CN118379086A - Data prediction method, apparatus, computer device, readable storage medium, and program product - Google Patents

Abstract

The present application relates to a data prediction method, device, computer equipment, computer-readable storage medium and computer program product. The method comprises: obtaining user feature data and resource conversion values of users of each stratified type; performing resource conversion sensitivity analysis and elastic data penetration index analysis on the user feature data of users of each stratified type according to a data prediction model to obtain sensitivity parameters and elastic index parameters; performing data penetration index result prediction on the sensitivity parameters, elastic index parameters and each resource conversion value based on the data prediction model to obtain the data penetration index results of users of each stratified type under each resource conversion value. The use of this method can improve the accuracy of data prediction.

Description

Data prediction method, device, computer equipment, readable storage medium and program product

Technical Field

The present application relates to the field of artificial intelligence technology, and in particular to a data prediction method, apparatus, computer equipment, computer-readable storage medium, and computer program product.

Background technique

With the development of artificial intelligence technology, the prediction of user consumption behavior penetration rate is becoming more and more common in the shared electric motorcycle industry. The penetration rate can help operating platforms analyze their market penetration level and then evaluate market potential.

In traditional technology, the Logit regression model (Logit model, assessment model) is used for market analysis. In the data related to the user consumption behavior of the shared electric motorcycle industry, the variables that affect the penetration rate of user consumption behavior are selected as independent variables, and the selected variables are imported into the Logit regression model for modeling. Then, based on the Logit regression model, the variables that affect the penetration rate of user consumption behavior in the user data of users of a single stratification type are analyzed and predicted to obtain the penetration rate of users of the current stratification type.

However, in traditional technologies, since the Logit regression model can only learn each user type independently, it leads to data sparsity problems, which in turn leads to poor accuracy in user penetration prediction.

Summary of the invention

Based on this, it is necessary to provide a data prediction method, apparatus, computer device, computer-readable storage medium and computer program product to address the above technical problems.

In a first aspect, the present application provides a data prediction method, comprising:

Obtain user feature data and resource conversion values of users of each stratification type;

Perform resource conversion sensitivity analysis and elastic data penetration index analysis on the user feature data of the users of each stratified type according to the data prediction model to obtain sensitivity parameters and elastic index parameters;

Based on the data prediction model, the sensitivity parameter, the elasticity index parameter and each resource conversion value are predicted for the data penetration index result, and the data penetration index result of each stratified type of user under each resource conversion value is obtained.

In one embodiment, the data penetration index result prediction is performed on the sensitivity parameter, the elasticity index parameter and each resource conversion value based on the data prediction model to obtain the data penetration index result of each stratified type of user under each resource conversion value, including:

Performing data processing on the elasticity index parameter according to the data evaluation structure in the data prediction model to obtain an initial data penetration index result;

The initial data penetration index result is modified and constrained based on the sensitivity parameter, the resource conversion value and the data assessment structure to obtain the data penetration index result of each user of the stratified type under each resource conversion value.

In one embodiment, before obtaining the user characteristic data and resource conversion value of each hierarchical type of user, the method further includes:

Obtain the neural network structure and initial sample set of the data prediction model to be trained;

Screening the initial sample set for an experimental sample set and a control sample set according to a propensity score matching rule, and constructing a target sample set based on the experimental sample set and the control sample set;

The neural network structure is trained according to the target sample set, and when the neural network structure meets the preset iteration conditions, a data evaluation structure is combined on the basis of the neural network structure to obtain a trained data prediction model.

In one embodiment, screening the initial sample set into an experimental sample set and a control sample set according to a propensity score matching rule, and constructing a target sample set based on the experimental sample set and the control sample set, includes:

Performing prediction processing on the initial sample set according to the propensity score prediction model to obtain a propensity score corresponding to each user sample in the initial sample set;

Taking the sample set of each stratified type of users in the initial sample set at each resource conversion value as the experimental sample set, and determining a control sample set matching the experimental sample set from other experimental sample sets of other resource conversion values of users of the stratified type based on the propensity score;

A target sample set is constructed based on each of the experimental sample sets and the control sample set corresponding to the experimental sample set.

In one embodiment, the step of using the sample set of each stratified type of user in the initial sample set at each resource conversion value as the experimental sample set, and determining a control sample set matching the experimental sample set from other experimental sample sets of other resource conversion values of users of the stratified type based on the propensity score, comprises:

In each of the stratified types, for each experimental sample set in the resource conversion value, a target propensity score matching the propensity score of each of the user samples in the experimental sample set is retrieved according to a nearest neighbor matching algorithm;

A control sample set matched with the experimental sample set is determined based on the target propensity score.

In one embodiment, constructing a target sample set based on each of the experimental sample sets and the control sample set corresponding to the experimental sample set includes:

Calculating the standardized mean difference between each of the experimental sample sets and the control sample set corresponding to the experimental sample set;

When the standardized mean difference satisfies a preset threshold condition, each of the experimental sample sets and the control sample set corresponding to the experimental sample set are determined as a target sample set.

In a second aspect, the present application also provides a data prediction device, comprising:

A first acquisition module is used to acquire user characteristic data and resource conversion values of users of each hierarchical type;

An analysis module, configured to perform resource conversion sensitivity analysis and elastic data penetration index analysis on the user characteristic data of the users of each stratified type according to a data prediction model, and obtain sensitivity parameters and elastic index parameters;

The prediction module is used to predict the data penetration index results of the sensitivity parameters, the elasticity index parameters and the resource conversion values based on the data prediction model, and obtain the data penetration index results of users of each stratified type under each resource conversion value.

In one of the embodiments, the prediction module is specifically used to perform data processing on the elasticity index parameter according to the data evaluation structure in the data prediction model to obtain an initial data penetration index result;

The initial data penetration index result is modified and constrained based on the sensitivity parameter, the resource conversion value and the data assessment structure to obtain the data penetration index result of each user of the stratified type under each resource conversion value.

In one embodiment, the device further comprises:

The second acquisition module is used to obtain the neural network structure and initial sample set of the data prediction model to be trained;

A screening module, used to screen the initial sample set into an experimental sample set and a control sample set according to a propensity score matching rule, and construct a target sample set based on the experimental sample set and the control sample set;

The training module is used to train the neural network structure according to the target sample set, and when the neural network structure meets the preset iteration conditions, a trained data prediction model is obtained by combining the data evaluation structure on the basis of the neural network structure.

In one embodiment, the screening module is specifically used to perform prediction processing on the initial sample set according to the propensity score prediction model to obtain the propensity score corresponding to each user sample in the initial sample set;

Taking the sample set of each stratified type of users in the initial sample set at each resource conversion value as the experimental sample set, and determining a control sample set matching the experimental sample set from other experimental sample sets of other resource conversion values of users of the stratified type based on the propensity score;

A target sample set is constructed based on each of the experimental sample sets and the control sample set corresponding to the experimental sample set.

In one embodiment, the screening module is specifically used to retrieve, for each experimental sample set in the resource conversion value in each of the stratified types, a target propensity score that matches the propensity score of each of the user samples in the experimental sample set according to a nearest neighbor matching algorithm;

A control sample set matched with the experimental sample set is determined based on the target propensity score.

In one embodiment, the screening module is specifically used to calculate the standardized mean difference between each of the experimental sample sets and the control sample set corresponding to the experimental sample set;

When the standardized mean difference satisfies a preset threshold condition, each of the experimental sample sets and the control sample set corresponding to the experimental sample set are determined as a target sample set.

In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

Obtain user feature data and resource conversion values of users of each stratification type;

Perform resource conversion sensitivity analysis and elastic data penetration index analysis on the user feature data of the users of each stratified type according to the data prediction model to obtain sensitivity parameters and elastic index parameters;

Based on the data prediction model, the sensitivity parameter, the elasticity index parameter and each resource conversion value are predicted for the data penetration index result, and the data penetration index result of each stratified type of user under each resource conversion value is obtained.

In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

Obtain user feature data and resource conversion values of users of each stratification type;

Perform resource conversion sensitivity analysis and elastic data penetration index analysis on the user feature data of the users of each stratified type according to the data prediction model to obtain sensitivity parameters and elastic index parameters;

Based on the data prediction model, the data penetration index result prediction is performed on the sensitivity parameter, the elasticity index parameter and each resource conversion value to obtain the data penetration index result of each stratified type of user under each resource conversion value.

In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

Obtain user feature data and resource conversion values of users of each stratification type;

Perform resource conversion sensitivity analysis and elastic data penetration index analysis on the user feature data of the users of each stratified type according to the data prediction model to obtain sensitivity parameters and elastic index parameters;

Based on the data prediction model, the sensitivity parameter, the elasticity index parameter and each resource conversion value are predicted for the data penetration index result, and the data penetration index result of each stratified type of user under each resource conversion value is obtained.

The above-mentioned data prediction method, device, computer equipment, computer-readable storage medium and computer program product obtain user feature data and resource conversion values of users of each stratification type; perform resource conversion sensitivity analysis and elastic data penetration index analysis on the user feature data of the users of each stratification type according to the data prediction model to obtain sensitivity parameters and elastic index parameters; perform data penetration index result prediction on the sensitivity parameters, the elastic index parameters and the resource conversion values based on the data prediction model to obtain the data penetration index results of the users of each stratification type under the resource conversion values. By adopting this method, the user feature data of the users of each stratification type are subjected to resource conversion sensitivity analysis and elastic data penetration index analysis according to the data prediction model, and the sensitivity parameters and elastic index parameters are analyzed and calculated based on the pre-learned combined information of the user features between the stratification types, so as to realize information sharing between the user feature data of different stratification types, avoid the problem of data sparsity, and predict the data penetration index results according to the sensitivity parameters and elastic index parameters to ensure that the data penetration index results are within a reasonable range, thereby improving the accuracy of data prediction.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments of the present application or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

FIG1 is a schematic diagram of a flow chart of a data prediction method in one embodiment;

FIG2 is a schematic diagram of a process for determining a data penetration index result based on an elasticity index parameter and a sensitivity parameter in one embodiment;

FIG3 is a schematic diagram of a process for training a neural network structure of a data prediction model in one embodiment;

FIG4 is a schematic diagram of a process of constructing a target sample set in one embodiment;

FIG5 is a schematic flow chart of a step of determining a control sample set corresponding to an experimental sample set in one embodiment;

FIG6 is a schematic flow chart of the steps of constructing a target sample set according to an experimental sample set and a control sample set in one embodiment;

FIG7 is a block diagram of a data prediction device in one embodiment;

FIG. 8 is a diagram showing the internal structure of a computer device in one embodiment.

Detailed ways

In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

In one embodiment, as shown in FIG1 , a data prediction method is provided. In the field of data analysis, the data prediction method is based on a data prediction model. According to the user characteristic data of the existing objective data, in the data analysis and statistical methods, it reflects the degree to which a certain variable is affected by the variable when the variable has different values (i.e., the sensitivity parameter), and the degree to which the change of the variable affects the data penetration index result (i.e., the elasticity index parameter), and then predicts the data penetration index result based on the sensitivity parameter and the elasticity index parameter. The data penetration index result is used to measure the popularity of a specific product, demand or effect in the target user group. This embodiment is illustrated by applying the method to a terminal. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

Step 102: Obtain user characteristic data and resource conversion values of users of each hierarchical type.

In the embodiment of the present application, the data prediction model can be a semi-black box model for predicting the penetration rate of riding card users in the field of shared travel electric motorcycles. The model structure of the data prediction model is described in detail in the following embodiment. The user stratification type can be divided according to the needs of the user role of the data prediction model. For example, according to the needs of the user role, the user stratification type can be divided into users, city group users, and strategic group users; the user role pre-sets the resource conversion value that needs to be predicted according to the needs, that is, the different discount rates corresponding to the amount of the riding card.

The terminal obtains the user portraits of users of each stratification type through the shared motorcycle operation platform according to the pre-divided user stratification types. The user portraits include features of multiple dimensions, such as age, gender, birth time, registration time, first ride time, user login client type, total number of orders in the last 3, 5, 7, 15, and 30 days, average number of orders, maximum number of orders, average duration of use, number of days since the last use of the car, competition intensity level of the city, vehicle efficiency level of the city, order scale level (volume) of the city, etc. Then, the terminal uses the user data and the pre-set discount rate as input data for the data prediction model.

After the terminal obtains the input user data, the data prediction model cleans and preprocesses the user data, including processing missing values, outliers and duplicate values in the user data, and normalizing or standardizing the user data of different dimensions to ensure the quality and consistency of the input user data. For the user data after data cleaning and preprocessing, the data prediction model extracts, transforms and combines the user data to obtain user feature data with more information and predictive capabilities. Optionally, for non-numeric features, the data prediction model needs to convert them into numerical data before processing, including unique hot encoding, label encoding or other encoding methods for classification features so that the data prediction model can understand and process them. For data with too high dimensions or redundant features, the data prediction model can perform feature dimensionality reduction operations to reduce computational complexity and improve the generalization ability of the model.

Step 104 , performing resource conversion sensitivity analysis and elastic data penetration index analysis on the user characteristic data of each stratified type of users according to the data prediction model, and obtaining sensitivity parameters and elastic index parameters.

In the embodiment of the present application, the data prediction model includes a neural network structure and a data assessment structure. Among them, the terminal uses the mapping relationship between the user feature data and the sensitivity parameters and elasticity index parameters learned by the neural network structure during the model training process to perform resource conversion sensitivity analysis and elastic data penetration index analysis on the user feature data of each stratified type of users, determine the sensitivity of users of different stratified types to discount prices and the elasticity of the data penetration index results of users of different stratified types to changes in resource conversion values, thereby obtaining the sensitivity parameters and elasticity index parameters of each user stratification type, and the sensitivity parameters and elasticity index parameters are used to predict the data penetration index results.

In an exemplary embodiment, the data prediction model learns the complex mapping relationship between the user feature data of each stratified type of users and the discount rate and penetration rate of the cycling card through training. After the training is completed, the data prediction model will use these learned mapping relationships to predict the penetration rate of each stratified type of users at different discount rates. By performing a sensitivity analysis on the change in discount rate, the sensitivity of different types of users to the discount price can be determined. For example, individual city users may be more sensitive to discount rates, while strategic group users may be relatively less sensitive. At the same time, the data prediction model will also analyze the elasticity of the penetration rate of users of different stratified types to changes in resource conversion values, that is, the sensitivity of the penetration rate to changes in discount rates, and then obtain the penetration rate for predicting users of each stratified type at different discount rates, where the penetration rate is the proportion of card-holding cycling users to all cycling users.

Step 106, based on the data prediction model, the sensitivity parameter, the elasticity index parameter and each resource conversion value are predicted for the data penetration index result, and the data penetration index result of each stratified type of user under each resource conversion value is obtained.

In the embodiment of the present application, the neural network structure of the data prediction model is connected to the data assessment structure to form a semi-black box model. After the terminal obtains the sensitivity parameter and the elasticity index parameter through the neural network structure, the sensitivity parameter and the elasticity index parameter are transmitted to the data assessment structure, wherein the definition of the data assessment structure is as follows:

in, Different user stratification types, For the The user of the hierarchical type has a resource conversion value of The data penetration index result at the time The user of each stratification type has a discount rate of The permeability at ; The resource conversion value, i.e., the discount rate, for example, 1, 2, 3, etc. is a sensitivity parameter, which is usually greater than 0, indicating that when the resource conversion value is smaller (for example, the discount rate is higher), the predicted data penetration index result is higher (the user penetration rate is higher); Elasticity indicator parameters for different user stratification types.

Based on the obtained sensitivity parameters, elasticity index parameters and resource conversion values of each stratified type of users, the terminal predicts the data penetration index results based on the data prediction model, that is, predicts the possible penetration rate results of each stratified type of users under different discount rates. The penetration rate results provide a basis for subsequent decision-making in the shared travel platform.

In an exemplary embodiment, in the medical field, the resource conversion value may refer to physiological, pathological or medical data indicators of patients after receiving treatment or taking specific health management measures, such as objective indicators such as blood pressure, blood sugar, and physiological function indicators. In medical data analysis, the sensitivity parameter can indicate the degree of response of the treatment plan, drug or treatment measure to the change of the patient's health indicator. The sensitivity parameter reflects the degree of influence of a specific treatment factor on the patient's physiological state; the elasticity index parameter is used to measure the degree of response of the change of a variable in the medical system or treatment plan to the patient's physiological state, which can cover multiple aspects from the patient's treatment results to the treatment resources in the treatment plan, and is used to evaluate the patient's physiological state when implementing different treatment strategies or adjusting resource allocation. The sensitivity parameter and the elasticity index parameter are both quantitative collinear factors based on actual data and analysis. The final data penetration index result can evaluate the expected results of the patient's treatment plan according to the data prediction model.

In the above data prediction method, resource conversion sensitivity analysis and elastic data penetration index analysis are performed on the user feature data of users of each stratification type according to the data prediction model. Based on the pre-learned combined information of user features between each stratification type, the sensitivity parameters and elasticity index parameters are analyzed and calculated. This can achieve information sharing between user feature data of different stratification types and avoid data sparsity problems. The data penetration index results are predicted based on the sensitivity parameters and elasticity index parameters to ensure that the data penetration index results are within a reasonable range, thereby improving the accuracy of data prediction.

In an exemplary embodiment, as shown in FIG2 , step 106 includes steps 202 to 204. Among them:

Step 202, performing data processing on elasticity index parameters according to the data evaluation structure in the data prediction model to obtain initial data penetration index results.

The data evaluation structure may be a Logit model structure.

In the embodiment of the present application, referring to the definition of the evaluation structure in the above embodiment, the terminal is based on the user penetration elasticity parameter A preliminary analysis is performed to obtain the initial penetration index results, taking the penetration rate prediction of the shared travel platform as an example, that is, the penetration rate elasticity parameter It reflects the user's sensitivity to the change in penetration rate under different discount rates. When it is larger, it means that users of this user type are more sensitive to the penetration rate when the discount changes, and the penetration rate changes more dramatically, that is, when the discount rate is larger, users of this user type are less willing to buy cycling cards.

Step 204, based on the sensitivity parameters, resource conversion values and data assessment structure, the initial data penetration index results are modified and constrained to obtain the data penetration index results of each hierarchical type of user under each resource conversion value.

In the embodiment of the present application, since only elasticity indicator parameters are used for prediction, the counter-realistic problem of violating diminishing marginal returns may occur. Therefore, in the embodiment of the present application, a neural network structure and a data assessment structure are introduced, and the sensitivity parameters obtained by the neural network prediction are used to correct and constrain the initial data penetration indicator results. The initial data penetration indicator results are further adjusted to more accurately reflect the data penetration indicator results of users of different stratified types under different resource conversion values, so as to ensure that the adjusted results meet the actual scenarios and business needs.

In this embodiment, the user data characteristics are converted into interpretable initial data penetration index results through the data assessment structure and elasticity index parameters in the data prediction model. The initial data penetration index results are corrected and constrained based on the sensitivity parameters, resource conversion values and data assessment structure to more accurately reflect the data penetration of users of different stratification types under different resource conversion values, ensure that the data penetration index results are within a reasonable range, and thereby improve the accuracy of data prediction.

In an exemplary embodiment, as shown in FIG3 , before step 102, the method further includes steps 302 to 306. Among them:

Step 302, obtaining the neural network structure and initial sample set of the data prediction model to be trained.

In the embodiment of the present application, the neural network structure for training is first determined, including the number of layers of the network, the number of neurons in each layer, the activation function, etc. Then, the initial sample set is collected and sorted, and the initial sample set will be used to train the neural network model. The initial sample set includes user data of users of different hierarchical types, and the user data of each hierarchical type contains multiple groups of sample sets under different resource conversion values and the data penetration indicator label corresponding to each sample set.

Step 304 , screening the initial sample set for the experimental sample set and the control sample set according to the propensity score matching rule, and constructing the target sample set based on the experimental sample set and the control sample set.

In the embodiment of the present application, the terminal selects the experimental sample set and the control sample set according to the propensity score matching rule to ensure the distribution consistency of the experimental sample set and the control sample set in certain key features. Then, a target sample set is constructed based on the experimental sample set and the control sample set for subsequent training and testing.

In an optional embodiment, the terminal can compare the distribution of the experimental sample set and the control sample set on some key covariates, and then select a method (e.g., nearest neighbor matching algorithm) to match the experimental sample set and the control sample set to ensure that the distribution of the two sample sets on the key covariates tends to be consistent. Among them, for the selection of key covariates in user data, variables that affect both intervention allocation and results are used as key covariates. For example, age may be a factor that affects the conversion of users to a certain discount, and it also affects the user's card-holding riding on a certain day, so it is used as one of the key covariates; variables affected by intervention items should be excluded. For example, discounts will affect different cycling card products, so the segmented cycling card features are already biased and should be excluded.

Step 306, training the neural network structure according to the target sample set, and when the neural network structure meets the preset iteration conditions, combining the data evaluation structure on the basis of the neural network structure to obtain a trained data prediction model.

Among them, the neural network structure can be a multi-layer MLP (Multilayer Perceptron).

In the embodiment of the present application, the terminal uses the target sample set to train the neural network structure, passes the target sample set to the input layer of the network, and propagates through the hidden layer of the network, and finally obtains the prediction result of the output layer. By optimizing the loss function, the output of the neural network structure is and As close as possible to the data penetration indicator label. The terminal can use the back propagation algorithm combined with the optimizer (e.g., gradient descent) to adjust the weights and biases in the neural network structure to minimize the loss function. When the loss value of the neural network structure meets the preset threshold or the number of iterations of the neural network structure meets the preset iteration condition, the terminal ends the training of the neural network structure and combines the trained neural network structure with the data assessment structure to obtain a trained data prediction model.

In this embodiment, the propensity score matching rule screens the initial sample set for the experimental sample set and the control sample set, which can ensure that the comparison between the experimental sample set and the control sample set is more comparable, reduce potential bias, avoid expensive control experiments to collect samples while ensuring sample quality, and improve the generalization ability of the model, thereby improving the accuracy of data prediction.

In an exemplary embodiment, as shown in FIG4 , step 304 includes steps 402 to 406. Among them:

Step 402 , predicting the initial sample set according to the propensity score prediction model, and obtaining the propensity score corresponding to each user sample in the initial sample set.

In an embodiment of the present application, the propensity score matching rule includes a strategy for matching the experimental sample set and the control sample set based on the propensity score prediction model and the subsequent propensity score based on the user sample. The propensity score prediction model can be a machine learning model, for example, a logistic regression model. The terminal predicts the propensity score of each user sample in the initial sample set according to the propensity score prediction model to obtain the propensity score of each user sample. The propensity score is used to reflect the similarity of the feature distribution between user samples under different resource conversion values.

Step 404, taking the sample set of each stratified type of users under each resource conversion value in the initial sample set as the experimental sample set, and determining a control sample set matching the experimental sample set from other experimental sample sets of other resource conversion values of users of the stratified type based on the propensity score.

In an embodiment of the present application, since the initial sample set includes multiple hierarchical types of users, each hierarchical type of users includes multiple further groupings of resource conversion values, wherein for each hierarchical type of users, multiple groups of experimental sample sets and control sample sets need to be determined, and the screening process of the experimental sample sets and the control sample sets is the same for each hierarchical type of users. Taking a certain hierarchical type of user as an example, the terminal uses the sample sets under each resource conversion value as experimental sample sets, and based on the propensity scores of each user sample in each experimental sample set, in other experimental sample sets under the grouping of other resource conversion values in the hierarchical type of users, determines the user samples in the control sample set that match the propensity scores of the user samples in the experimental sample set under the current resource conversion value, thereby obtaining the control sample set. On this basis, each experimental sample set under a resource conversion value contains one or more control sample sets.

In a specific embodiment, taking a user of a certain stratified type as an example, the terminal has 30%, 40%, and 50% off cycling users in this stratified type every day, wherein the number of users at different discounts is inconsistent, and the number of user card holders is also inconsistent, and the relationship between discount and user penetration cannot be directly compared. Then, through the screening of the experimental sample set and the control sample set, each discount group will match the control samples with the same number of user samples as its own from other discount groups to form a control sample set, that is, the cycling users at 30% off are matched with the cycling users at 40% off to obtain the control sample set 4', the cycling users at 30% off are matched with the cycling users at 50% off to obtain the control sample set 5', and similarly, the cycling users at 40% off and 50% off can obtain the control sample set 3' and the control sample set 5' respectively; the control sample set 3' and the control sample set 4', that is, the sample sequence is [(3,4',5'), (3',4,5'), (3',4',5)].

Step 406: construct a target sample set based on each experimental sample set and a control sample set corresponding to the experimental sample set.

In the embodiment of the present application, the terminal combines each experimental sample set with its corresponding control sample set to construct a target sample set. Each target sample set includes multiple experimental sample sets and control sample sets corresponding to the experimental sample sets, ensuring the balance of the sample feature distribution of the experimental sample sets and the control sample sets in terms of tendency.

In an optional embodiment, the terminal can perform secondary screening on the multiple groups of sample sequences constructed daily according to the law of diminishing marginal returns. The law of diminishing marginal returns means that as the number of samples increases, the benefits brought by the additional samples gradually decrease. Specifically, the terminal can evaluate the multiple groups of sample sequences constructed daily, calculate the expected benefits or benefits of each group of sample sequences, and sort the sample sequences according to the expected benefits or benefits, from high to low. Then, according to the law of diminishing marginal returns, the terminal selects sample sequences that meet specific conditions for retention, and eliminates sample sequences with lower returns. Among them, the specific conditions of the law of diminishing marginal returns can be to set a threshold or rule, and only retain sample sequences with expected returns higher than the threshold, or only retain sample sequences with returns in the top few percent. The terminal uses the sample sequence consisting of the experimental sample set and the control sample set that have completed the secondary screening as the target sample set.

In this embodiment, by determining the control sample set that matches the experimental sample set through the propensity score, it can ensure that the experimental group and the control group are more balanced after the propensity score matching, improve the generalization of the data prediction model training, improve the data processing accuracy of the data prediction model, and thus improve the accuracy of data prediction.

In an exemplary embodiment, as shown in FIG5 , step 404 includes steps 502 to 504. Among them:

Step 502 : In each stratification type, for each experimental sample set in the resource conversion value, a target propensity score matching the propensity score of each user sample in the experimental sample set is retrieved according to a nearest neighbor matching algorithm.

In the embodiment of the present application, the terminal uses the nearest neighbor matching algorithm (Nearest Neighbour Matching), and for each stratified type of user, for each experimental sample set of resource conversion values, searches for a target propensity score that has a high degree of matching with the propensity score of the user sample in the current experimental sample set in other experimental sample sets of other resource conversion values. According to the matching degree of the propensity score, a certain threshold or criterion can be set. For example, if the difference between the nearest neighbor propensity scores is less than a certain threshold, the two samples are considered to match, which can ensure that the matching degree of the experimental group and the control group in terms of propensity reaches the expected level.

Step 504 , determining a control sample set that matches the experimental sample set based on the target propensity score.

In the embodiment of the present application, the terminal determines the user samples corresponding to the current experimental sample set in other experimental sample sets based on the user samples corresponding to the target propensity scores, and then obtains the control sample set.

In this embodiment, by determining the control sample set that matches the experimental sample set through the propensity score, it can ensure that the experimental group and the control group are more balanced after the propensity score matching, improve the generalization of the data prediction model training, improve the data processing accuracy of the data prediction model, and thus improve the accuracy of data prediction.

In an exemplary embodiment, as shown in FIG6 , step 406 includes steps 602 to 604. Among them:

Step 602: Calculate the standardized mean difference between each experimental sample set and the control sample set corresponding to the experimental sample set.

In the embodiment of the present application, the terminal uses the Standardized Mean Difference (SMD) to test whether the distribution of the covariates between the user samples matched in the experimental sample set and each control sample set is similar, wherein the standardized mean difference SMD of the covariate distribution between the user samples is calculated as follows:

in, and are the average characteristics of users under two different resource conversion values compared under a certain stratification type, and represents the respective characteristic variance.

Step 604 : when the standardized mean difference satisfies a preset threshold condition, each experimental sample set and a control sample set corresponding to the experimental sample set are determined as a target sample set.

In the embodiment of the present application, the terminal determines whether the standardized mean difference of the experimental sample set and the control sample set under each stratification type meets the training requirements according to the preset threshold conditions. If the standardized mean difference meets the preset threshold conditions, the experimental sample set and the control sample set have similar feature distributions under the stratification type and can be used as the target sample set. If the standardized mean difference of the experimental sample set and the control sample set under certain stratification types does not meet the preset threshold conditions, the user sample adjustment after propensity score matching or the recalculation of the propensity score can be performed, or the control sample can be reselected to ensure better matching of the experimental sample set.

In this embodiment, by calculating the standardized mean difference and determining the target sample set according to the preset threshold condition, the training effect of the data prediction model can be ensured while maintaining the balance of the experimental sample set and the control sample set under each stratification type, thereby improving the accuracy of data prediction.

It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

Based on the same inventive concept, the embodiment of the present application also provides a data prediction device for implementing the data prediction method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more data prediction device embodiments provided below can refer to the limitations on the data prediction method above, and will not be repeated here.

In an exemplary embodiment, as shown in FIG7 , a data prediction device 700 is provided, comprising: a first acquisition module 701, an analysis module 702 and a prediction module 703, wherein:

The first acquisition module 701 is used to acquire user characteristic data and resource conversion values of users of each hierarchical type;

An analysis module 702 is used to perform resource conversion sensitivity analysis and elastic data penetration index analysis on user feature data of each stratified type of user according to a data prediction model to obtain sensitivity parameters and elastic index parameters;

The prediction module 703 is used to predict the data penetration index results of the sensitivity parameters, elasticity index parameters and each resource conversion value based on the data prediction model, and obtain the data penetration index results of each layer type of users under each resource conversion value.

In one embodiment, the prediction module 703 is specifically used to perform data processing on elasticity index parameters according to the data evaluation structure in the data prediction model to obtain an initial data penetration index result;

Based on the sensitivity parameters, resource conversion values and data assessment structure, the initial data penetration index results are modified and constrained to obtain the data penetration index results of each stratified type of user under each resource conversion value.

In one embodiment, the apparatus 700 further includes:

The second acquisition module is used to obtain the neural network structure and initial sample set of the data prediction model to be trained;

A screening module is used to screen the initial sample set into an experimental sample set and a control sample set according to the propensity score matching rule, and to construct a target sample set based on the experimental sample set and the control sample set;

The training module is used to train the neural network structure according to the target sample set. When the neural network structure meets the preset iteration conditions, the data evaluation structure is combined with the neural network structure to obtain a trained data prediction model.

In one embodiment, the screening module is specifically used to perform prediction processing on the initial sample set according to the propensity score prediction model to obtain the propensity score corresponding to each user sample in the initial sample set;

The sample set of each stratified type of users in the initial sample set under each resource conversion value is used as the experimental sample set, and based on the propensity score, a control sample set matching the experimental sample set is determined from other experimental sample sets of other resource conversion values of the stratified type of users;

A target sample set is constructed based on each experimental sample set and a control sample set corresponding to the experimental sample set.

In one embodiment, the screening module is specifically used to retrieve, for each experimental sample set in each resource conversion value in each stratification type, a target propensity score that matches the propensity score of each user sample in the experimental sample set according to a nearest neighbor matching algorithm;

A control sample set matched to the experimental sample set was determined based on the target propensity score.

In one embodiment, the screening module is specifically used to calculate the standardized mean difference between each experimental sample set and a control sample set corresponding to the experimental sample set;

When the standardized mean difference meets the preset threshold condition, each experimental sample set and the control sample set corresponding to the experimental sample set are determined as the target sample set.

Each module in the above data prediction device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG8. The computer device includes a processor, a memory, an input/output interface (Input/Output, referred to as I/O) and a communication interface. The processor, the memory and the input/output interface are connected via a system bus, and the communication interface is connected to the system bus via the input/output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store user feature data. The input/output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a data prediction method is implemented.

Those skilled in the art will understand that the structure shown in FIG. 8 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

Obtain user feature data and resource conversion values of users of each stratification type;

According to the data prediction model, the resource conversion sensitivity analysis and elastic data penetration index analysis are performed on the user characteristic data of each stratified type of users to obtain the sensitivity parameters and elastic index parameters;

Based on the data prediction model, the data penetration index results are predicted for the sensitivity parameters, elasticity index parameters and each resource conversion value, and the data penetration index results of users of each stratification type under each resource conversion value are obtained.

In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

Perform data processing on elasticity index parameters according to the data evaluation structure in the data prediction model to obtain initial data penetration index results;

Based on the sensitivity parameters, resource conversion values and data assessment structure, the initial data penetration index results are modified and constrained to obtain the data penetration index results of each stratified type of user under each resource conversion value.

In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

Obtain the neural network structure and initial sample set of the data prediction model to be trained;

The initial sample set is screened for the experimental sample set and the control sample set according to the propensity score matching rule, and the target sample set is constructed based on the experimental sample set and the control sample set;

The neural network structure is trained according to the target sample set. When the neural network structure meets the preset iteration conditions, the data evaluation structure is combined with the neural network structure to obtain a trained data prediction model.

In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

The initial sample set is predicted according to the propensity score prediction model to obtain the propensity score corresponding to each user sample in the initial sample set;

The sample set of each stratified type of users in the initial sample set under each resource conversion value is used as the experimental sample set, and based on the propensity score, a control sample set matching the experimental sample set is determined from other experimental sample sets of other resource conversion values of the stratified type of users;

A target sample set is constructed based on each experimental sample set and a control sample set corresponding to the experimental sample set.

In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

In each stratification type, for each experimental sample set in the resource conversion value, a target propensity score matching the propensity score of each user sample in the experimental sample set is retrieved according to the nearest neighbor matching algorithm;

A control sample set matched to the experimental sample set was determined based on the target propensity score.

In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

Calculate the standardized mean difference between each experimental sample set and the control sample set corresponding to the experimental sample set;

When the standardized mean difference meets the preset threshold condition, each experimental sample set and the control sample set corresponding to the experimental sample set are determined as the target sample set.

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims (10)

1. A method of data prediction, the method comprising:

Acquiring user characteristic data and resource conversion values of users of each layering type;

Performing resource conversion sensitivity analysis and elastic data penetration index analysis on the user characteristic data of each layering type user according to a data prediction model to obtain sensitivity parameters and elastic index parameters;

And predicting the data permeation index result based on the data prediction model for the sensitivity parameter, the elasticity index parameter and the resource conversion values to obtain the data permeation index result of the user of each layering type under each resource conversion value.

2. The method according to claim 1, wherein the predicting the sensitivity parameter, the elasticity index parameter, and the resource conversion values based on the data prediction model to obtain the data penetration index result of each hierarchical type of user under each resource conversion value includes:

performing data processing on the elastic index parameters according to a data evaluation structure in the data prediction model to obtain an initial data penetration index result;

and correcting and restraining the initial data penetration index result based on the sensitivity parameter, the resource conversion value and the data evaluation structure to obtain the data penetration index result of each hierarchical type user under each resource conversion value.

3. The method of claim 1, wherein prior to the obtaining user characteristic data and resource conversion values for each hierarchical type of user, the method further comprises:

acquiring a neural network structure and an initial sample set of a data prediction model to be trained;

Screening an experimental sample set and a control sample set according to a tendency score matching rule, and constructing a target sample set based on the experimental sample set and the control sample set;

Training the neural network structure according to the target sample set, and combining the data evaluation structure on the basis of the neural network structure when the neural network structure meets the preset iteration condition to obtain a trained data prediction model.

4. A method according to claim 3, wherein said screening of the initial sample set for experimental and control sample sets according to a trend score matching rule and constructing a target sample set based on the experimental and control sample sets comprises:

Carrying out prediction processing on the initial sample set according to a tendency score prediction model to obtain tendency scores corresponding to each user sample in the initial sample set;

Taking a sample set of each hierarchical type user under each resource conversion value in the initial sample set as an experimental sample set, and determining a control sample set matched with the experimental sample set in other experimental sample sets of other resource conversion values of the hierarchical type user based on the tendency score;

And constructing a target sample set based on each experimental sample set and the control sample set corresponding to the experimental sample set.

5. The method of claim 4, wherein the determining a control sample set that matches the experimental sample set in other experimental sample sets of other resource conversion values of users of the hierarchical type based on the tendency score using the sample set of each hierarchical type user at each resource conversion value in the initial sample set as the experimental sample set comprises:

In each hierarchical type, searching a target tendency score matched with the tendency score of each user sample in the experimental sample set according to a nearest neighbor matching algorithm aiming at the experimental sample set in each resource conversion value;

A control sample set that matches the experimental sample set is determined based on the target propensity score.

6. The method of claim 4, wherein the constructing a target sample set based on each of the experimental sample set and the control sample set corresponding to the experimental sample set comprises:

Calculating a standardized mean difference between each experimental sample set and the control sample set corresponding to the experimental sample set;

And under the condition that the standardized mean value difference meets a preset threshold value condition, determining each experimental sample set and the control sample set corresponding to the experimental sample set as a target sample set.

7. A data prediction apparatus, the apparatus comprising:

The first acquisition module is used for acquiring user characteristic data and resource conversion values of users of each layering type;

The analysis module is used for carrying out resource conversion sensitivity analysis and elastic data penetration index analysis on the user characteristic data of the users of each layering type according to the data prediction model to obtain sensitivity parameters and elastic index parameters;

and the prediction module is used for predicting the data permeability index result of the sensitivity parameter, the elasticity index parameter and the resource conversion values based on the data prediction model to obtain the data permeability index result of the hierarchical type user under the resource conversion values.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 6 when the computer program is executed.

9. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.