gain set includes: for each first interference value in the interference set, obtaining the difference between the first gain corresponding to the first interference value in the first gain set and the second gain corresponding to the first interference value in the second gain set, and using the difference as the bias value, and further obtaining a plurality of bias values, and using the plurality of bias values as the bias set; constructing a bias model based on the second observation data set for constructing the first model and the bias set; wherein, the bias value is used to describe the error between the predicted influence degree of the first interference value on the target result by the initial gain model and the actual influence degree of the first interference value on the target result. The difference between the first gain that conforms to the actual situation determined by the first model and the second gain determined by the initial gain model can accurately obtain the selection bias existing in the initial gain model.
[0011] In some alternative embodiments, constructing the bias model based on the second observation data set for constructing the first model and the bias set includes: fitting the bias set based on the second observation data set for constructing the first model to obtain a bias model. Fitting the bias data in the bias set through the observation data improves the accuracy of the bias model.
[0012] In some alternative embodiments, the method further includes: randomly selecting some observation data from the third observation data set to form a control set; forming an experimental set with the data in the third observation data set other than the control set; constructing the initial gain model based on the control set and the experimental set. By selecting the control set and the experimental set to train the initial gain model, the confusion bias of the initial gain model is reduced.
[0013] In some alternative embodiments, the method further includes: obtaining a to-be-predicted interference value of the target interference item; inputting the to-be-predicted interference value into the target gain model, and taking the sum of a first predicted value of the deviation model in the target gain model for the to-be-predicted interference value and a second predicted value of the initial gain model in the target gain model as the output of the target gain model. By learning the selection bias between the initial gain model and the first model based on the observation data without selection bias, a deviation model is constructed. The target gain model is constructed by combining the deviation model and the initial gain model, so that the target gain model has no selection bias.
[0014] In some alternative embodiments, the method further includes: obtaining a second observation data set, where the second observation data set is data that has not been manually adjusted at the actual collection location, and there is no duplicate data between the second observation data set and the first observation data set; constructing the first model based on the second observation data set. The first model constructed using the observation data has no selection bias, so as to better predict the selection bias existing in the initial gain model. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] One or more embodiments are exemplarily illustrated by pictures in the corresponding drawings, and these exemplary illustrations do not constitute a limitation on the embodiments.
[0016] Figure 1 is a schematic flowchart of a method for constructing a gain model based on credit data provided by an embodiment of the present application;
[0017] Figure 2 is a schematic structural diagram of the construction and application of a first model provided by an embodiment of the present application;
[0018] Figure 3 is a schematic structural diagram of the construction and application of an initial gain model provided by an embodiment of the present application;
[0019] Figure 4 is a schematic structural diagram of the construction of a deviation model provided by an embodiment of the present application;
[0020] Figure 5 is a schematic structural diagram of the construction of a target gain model provided by an embodiment of the present application;
[0021] Figure 6 is a schematic structural diagram of a device for constructing a gain model based on credit data provided by another embodiment of the present application;
[0022] Figure 7 is a schematic structural diagram of an electronic device provided by another embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the following will elaborate on each embodiment of this application in conjunction with the accompanying drawings. However, those of ordinary skill in the art can understand that in each embodiment of this application, many technical details are presented to help readers better understand this application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions claimed in this application can still be implemented. The division of the following embodiments is for convenience of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined and cross-referenced with each other on the premise of not being contradictory.
[0024] It should be noted that the acquisition or use of the data in the embodiments of this application requires user consent. Relevant data can only be obtained after the user authorizes and permits it, and the acquisition or use of the data complies with the provisions of relevant laws and regulations.
[0025] In the field of artificial intelligence technology, when training a gain model, experimental groups and control groups are usually selected from a large amount of data to train the gain model. The gain model trained based on this method has the problem of selection bias. To solve this problem, the prior art collects observation data with the same distribution as the control group data from the experimental group and adds the observation data to the control group to improve the prediction accuracy of the model for data without intervention. However, this method cannot solve the problem of selection bias in the trained gain model and may instead introduce confounding bias. In addition, a baseline model can be constructed through observation data, and the residual of the baseline model on random data can be learned using experimental data to construct a residual model. The gain model is obtained by combining the baseline model and the residual model. This method does not effectively utilize the experimental data that truly contains the unbiased effect of the intervention, and the trained gain model still has the problem of selection bias. Therefore, in the prior art, the problem of selection bias in the gain model cannot be solved when training the gain model.
[0026] To solve the above technical problem of selection bias in the gain model, the present invention proposes a method for constructing a gain model based on credit data. The following specifically describes the implementation details of the method for constructing a gain model based on credit data in this embodiment. The following content is only the implementation details provided for convenience of understanding and is not necessary for implementing this solution.
[0027] Embodiment 1:
[0028] The method for constructing a gain model based on credit data in this embodiment can be applied to an electronic device with communication, computing, and data storage capabilities. Its specific process can be as Figure 1 shown, including the following steps Step 101 - Step 106:
[0029] Step 101: Input the obtained first observation data set into the first model to obtain a first result set corresponding to the first observation data set. The first observation data set consists of actually collected credit data. The first model is used to predict the result value of the target result, and the target result includes the repeated purchase ratio of credit products and the purchase rate of credit products.
[0030] Specifically, the first result set stores multiple first results, and each first result is a piece of data.
[0031] In some examples, the step of inputting the obtained first observation data set into the first model to obtain a first result set corresponding to the first observation data set includes:
[0032] For each first observation data in the first observation data set, input the first observation data into the first model to obtain a first result, construct a first correspondence between the first result and the first observation data, and then obtain multiple first results and multiple first correspondences; form the first result set based on the multiple first results and store the multiple first correspondences.
[0033] In some examples, the method further includes: obtaining a second observation data set, where the second observation data set is data that has not been manually adjusted in actual collection, and there is no duplicate data between the second observation data set and the first observation data set; constructing the first model based on the second observation data set.
[0034] Exemplarily, taking the credit scenario in the Internet finance field as an example, in order to obtain more benefits by controlling the price fluctuation of the loan platform to increase the repurchase rate of users. It is necessary to construct an accurate price gain repurchase rate model to accurately obtain the impact degree of price fluctuation on the user repurchase rate. To address the problem of selection bias in the model caused by selecting some data from the massive data as samples to train the price gain repurchase rate model, refer to Figure 2 , a price response model, that is, the first model, can be constructed using a large amount of market observation data. Denote the first model as F(T, X), where T represents the price, as the independent variable, and X represents other factors that affect the actual repurchase rate except the price, as the covariate, and (T, X) represents the input parameters acceptable to the first model. Refer to Figure 2, regarding the first observation data as (t, x), predicting the first observation data based on the first model, the actual repurchase rate corresponding to the first observation data, i.e., the first result, can be accurately obtained, and the first result is denoted as Y1 = F(t, x). However, there are multiple factors affecting the actual repurchase rate in the observation data, not just the price factor. Therefore, it can be known that the first result obtained by predicting the first observation data based on the first model is approximate to the actual result corresponding to the first observation data. However, this first model cannot well capture the causal relationship between a certain factor in the first observation data and the first result alone.
[0035] Step 102, determine a second result set based on the first model, the first observation data set, and the obtained interference set, where the interference set includes multiple first interference values of a target interference item.
[0036] Specifically, the target interference item is used as an independent variable, and the first interference value is used to indicate the change value of the independent variable, that is, the interference degree of the target interference item.
[0037] Continuing with the previous example, the target interference item is the price, and the first interference value is the change value of the price, which is denoted as ΔT.
[0038] In some examples, in the aforementioned step 102, the determining the second result set based on the first model, the first observation data set, and the obtained interference set includes: for each first observation data in the first observation data set, randomly generate a first interference value, and then generate multiple first interference values, and form the interference set based on the multiple first interference values; for each first observation data in the first observation data set, add the first interference value corresponding to the first observation data in the interference set to the first observation data to obtain the data to be predicted, and use the first model to predict the data to be predicted to obtain a second result, and then obtain multiple second results, and use the multiple second results as the second result set.
[0039] Specifically, for each first observation data in the first observation data set, randomly generate a first interference value, construct an interference for adding the first interference value to the first observation data to obtain the data to be predicted, use the first model to predict the data to be predicted to obtain a second result, construct a second corresponding relationship in one-to-one correspondence between the first observation data, the first interference value, the data to be predicted, and the second result, obtain multiple second corresponding relationships, and store the multiple second corresponding relationships.
[0040] Continuing with the previous example, see Figure 2, the data to be predicted is the observed data with only the price changed. The data to be predicted is denoted as (t + ΔT, x), and the second result is the actual repurchase rate corresponding to the data to be predicted, denoted as Y2 = F(t + ΔT, x).
[0041] Step 103, determine a first gain set based on the first result set and the second result set, where the first gain in the first gain set corresponds one-to-one to the first interference value in the interference set.
[0042] In some examples, for each first result in the first result set, obtain the second result corresponding to the first observed data corresponding to the first result, subtract the second result from the first result to obtain a first gain, construct a third correspondence where the first gain corresponds one-to-one to the first interference value corresponding to the first observed data, thereby obtaining multiple first gains and multiple third correspondences; form a first gain set based on the multiple first gains and store the multiple third correspondences.
[0043] Specifically, the first gain is used to represent that when the price fluctuates by the first interference value, the fluctuation of the user repurchase rate is the first gain, that is, when the price increases or decreases by the first interference value, the user repurchase rate decreases or increases by the first gain. However, since the price response model constructed based on the observed data predicts the corresponding user repurchase rate based on the observed data, and there are other factors in the observed data that directly affect the user repurchase rate, or there are other indirect factors that affect the user repurchase rate by affecting the price. Therefore, only by the way of taking the difference, the influence degree of price change on the user repurchase rate cannot be accurately determined. That is, the first gain includes the influence of the real price fluctuation on the fluctuation of the user repurchase rate, and the influence of other factors on the fluctuation of the user repurchase rate (confounding bias), and the first gain cannot directly reflect the influence degree of price fluctuation on the user repurchase rate.
[0044] Continuing with the above example, subtracting the second result from the first result, the first gain can be expressed by the following formula:
[0045] τ1 = Y2 - Y1
[0046] Step 104, obtain a second gain set corresponding to the interference set based on the interference set and the initial gain model, where the initial gain model is used to predict the influence degree of the interference degree of the target interference item on the target result.
[0047] In some examples, for each first interference in the interference set, input the first interference into the initial gain model to obtain a second gain, construct a fourth correspondence between the second gain and the first interference, thereby obtaining multiple second gains and multiple fourth correspondences, form the second gain set based on the multiple second gains and store the fourth correspondences.
[0048] Continuing with the above example, refer to Figure 3 , denote the initial gain model as θ(Δt, X), where Δt represents the price fluctuation value and serves as the independent variable, X represents factors other than price fluctuations that affect the fluctuation of the actual repurchase rate and serves as the covariate, (Δt, X) is the input parameter of the initial gain model. Input the first interference ΔT into the initial gain model to obtain the second gain, and denote the second gain as τ2 = θ(ΔT, x).
[0049] In some examples, the method further includes: randomly selecting some observation data from the third observation dataset to form a control set; using the data in the third observation dataset other than the control set to form an experimental set; constructing the initial gain model based on the control set and the experimental set.
[0050] Specifically, the third observation dataset, the second observation dataset, and the first observation dataset all belong to the overall market observation dataset.
[0051] Continuing with the above example, refer to Figure 3 , use prior knowledge to select some data from the third observation dataset for constructing the initial gain model. Randomly divide the selected part of the data into a control group and an experimental group. Do not intervene in the price for the control group, and intervene in the price by reducing or increasing the price for the experimental group. Thus, the initial gain model constructed based on the control group and the experimental group excludes the interference of other covariates except price fluctuations on the influence degree of the user repurchase rate, making the initial gain model have a low confounding bias.
[0052] Therefore, it can be known that although the initially constructed model excludes the influence degree of the covariate on the repurchase rate fluctuation, when selecting the observation data for constructing the initial gain model, it is selected according to prior knowledge, resulting in a selection bias in the initial gain model. Therefore, for the second gain, it includes the influence of the real price fluctuation on the user repurchase rate fluctuation, the influence of other factors on the user repurchase rate fluctuation (low confounding bias), and the selection bias caused by selecting some data according to prior knowledge.
[0053] Step 105, construct a bias model based on the interference set, the second observation dataset for constructing the first model, the first gain set, and the second gain set;
[0054] In some examples, in the aforementioned step 105, constructing the deviation model based on the interference set, the second observation data set for constructing the first model, the first gain set, and the second gain set includes: for each first interference value in the interference set, obtaining the difference between the first gain corresponding to the first interference value in the first gain set and the second gain corresponding to the first interference value in the second gain set, using the difference as the deviation value, thereby obtaining a plurality of deviation values, and using the plurality of deviation values as the deviation set; constructing a deviation model based on the second observation data set for constructing the first model and the deviation set; wherein, the deviation value is used to describe the error between the predicted influence degree of the initial gain model on the prediction of the target result by the first interference value and the actual influence degree of the first interference value on the target result.
[0055] Continuing with the above example, the difference between the first gain minus the confounding bias is the actual influence degree of the first interference value on the target result, and the difference between the second gain minus the selection bias is the actual influence degree of the first interference value on the target result, which can be specifically expressed by the following formula:
[0056] τ1 - confounding bias = τ2 - selection bias - low confounding bias
[0057] That is, the deviation value is equal to the difference between the confounding bias except for the low confounding bias in the confounding bias and the selection bias, where the low confounding bias is a part of the confounding bias.
[0058] In some examples, constructing the deviation model based on the second observation data set for constructing the first model and the deviation set includes: fitting the deviation set based on the second observation data set for constructing the first model to obtain a deviation model.
[0059] Continuing with the above example, reference can be made to Figure 4 , subtracting the first result from the second result to obtain the first gain, subtracting the second gain from the first gain to obtain the deviation, fitting the deviation using the second observation data set for constructing the first model, constructing a deviation model, and denoting the deviation model as f(ΔT, x).
[0060] Step 106, constructing a target gain model based on the deviation model and the initial gain model.
[0061] Continuing with the above example, the target gain model can be expressed by the following formula:
[0062] effect = θ(ΔT, x) + f(ΔT, x)
[0063] wherein, the effect is the target gain model.
[0064] In some examples, the method further includes: obtaining a to-be-predicted interference value of the target interference item; inputting the to-be-predicted interference value into the target gain model, and taking the sum of a first predicted value of the deviation model in the target gain model for the to-be-predicted interference value and a second predicted value of the initial gain model in the target gain model as the output of the target gain model.
[0065] For ease of understanding of this solution, refer to Figure 5 , when obtaining the gain model, an initial gain model can be constructed based on experimental data. The initial gain model is used to predict the first influence degree of the first intervention degree of the intervention item on the result. A first model is constructed using observational data. The first model is used to predict the result corresponding to the observational data. The first result is obtained by predicting the target observational data before the intervention is applied using the first model, and the second result is obtained by predicting the target intervention data after the first intervention degree is applied to the target observational data. Thus, the second influence degree of the intervention with the first intervention degree determined based on the first model on the result is indirectly obtained. By taking the difference between the second influence degree and the first influence degree, the total deviation existing in the first model and the initial gain model in predicting the influence degree of the intervention degree of this intervention item on the result can be obtained. The deviation model is obtained by fitting the total deviation using observational data similar to that used to construct the first model. The deviation existing in the initial gain model in predicting the influence degree of the intervention degree of this intervention item on the result can be obtained based on the deviation model. Thus, the sum of the deviation model and the initial gain model is used as the target deviation model, so that there are no problems of selection bias and confounding bias in obtaining the target deviation model.
[0066] In summary, for the method for constructing a gain model based on credit data provided in this application, the obtained first observation data set is input into the first model to obtain a first result set corresponding to the first observation data set, where the first observation data set is composed of actually collected credit data, and the first model is used to predict the result value of the target result based on the observation data, and the target result includes the repeated purchase ratio of credit products and the purchase rate of credit products; a second result set is determined based on the first model, the first observation data set, and the obtained interference set, where the interference set includes multiple first interference values of the target interference item; a first gain set is determined based on the first result set and the second result set, where the first gain in the first gain set corresponds one-to-one with the first interference value in the interference set; a second gain set corresponding to the interference set is obtained based on the interference set and the initial gain model, where the initial gain model is used to predict the influence degree of the interference degree of the target interference item on the target result; a deviation model is constructed based on the interference set, the second observation data set for constructing the first model, the first gain set, and the second gain set; a target gain model is constructed based on the deviation model and the initial gain model. By constructing the first model using the observation data on the basis of the initial gain model to predict the second result of the observation data with the added interference value and the first result of the observation data, the actual influence degree of adding the interference value on the target result is determined, and the predicted influence degree of the interference value on the target result is predicted by using the initial gain model. The selection bias existing in the initial gain model can be determined through the actual influence degree and the predicted influence degree, so that a deviation model can be constructed based on the determined deviation, and the deviation model is used to make up for the selection bias existing in the initial gain model to obtain a target gain model without selection bias. It is realized to use the observation data to alleviate the selection bias existing in the initial gain model.
[0067] Specifically, in this solution, on the basis of the initial gain model, the first model constructed by using the observation data makes up for the low confusion bias and selection bias existing in the initial gain model. It not only makes full use of the unbiased effect of the random experiment data that truly contains the intervention to construct the initial gain model to fully reflect the influence degree of the intervention on the target result, but also makes full use of the feature of the observation data without selection bias to remove the selection bias in the initial gain model. The method for constructing a gain model based on credit data proposed in this solution enables the obtained target gain model to truly reflect the causal relationship between the intervention item and the target result, has high application value, high flexibility, and is convenient for expansion. It also improves the prediction accuracy of the target gain model.
[0068] Embodiment 2:
[0069] Another embodiment of the present application relates to an apparatus for constructing a gain model based on credit data. The implementation details of the apparatus for constructing a gain model based on credit data in this embodiment will be specifically described below. The following content is only the implementation details provided for convenient understanding and is not necessary for implementing this solution. The schematic diagram of the apparatus for constructing a gain model based on credit data in this embodiment can be as shown in Figure 6 shown, including an obtaining module 601, a determining module 602, and a constructing module 603.
[0070] The obtaining module 601 is configured to input the acquired first observation data set into the first model to obtain a first result set corresponding to the first observation data set, where the first observation data set is composed of actually collected credit data, and the first model is used to predict the result value of the target result based on the observation data, and the target result includes the credit product repeat purchase ratio and the credit product purchase rate;
[0071] The determining module 602 is configured to determine a second result set based on the first model, the first observation data set, and the acquired interference set, where the interference set includes a plurality of first interference values of the target interference item;
[0072] The determining module 602 is further configured to determine a first gain set based on the first result set and the second result set, where the first gain in the first gain set corresponds one-to-one with the first interference value in the interference set;
[0073] The obtaining module 601 is further configured to obtain a second gain set corresponding to the interference set based on the interference set and the initial gain model, where the initial gain model is used to predict the influence degree of the interference degree of the target interference item on the target result;
[0074] The constructing module 603 is configured to construct a deviation model based on the interference set, the second observation data set for constructing the first model, the first gain set, and the second gain set;
[0075] The constructing module 603 is further configured to construct a target gain model based on the deviation model and the initial gain model.
[0076] In some examples, when the device is used to determine a second result set based on the first model, the first observation data set, and the obtained interference set, it is specifically configured to: for each first observation data in the first observation data set, randomly generate a first interference value, and then generate a plurality of first interference values, and form the interference set based on the plurality of first interference values; for each first observation data in the first observation data set, add the first interference value corresponding to the first observation data in the interference set to the first observation data to obtain the data to be predicted, and use the first model to predict the data to be predicted to obtain a second result, and then obtain a plurality of second results, and use the plurality of second results as the second result set.
[0077] In some examples, when the device is used to construct a deviation model based on the interference set, the second observation data set for constructing the first model, the first gain set, and the second gain set, it is specifically configured to: for each first interference value in the interference set, obtain the difference between the first gain corresponding to the first interference value in the first gain set and the second gain corresponding to the first interference value in the second gain set, and use the difference as the deviation value, and then obtain a plurality of deviation values, and use the plurality of deviation values as the deviation set; construct a deviation model based on the second observation data set for constructing the first model and the deviation set; wherein, the deviation value is used to describe the error between the predicted influence degree of the initial gain model on the prediction of the target result by the first interference value and the actual influence degree of the first interference value on the target result.
[0078] In some examples, when the device is used to construct a deviation model based on the second observation data set for constructing the first model and the deviation set, it is specifically configured to: fit the deviation set based on the second observation data set for constructing the first model to obtain a deviation model.
[0079] In some examples, the device is further configured to: randomly select some observation data from the third observation data set to form a control set; form an experimental set from the data in the third observation data set except the control set; construct the initial gain model based on the control set and the experimental set.
[0080] In some examples, the device is further configured to: obtain the to-be-predicted interference value of the target interference item; input the to-be-predicted interference value into the target gain model, and use the sum of the first prediction value of the deviation model for the to-be-predicted interference value in the target gain model and the second prediction value of the initial gain model in the target gain model as the output of the target gain model.
[0081] In some examples, the device is further configured to: obtain a second observation data set, where the second observation data set is data that has not been manually adjusted at the actual collection location, and there is no duplicate data between the second observation data set and the first observation data set; construct the first model based on the second observation data set.
[0082] It is worth mentioning that each module involved in this embodiment is a logical module. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or can be implemented as a combination of multiple physical units. In addition, to highlight the innovative part of this application, units that are not closely related to solving the technical problems proposed in this application are not introduced in this embodiment, but this does not mean that there are no other units in this embodiment.
[0083] Embodiment Three:
[0084] Another embodiment of the present application relates to an electronic device, as Figure 7 shown, including: at least one processor 901; and a memory 902 communicatively connected to the at least one processor 901; wherein, the memory 902 stores instructions executable by the at least one processor 901, and the instructions are executed by the at least one processor 901 to enable the at least one processor 901 to execute the method for constructing a gain model based on credit data in the above embodiments.
[0085] Among them, the memory and the processor are connected in a bus manner. The bus can include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art, so they will not be further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, and provides a unit for communicating with various other devices on the transmission medium. The data processed by the processor is transmitted over a wireless medium via an antenna. Further, the antenna also receives data and transmits the data to the processor.
[0086] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interface, voltage regulation, power management, and other control functions. The memory can be used to store data used by the processor when performing operations.
[0087] Embodiment Four:
[0088] Another embodiment of the present application relates to a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the above method embodiments are implemented.
[0089] That is, those skilled in the art can understand that all or part of the steps in implementing the above method embodiments can be completed by instructing relevant hardware through a program. The program is stored in a storage medium, including several instructions for causing a device (which can be a single-chip microcomputer, a chip, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs that can store program codes.
[0090] Those of ordinary skill in the art can understand that the above embodiments are specific embodiments for implementing the present application, and in practical applications, various changes can be made in form and details without departing from the spirit and scope of the present application.
Claims
1. A method for constructing a gain model based on credit data, characterized in that Including: Input the obtained first observation data set into the first model to obtain a first result set corresponding to the first observation data set, where the first observation data set consists of actually collected credit data, and the first model is used to predict the result value of the target result, and the target result includes the repeated purchase ratio of credit products and the procurement rate of credit products; Determine a second result set based on the first model, the first observation data set, and the obtained interference set, where the interference set includes multiple first interference values of the target interference item; Determine a first gain set based on the first result set and the second result set, where the first gain in the first gain set corresponds one-to-one with the first interference value in the interference set; Based on the interference set and the initial gain model, obtain a second gain set corresponding to the interference set, where the initial gain model is used to predict the influence degree of the interference degree of the target interference item on the target result; Construct a deviation model based on the interference set, the second observation data set for constructing the first model, the first gain set, and the second gain set; Construct a target gain model based on the deviation model and the initial gain model.
2. The method for constructing a gain model based on credit data according to claim 1, wherein The determining the second result set based on the first model, the first observation data set, and the obtained interference set includes: For each first observation data in the first observation data set, randomly generate a first interference value, and then generate multiple first interference values, and form the interference set based on the multiple first interference values; For each first observation data in the first observation data set, add the first interference value corresponding to the first observation data in the interference set to the first observation data to obtain data to be predicted, and use the first model to predict the data to be predicted to obtain a second result, and then obtain multiple second results, and use the multiple second results as the second result set.
3. The method for constructing a gain model based on credit data according to claim 1, wherein The constructing the deviation model based on the interference set, the second observation data set for constructing the first model, the first gain set, and the second gain set includes: For each first interference value in the interference set, obtain the difference between the first gain corresponding to the first interference value in the first gain set and the second gain corresponding to the first interference value in the second gain set, and use the difference as the deviation value, and then obtain multiple deviation values, and use the multiple deviation values as the deviation set; Construct a deviation model based on the second observation data set for constructing the first model and the deviation set; Wherein, the deviation value is used to describe the error between the prediction influence degree of the initial gain model predicting the first interference value on the target result and the actual influence degree of the first interference value on the target result.
4. The method for constructing a gain model based on credit data according to claim 3, wherein, The constructing the deviation model based on the second observation data set for constructing the first model and the deviation set includes: Fit the deviation set based on the second observation data set for constructing the first model to obtain a deviation model.
5. The method for constructing a gain model based on credit data according to claim 1, wherein The method further includes: Randomly select some observation data from the third observation data set to form a control set; Form an experimental set from the data in the third observation dataset other than the control set; Construct the initial gain model based on the control set and the experimental set.
6. The method for constructing a gain model based on credit data according to any one of claims 1 to 5, characterized in that The method further includes: Obtain the to-be-predicted interference value of the target interference item; Input the to-be-predicted interference value into the target gain model, and use the sum of the first prediction value of the deviation model in the target gain model for the to-be-predicted interference value and the second prediction value of the initial gain model in the target gain model as the output of the target gain model.
7. The method for constructing a gain model based on credit data according to any one of claims 1 to 5, characterized in that The method further includes: Obtain a second observation dataset, where the second observation dataset is data that has not been artificially adjusted at the actual collection location, and there is no duplicate data between the second observation dataset and the first observation dataset; Construct the first model based on the second observation dataset.
8. An apparatus for constructing a gain model based on credit data, characterized in that, It includes: A obtaining module, configured to input the obtained first observation dataset into the first model to obtain a first result set corresponding to the first observation dataset, where the first observation dataset is composed of actual collected credit data, and the first model is used to predict the result value of the target result, and the target result includes the credit product repeated purchase ratio and the credit product purchase rate; A determining module, configured to determine a second result set based on the first model, the first observation dataset, and the obtained interference set, where the interference set includes multiple first interference values of the target interference item; The determining module is further configured to determine a first gain set based on the first result set and the second result set, where the first gain in the first gain set corresponds one-to-one with the first interference value in the interference set; The obtaining module is further configured to obtain a second gain set corresponding to the interference set based on the interference set and the initial gain model, where the initial gain model is used to predict the influence degree of the interference degree of the target interference item on the target result; A constructing module, configured to construct a deviation model based on the interference set, the second observation dataset for constructing the first model, the first gain set, and the second gain set; The constructing module is further configured to construct a target gain model based on the deviation model and the initial gain model.
9. An electronic device, characterized in that, It includes: At least one processor; And, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the method for constructing a gain model based on credit data as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method for constructing a gain model based on credit data as described in any one of claims 1 to 7.