Data estimation method and device

By decentralizing processing and building a preset fixed-effect model, using tool variables to replace endogenous independent variables, the parameter deviation caused by endogenous problems in regression analysis is solved, and the accuracy of user future behavior prediction is improved.

CN112380273BActive Publication Date: 2024-12-03BEIJING DAJIA INTERNET INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011255753.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-11-11
Publication Date
2024-12-03
Estimated Expiration
2040-11-11

AI Technical Summary

Technical Problem

The existing regression analysis methods cause deviations in parameter estimates due to endogenous problems, which leads to inaccurate user future behavior estimates.

Method used

By obtaining the user's historical behavior data and user portrait data, decentralized processing and regression analysis are carried out, pre-processed data related to the first sub-behavior data is used as tool variables to replace the first sub-behavior data, and a preset fixed effect model is constructed. It indicates the causal relationship between historical behavior data and target behavior data, and then estimates the user's future behavior.

Benefits of technology

The parameter estimation deviation caused by endogenous problems in regression analysis is solved, ensuring the consistency of the estimated parameters and improving the accuracy of user's future behavior estimates.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112380273B_ABST
    Figure CN112380273B_ABST
Patent Text Reader

Abstract

The present disclosure is about a data estimation method and device, the method includes obtaining historical behavior data and historical user portrait data of a user in a first historical time period, determining first pre-processed data; obtaining target behavior data of a user in a second historical time period; based on the historical user portrait data, decentralizing the first sub-behavior data, the second sub-behavior data, the target behavior data and the first pre-processed data to obtain the third sub-behavior data, the fourth sub-behavior data, the target behavior processing data and the second pre-processed data; regressing the second pre-processed data and the third sub-behavior data to obtain an estimated value of the third sub-behavior data; regressing the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain an estimated parameter; obtaining the current behavior data and current user portrait data of the user at the current moment, and estimating the user's future behavior data based on the estimated parameters. The present disclosure can solve the endogeneity problem and accurately estimate the user's future behavior data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of computer processing technology, and in particular to a data estimation method and device. Background Art

[0002] Currently, many business scenarios require the analysis of user behavior data. For example, in the video field such as short videos and live broadcasts, practical problems can be analyzed and solved by obtaining massive user behavior data. As a common analysis model, regression models are often used to study the relationship between dependent variables and independent variables. Regression can help explore the correlation between variables and obtain quantitative analysis results. For example, as live e-commerce gradually enters everyone's lives, in order to analyze the impact of users' purchasing behavior in e-commerce live broadcast rooms on their future reward behavior, regression analysis is usually performed using the user's purchase amount in the live broadcast room as the independent variable and the user's future reward amount as the dependent variable.

[0003] However, the above regression analysis method may cause endogeneity problems due to omitted variables or selection bias. For example, the amount of purchases made by a user on that day (independent variable) and the amount of rewards given afterwards (dependent variable) are affected by the user's personal characteristics. Endogeneity problems may cause parameter estimates to be biased and inconsistent. The consistency of parameter estimates means that when the sample size is large, the parameters estimated by the sample will approach the true parameters of the population infinitely. When the consistency is destroyed, the estimation result loses its reference value.

[0004] It can be seen that the regression analysis method in the relevant technology may cause deviations in parameter estimates due to endogeneity problems, which in turn leads to inaccurate predictions of users' future behaviors. Summary of the invention

[0005] The present disclosure provides a data prediction method and device to at least solve the problem that the regression analysis method in the related art may cause parameter estimation deviation due to endogenous problems, thereby causing inaccurate prediction of user future behavior. The technical solution of the present disclosure is as follows:

[0006] According to a first aspect of an embodiment of the present disclosure, a data estimation method is provided, comprising:

[0007] Acquire historical behavior data and historical user portrait data of a user in a first historical time period, and determine first preprocessing data related to first sub-behavior data, wherein the historical behavior data includes the first sub-behavior data and the second sub-behavior data;

[0008] Acquire target behavior data of the user in a second historical time period, wherein the second historical time period is a time period after the first historical time period;

[0009] Based on the historical user portrait data, the first sub-behavior data, the second sub-behavior data, the target behavior data and the first pre-processed data are decentralized to obtain corresponding third sub-behavior data, fourth sub-behavior data, target behavior processed data and second pre-processed data;

[0010] Performing regression processing on the second preprocessed data and the third sub-behavior data to obtain an estimated value of the third sub-behavior data;

[0011] Performing regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain an estimated parameter, wherein the estimated parameter is used to represent the causal relationship between the historical behavior data and the target behavior data;

[0012] The current behavior data and current user portrait data of the user at the current moment are obtained, and based on the estimated parameters, the user's future behavior data is estimated.

[0013] Optionally, the acquiring of historical behavior data and historical user portrait data of the user in the first historical time period, and determining first preprocessing data related to the first sub-behavior data, includes:

[0014] Preprocessing the service function of the service to be analyzed within the first historical time period to obtain first preprocessed data of the user;

[0015] Obtain the preprocessed historical behavior data of the user in the first historical time period, and obtain historical user portrait data, wherein the first preprocessed data is an instrumental variable of the first sub-behavior data.

[0016] Optionally, performing regression processing on the second preprocessed data and the third sub-behavior data to obtain an estimated value of the third sub-behavior data includes:

[0017] Performing regression processing on the second preprocessed data, the fourth sub-behavior data, and the third sub-behavior data to obtain coefficient estimates of the second preprocessed data and the fourth sub-behavior data;

[0018] Based on the coefficient estimate, the second pre-processed data and the fourth sub-behavior data, a third sub-behavior data estimate is determined.

[0019] Optionally, performing regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processed data to obtain the estimated parameters includes:

[0020] A multivariate linear regression model is used to perform regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain model parameters of the multivariate linear regression model, and the model parameters of the multivariate linear regression model are used as the estimated parameters.

[0021] Optionally, the target behavior data is an n-dimensional column vector, the historical behavior data is an n×k matrix, and the historical user portrait data is an n×r matrix, where n is the number of users, k is the number of historical behavior data, and r is the number of categories of historical user portrait data, and n, k, and r are all positive integers;

[0022] The decentralizing the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data based on the historical user portrait data includes:

[0023] For each column of the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data, the data are grouped according to a column of the historical user portrait data and the average is calculated each time, and the average of the group to which the data belongs is subtracted from each data in each column of the data until the data in each column of the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data no longer changes, wherein each column of data corresponding to the same historical user portrait data is a group.

[0024] Optionally, after performing regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain the estimated parameters, before obtaining the current behavior data and the current user portrait data of the user at the current moment and estimating the user's future behavior data based on the estimated parameters, the method further includes:

[0025] Based on the estimated parameters, a preset fixed effect model is constructed, wherein the endogenous independent variable of the preset fixed effect model is the first sub-behavior data of the user, the exogenous independent variable of the preset fixed effect model is the second sub-behavior data of the user, the dependent variable of the preset fixed effect model is the target behavior data of the user, and the model parameters of the preset fixed effect model are the estimated parameters;

[0026] The obtaining of the current behavior data and the current user portrait data of the user at the current moment, and estimating the user's future behavior data based on the estimation parameters, includes:

[0027] The current behavior data and current user portrait data of the user at the current moment are obtained, and the user's future behavior data is estimated using the preset fixed effect model.

[0028] Optionally, constructing a preset fixed effect model based on the estimated parameters includes:

[0029] Determine a target behavior processing data estimation value based on the third sub-behavior data, the fourth sub-behavior data and the estimated parameter;

[0030] Determining a residual term in the preset fixed effect model based on a difference between the target behavior processed data and an estimated value of the target behavior processed data;

[0031] Based on the estimated parameters and the residual terms, the preset fixed effect model is constructed.

[0032] Optionally, after determining the residual term in the preset fixed effect model, the method further includes:

[0033] Obtain the correlation between each type of historical user portrait data, and cluster each column of historical user portrait data based on the correlation;

[0034] Determine the covariance matrix of the model parameters according to the residual items corresponding to the historical user portrait data after clustering;

[0035] Based on the diagonal elements in the covariance matrix, estimated standard errors of the model parameters are determined.

[0036] Optionally, the fixed effect variable of the preset fixed effect model is user portrait data;

[0037] After determining the residual term in the preset fixed effect model and before constructing the preset fixed effect model based on the estimated parameters and the residual term, the method further includes:

[0038] Determining fixed effect parameters in the preset fixed effect model based on the historical behavior data, the target behavior data, the historical user portrait data, the model parameters and the residual term;

[0039] The step of constructing a preset fixed effect model based on the estimated parameters and the residual term includes:

[0040] The preset fixed effect model is constructed based on the estimated parameters, the residual terms and the fixed effect parameters.

[0041] Optionally, the residual term is an n-dimensional column vector; after determining the fixed effect parameters in the preset fixed effect model, the method further includes:

[0042] Sampling n data in the residual term n times to obtain a target residual term consisting of n sampled data, and repeating the process of sampling n data in the residual term n times until m target residual terms are obtained, where n is a positive integer and m is a positive integer;

[0043] Determining m fixed effect parameter estimates based on the historical behavior data, the target behavior data, the historical user profile data, the model parameters, and the m target residual terms;

[0044] Based on the m fixed effect parameter estimates, the estimated standard errors of the fixed effect parameters are determined.

[0045] According to a second aspect of an embodiment of the present disclosure, there is provided a data estimation device, comprising:

[0046] A first acquisition module is configured to acquire historical behavior data and historical user portrait data of a user in a first historical time period, and determine first preprocessing data related to first sub-behavior data, wherein the historical behavior data includes the first sub-behavior data and the second sub-behavior data;

[0047] A second acquisition module is configured to execute acquisition of target behavior data of users in a second historical time period, wherein the second historical time period is a time period after the first historical time period;

[0048] A first processing module is configured to perform decentralized processing on the first sub-behavior data, the second sub-behavior data, the target behavior data and the first pre-processed data based on the historical user portrait data to obtain corresponding third sub-behavior data, fourth sub-behavior data, target behavior processing data and second pre-processed data;

[0049] A second processing module is configured to perform regression processing on the second pre-processed data and the third sub-behavior data to obtain an estimated value of the third sub-behavior data;

[0050] a third processing module, configured to perform regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain an estimated parameter, wherein the estimated parameter is used to represent a causal relationship between the historical behavior data and the target behavior data;

[0051] The estimation module is configured to obtain the current behavior data and current user portrait data of the user at the current moment, and estimate the user's future behavior data based on the estimation parameters.

[0052] Optionally, the first acquisition module includes:

[0053] A first processing unit is configured to perform preprocessing on a service function of the service to be analyzed within the first historical time period, and obtain first preprocessing data of the user;

[0054] The acquisition unit is configured to execute acquisition of the preprocessed historical behavior data of the user in the first historical time period, and to acquire historical user portrait data, wherein the first preprocessed data is an instrumental variable of the first sub-behavior data.

[0055] Optionally, the second processing module includes:

[0056] A second processing unit is configured to perform regression processing on the second preprocessed data, the fourth sub-behavior data and the third sub-behavior data to obtain coefficient estimates of the second preprocessed data and the fourth sub-behavior data;

[0057] The first determining unit is configured to determine a third sub-behavior data estimation value based on the coefficient estimation, the second pre-processed data and the fourth sub-behavior data.

[0058] Optionally, the third processing module is configured to perform regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data using a multivariate linear regression model to obtain model parameters of the multivariate linear regression model, and use the model parameters of the multivariate linear regression model as the estimated parameters.

[0059] Optionally, the target behavior data is an n-dimensional column vector, the historical behavior data is an n×k matrix, and the historical user portrait data is an n×r matrix, where n is the number of users, k is the number of historical behavior data, and r is the number of categories of historical user portrait data, and n, k, and r are all positive integers;

[0060] The first processing module is configured to perform averaging on each column of the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data, each time by grouping one column of data in the historical user portrait data, and subtracting the mean of the group to which the data belongs from each data in each column of data until each column of the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data no longer changes, wherein each column of data corresponding to the same historical user portrait data is a group.

[0061] Optionally, the data estimation device further includes:

[0062] A construction module is configured to execute construction of a preset fixed effect model based on the estimated parameters, wherein the endogenous independent variable of the preset fixed effect model is the first sub-behavior data of the user, the exogenous independent variable of the preset fixed effect model is the second sub-behavior data of the user, the dependent variable of the preset fixed effect model is the target behavior data of the user, and the model parameters of the preset fixed effect model are the estimated parameters;

[0063] The estimation module is configured to obtain the current behavior data and current user portrait data of the user at the current moment, and use the preset fixed effect model to estimate the user's future behavior data.

[0064] Optionally, the building blocks include:

[0065] A second determining unit is configured to determine a target behavior processing data estimation value based on the third sub-behavior data, the fourth sub-behavior data and the estimated parameter;

[0066] A third determining unit is configured to determine a residual term in the preset fixed effect model based on a difference between the target behavior processed data and the target behavior processed data estimate;

[0067] A construction unit is configured to construct the preset fixed effect model based on the estimated parameters and the residual term.

[0068] Optionally, the data estimation device further includes:

[0069] A clustering module is configured to obtain the correlation between each type of historical user portrait data, and cluster each column of historical user portrait data based on the correlation;

[0070] A first determination module is configured to determine the covariance matrix of the model parameters according to the residual term corresponding to the clustered historical user portrait data;

[0071] The second determination module is configured to determine the estimated standard errors of the model parameters based on the diagonal elements in the covariance matrix.

[0072] Optionally, the fixed effect variable of the preset fixed effect model is user portrait data;

[0073] The data estimation device also includes:

[0074] A third determination module is configured to determine the fixed effect parameters in the preset fixed effect model based on the historical behavior data, the target behavior data, the historical user portrait data, the model parameters and the residual term;

[0075] The construction module is configured to construct the preset fixed effect model based on the estimated parameters, the residual terms and the fixed effect parameters.

[0076] Optionally, the residual term is an n-dimensional column vector;

[0077] The data estimation device also includes:

[0078] a sampling module configured to perform n sampling on the n data in the residual term n times to obtain a target residual term consisting of the n sampled data, and repeat the process of performing n sampling on the n data in the residual term n times until m target residual terms are obtained, where n is a positive integer and m is a positive integer;

[0079] A fourth determination module is configured to determine m fixed effect parameter estimates based on the historical behavior data, the target behavior data, the historical user portrait data, the model parameters and the m target residual terms;

[0080] The fifth determination module is configured to determine the estimated standard error of the fixed effect parameter based on the m fixed effect parameter estimates.

[0081] According to a third aspect of an embodiment of the present disclosure, a data estimation device is provided, including:

[0082] processor;

[0083] a memory for storing instructions executable by the processor;

[0084] The processor is configured to execute the instructions to implement the data estimation method described in the first aspect above.

[0085] According to a fourth aspect of an embodiment of the present disclosure, a storage medium is provided. When instructions in the storage medium are executed by a data estimation device, the data estimation device can execute the data estimation method described in the first aspect above.

[0086] According to a fifth aspect of an embodiment of the present disclosure, a computer program product is provided, comprising executable instructions, which, when executed on a computer, enable the computer to execute the data estimation method described in the first aspect.

[0087] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects:

[0088] Acquire the historical behavior data and historical user portrait data of the user in a first historical time period, and determine the first preprocessing data related to the first sub-behavior data, wherein the historical behavior data includes the first sub-behavior data and the second sub-behavior data; acquire the target behavior data of the user in a second historical time period, wherein the second historical time period is the time period after the first historical time period; based on the historical user portrait data, decentralize the first sub-behavior data, the second sub-behavior data, the target behavior data and the first preprocessing data to obtain the corresponding third sub-behavior data, fourth sub-behavior data, target behavior processing data and second preprocessing data; perform regression processing on the second preprocessing data and the third sub-behavior data to obtain an estimated value of the third sub-behavior data; perform regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain an estimated parameter, wherein the estimated parameter is used to represent the causal relationship between the historical behavior data and the target behavior data; acquire the current behavior data and current user portrait data of the user at the current moment, and estimate the user's future behavior data based on the estimated parameters. In this way, by using the first preprocessed data related to the first sub-behavior data as an instrumental variable to replace the first sub-behavior data to regress the target behavior data, the problem of variable endogeneity can be solved, thereby ensuring that the estimated parameters obtained by regression are consistent, and then ensuring that based on the estimated parameters and the user's current behavior data and portrait data, the user's future behavior data can be more accurately predicted.

[0089] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0091] Figure 1 The figure is a flow chart of a data estimation method according to an exemplary embodiment.

[0092] Figure 2 The present invention is a flowchart showing a method of using a fixed effect model to analyze the causal relationship between user historical behavior data and target behavior data in a short video application according to an exemplary embodiment.

[0093] Figure 3 It is a block diagram of a data estimation device according to an exemplary embodiment.

[0094] Figure 4 It is a block diagram of a device according to an exemplary embodiment. DETAILED DESCRIPTION

[0095] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0096] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0097] This solution can be applied to application scenarios that require analysis and prediction of user behavior, for example, it can be applied to audio and video (such as short videos), live broadcast and other video fields. The video field is only an example and is not limited by the embodiments of the present disclosure. In one application scenario, it can be used to analyze the impact of the first digital currency transfer behavior (such as purchase behavior data, in one example, the purchase behavior data can be the purchase amount) in the user's live broadcast room on the future second digital currency transfer behavior (such as reward behavior data, in one example, the reward behavior data can be the reward amount). The embodiment of the present disclosure uses the pre-processed data related to the first sub-behavior data (such as whether to block the live broadcast reminder of the user's attention page) as an instrumental variable during the analysis to solve the endogeneity problem that exists when the first sub-behavior data is used as an independent variable in the regression analysis, and adds the influence of user portrait data (i.e., fixed effect), which can improve the accuracy and credibility of the analysis results, and can help solve the situation where the consistency of coefficient estimation is destroyed due to endogeneity problems, thereby obtaining more reliable causal conclusions and more accurate estimated data.

[0098] Figure 1 is a flow chart of a data estimation method according to an exemplary embodiment. Figure 1 As shown, the method includes the following steps.

[0099] In step S11, historical behavior data and historical user portrait data of the user in a first historical time period are obtained, and first preprocessing data related to first sub-behavior data is determined, wherein the historical behavior data includes the first sub-behavior data and the second sub-behavior data.

[0100] In step S12, target behavior data of the user in a second historical time period is obtained, wherein the second historical time period is a time period after the first historical time period.

[0101] The above-mentioned historical behavior data of the user may be some behavior data generated by the user when using the target application, and the above-mentioned target behavior data may be other behavior data generated by the user when using the target application, and the historical behavior data of the user for a certain period of time will have a certain impact on the user's subsequent target behavior data, that is, there is a certain causal relationship between the two. The target application can be any application that needs to study the impact of certain behavior data of the user on another behavior data during the user's use. For example, when it is necessary to analyze the impact of the user's purchase amount, viewing time, etc. in the live broadcast room on the future reward amount when the user uses the live broadcast application, the live broadcast application is the target application, the user's purchase amount, viewing time, etc. in the live broadcast room are historical behavior data, and the reward amount is the target behavior data.

[0102] Among them, the historical behavior data includes first sub-behavior data and second sub-behavior data. In analyzing the causal relationship between the historical behavior data and the target behavior data, the historical behavior data is the independent variable and the target behavior data is the dependent variable. In the embodiment of the present disclosure, in order to solve the common variable endogeneity problem in causal analysis, it is necessary to select appropriate instrumental variables to replace the independent variables with endogeneity problems for regression analysis. Therefore, after determining the historical behavior data as the independent variable, it is necessary to further determine which behavior data are endogenous independent variables and which behavior data are exogenous independent variables. Therefore, the first sub-behavior data and the second sub-behavior data in the historical behavior data can refer to endogenous independent variables and exogenous independent variables, respectively, and which specific behavior data are endogenous independent variables and which variables are exogenous independent variables can be set accordingly according to actual conditions. For example, when analyzing the impact of a user's live broadcast room purchase amount on the subsequent reward amount, since the purchase amount will also be affected by the user's personal characteristics, that is, there is an endogeneity problem when the purchase amount is used as an independent variable, the purchase amount can be used as an endogenous independent variable in the analysis. The first sub-behavior data can refer to the purchase amount, and the second sub-behavior data can refer to other live broadcast room behavior data that affects the reward amount and does not have an endogeneity problem.

[0103] The first historical time period may be any historical time period, such as yesterday, the day before yesterday, or the past week, and the second historical time period may be a time period after the first historical time period, that is, the user's target behavior data is generated after the historical behavior data. For example, when it is necessary to analyze the impact of the user's first day's live broadcast purchase amount, viewing time, etc. on the user's second day's reward amount, the first historical time period is the first day, and the second historical time period is the second day.

[0104] The above-mentioned historical user portrait data can be data that characterizes the individual characteristics of the user who generated the above-mentioned historical behavior data and target behavior data, for example, the user's geographical characteristics, age, gender, model used, ID, preferences, etc.

[0105] In one embodiment, the first sub-behavior data, the second sub-behavior data and the target behavior data may be some operation behavior data generated by the user when using the target application, such as Figure 2 As shown, taking the use of live broadcast applications as an example, the first sub-behavior data may include the purchase amount in the live broadcast room, the second sub-behavior data may include the live broadcast viewing time, the target behavior data may be the reward amount in the live broadcast room, and the user portrait data may include age, gender, region, model, time, ID, etc.

[0106] In the embodiment of the present disclosure, the user behavior data and user portrait data for the specific historical time period that needs to be obtained can be selected according to actual needs.

[0107] In the related art, when directly using the user's historical behavior data (as an independent variable) to perform regression analysis on the user's target behavior data (as a dependent variable), there will be an endogeneity problem of the dependent variable, such as the user's daily purchase amount (independent variable) and the subsequent reward behavior (dependent variable) are affected by the user's personal characteristics (user portrait data), which destroys the consistency of parameter estimation. Therefore, in the embodiment of the present disclosure, in order to solve the above-mentioned endogeneity problem, for the endogenous independent variable (such as the purchase amount) in the regression analysis, the pre-processed data with correlation therewith is used as an instrumental variable, and the instrumental variable is used to replace the endogenous independent variable, and the dependent variable is regressed, which can help users obtain more accurate quantitative analysis results. Among them, the endogenous independent variable can be determined according to the actual situation in the specific application. When it is found that there is an endogeneity problem when a certain data is used as an independent variable, such as being related to the omitted independent variable, it can be used as an endogenous independent variable.

[0108] Among them, the above-mentioned first preprocessed data can be data generated by randomly preprocessing the user before the user generates the historical behavior data, and the first preprocessed data will have a greater impact on the first sub-behavior data (endogenous independent variable) in the user's historical behavior data, but will not affect other behavior data in the historical behavior data except the first sub-behavior data, such as the second sub-behavior data (exogenous independent variable), that is, the first preprocessed data is highly correlated with the first sub-behavior data, but is not correlated with the second sub-behavior data. Therefore, the above-mentioned first preprocessed data related to the first sub-behavior data can be determined based on whether the user accounts that generate the historical behavior data have undergone the preprocessing. For example, 0 and 1 are used to represent no preprocessing and preprocessing, respectively, so that the first preprocessed data corresponding to the user who has undergone preprocessing can be recorded as 1, and the first preprocessed data corresponding to the user who has not undergone preprocessing can be recorded as 0.

[0109] That is, the acquiring of the historical behavior data and the historical user portrait data of the user in the first historical time period and determining the first pre-processed data related to the first sub-behavior data may include:

[0110] Preprocessing the service function of the service to be analyzed within the first historical time period to obtain first preprocessed data of the user;

[0111] Obtain the preprocessed historical behavior data of the user in the first historical time period, and obtain historical user portrait data, wherein the first preprocessed data is an instrumental variable of the first sub-behavior data.

[0112] The above-mentioned business to be analyzed may refer to a business that currently needs to analyze the causal relationship between business data, for example, a live broadcast business, which currently needs to analyze the causal relationship between the user's live broadcast room purchase amount and the subsequent reward amount in the live broadcast business.

[0113] In this implementation, in order to obtain the first preprocessed data as an instrumental variable, a certain business function of the business to be analyzed can be preprocessed to affect the first sub-behavior data of the user in the business to be analyzed through the preprocessing, that is, the business function to be preprocessed is a function that may have an impact on the first sub-behavior data of the user. In practical applications, based on the user behavior data as an endogenous independent variable in the business to be analyzed, it is possible to analyze which business function can be processed to significantly affect the user behavior data, and then determine the business function that needs to be preprocessed, and the effectiveness of the obtained instrumental variable can be verified using relevant data after preprocessing, that is, by regressing the first sub-behavior data of the user after preprocessing using the preprocessed data, verifying whether the coefficient estimate γ1 obtained by the regression is significant, such as by calculating and judging whether its estimated standard error is less than a certain value to determine whether it is significant. If it is significant, it means that the instrumental variable is valid, otherwise it is necessary to reselect other business functions for preprocessing to obtain a valid instrumental variable.

[0114] Taking the live broadcast business as an example, the first sub-behavior data is the purchase amount of the live broadcast room generated by the user when using the live broadcast business. The pre-processed business function is the live broadcast reminder function in the live broadcast business. By blocking the live broadcast reminder function in the live broadcast business used by the user, the exposure of the live broadcast room can be reduced, and the channel for the user to enter the live broadcast room can be closed, which reduces the possibility of the user entering the live broadcast room, and can directly affect the user's purchase amount in the live broadcast room.

[0115] The above-mentioned preprocessing of the service functions of the service to be analyzed may be to randomly preprocess the service functions used by some users, and not process the service functions used by some users, to observe whether the first sub-behavior data of the preprocessed users is significantly different from the first sub-behavior data of the unpreprocessed users. If there is a significant difference, it means that the preprocessing is highly correlated with the first sub-behavior data.

[0116] Specifically, a group test (ABtset) can be used to randomly divide multiple users into a control group and an experimental group. The service functions used by the control group users are not preprocessed, and the service functions used by the experimental group users are preprocessed. Then, the historical behavior data of the control group users and the experimental group users in the first historical period, as well as the historical user portrait data, are obtained respectively. The historical behavior data includes the first sub-behavior data and the second sub-behavior data. Among them, 0 and 1 can be used to represent no processing and preprocessing, respectively, then the preprocessed data of the control group users is 0, and the preprocessed data of the experimental group users is 1. Of course, the user's preprocessed data can also be used to regress its first sub-behavior data, and the coefficient estimate obtained by analysis is significantly not equal to 0 to verify that the preprocessing is highly correlated with the first sub-behavior data.

[0117] For example, still taking the analysis of the impact of the user's live broadcast room purchase amount on the future reward amount as an example, the preprocessing can be to block the live broadcast reminder function in the live broadcast service used by the user, such as blocking the live broadcast reminder that the user is concerned about or blocking the live broadcast entrance of the user's attention page. In this way, by closing the reminder of the live broadcast room that the user is concerned about, the probability of the user entering the live broadcast room will be greatly reduced, which in turn will cause a significant change in the user's live broadcast room purchase amount. At the same time, for each user, whether or not to fall into the blocking experiment is completely random. Therefore, for a completely random variable (the first preprocessed data), it must be uncorrelated with the model residual term, so that the first preprocessed data can be used as an instrumental variable to solve the endogeneity problem that exists when the first sub-behavior data is used as an independent variable.

[0118] It should be noted that when there are multiple endogenous independent variables in the regression analysis, that is, when multiple behavioral data in the historical behavioral data as independent variables are endogenous, multiple instrumental variables can be used to replace the multiple endogenous independent variables respectively, and each instrumental variable is independent of each other, and each instrumental variable is only highly correlated with the independent variable it replaces. The disclosed embodiment only takes the first sub-behavior data as an endogenous independent variable as an example for explanation, but it is not limited to the case of only one endogenous independent variable. When there are multiple endogenous independent variables, a similar processing method can be used to solve the endogeneity problem.

[0119] In addition, the second sub-behavior data in the embodiment of the present disclosure may refer to other behavior data in the user's historical behavior data except for the first sub-behavior data that is endogenous, that is, the second sub-behavior data may be other exogenous behavior data, and the second sub-behavior data may not be limited to a certain type of behavior data, that is, it may include multiple types of behavior data, that is, in the embodiment of the present disclosure, the first sub-behavior data may refer to the endogenous independent variable in the historical behavior data as an independent variable, and the second sub-behavior data may refer to the exogenous independent variable in the historical behavior data as an independent variable. Among them, endogeneity means that the independent variable in the model is related to the model residual term, and exogeneity means that the independent variable in the model is not related to the model residual term. Accordingly, the independent variable related to the model residual term is called an endogenous independent variable, and the independent variable unrelated to the model residual term is called an exogenous independent variable.

[0120] It should also be noted that the embodiment of the present disclosure does not limit the execution order of the above-mentioned step S11 and step S12. The above-mentioned step S11 and step S12 can be executed successively or in parallel. For example, step S11 can be executed first and then step S12, or step S12 can be executed first and then step S11, or step S11 and step S12 can be executed at the same time.

[0121] In step S13, based on the historical user portrait data, the first sub-behavior data, the second sub-behavior data, the target behavior data and the first pre-processed data are decentralized to obtain corresponding third sub-behavior data, fourth sub-behavior data, target behavior processing data and second pre-processed data.

[0122] Since user portrait data is user individual characteristic data, usually when a user uses an application, the user's usage habits of the application will not change in the short term, that is, it can have a relatively fixed impact on the user's behavior data of the application, which we can call a fixed effect variable. Therefore, in order to accurately analyze the impact of the user's historical behavior data on its target behavior data, the impact of the user portrait data on the target behavior data should be removed. To this end, we can eliminate the impact of the user's individual characteristic data on the target behavior data based on the historical user portrait data.

[0123] In one embodiment, the historical behavior data and the target behavior data can be decentralized based on the historical user portrait data to remove the influence of the historical user portrait data on the target behavior data, and obtain the third sub-behavior data, fourth sub-behavior data and target behavior processing data corresponding to the first sub-behavior data, the second sub-behavior data and the target behavior data respectively after the decentralized processing, so as to facilitate the regression analysis of the target behavior processing data based on the third sub-behavior data and the fourth sub-behavior data. Similarly, the first pre-processed data can also be decentralized based on the historical user portrait data to obtain the second pre-processed data, so as to perform a regression analysis on the third sub-behavior data based on the second pre-processed data to determine the influence coefficient of the first pre-processed data on the first sub-behavior data.

[0124] Among them, the decentralized processing can be to group and average the data sets of multiple users (including first sub-behavior data, second sub-behavior data, target behavior data and first preprocessed data) according to the historical user portrait data of multiple users, that is, the data sets of users with the same user portrait data (such as the same age or gender) are divided into a group, and the mean of each group is calculated, and then the data set of each group is subtracted from the mean of the group. The final result is a data set after removing the influence of the user portrait data on the target behavior data.

[0125] Optionally, the target behavior data is an n-dimensional column vector, the historical behavior data is an n×k matrix, and the historical user portrait data is an n×r matrix, where n is the number of users, k is the number of historical behavior data, and r is the number of categories of historical user portrait data, and n, k, and r are all positive integers;

[0126] The decentralizing the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data based on the historical user portrait data includes:

[0127] For each column of the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data, the following processing is repeatedly performed: grouping and averaging a column of the historical user portrait data, and subtracting the mean of the group to which the data belongs from each data in each column of the data;

[0128] Until each column of data in the first sub-behavior data, the second sub-behavior data, the target behavior data and the first preprocessed data no longer changes, wherein each column of data corresponding to the same historical user portrait data is a group.

[0129] In one embodiment, when obtaining the historical behavior data, target behavior data and historical user portrait data of a user, k types of historical behavior data of n users may be obtained, 1 type of target behavior data of the n users may be obtained, and r types of historical user portrait data of the n users may be obtained. Accordingly, the first preprocessed data of the n users may be determined, wherein the target behavior data is different from the k types of historical behavior data, and the k types of historical behavior data may include first sub-behavior data and second sub-behavior data. When the types of the first sub-behavior data and the second sub-behavior data are both 1, k is equal to 2. Specifically, the target behavior data of the n users may be processed into an expression form of an n-dimensional column vector. Similarly, the k types of historical behavior data of the n users may be processed into an n×k matrix, wherein each row of data in the matrix represents k types of historical behavior data of a certain user among the n users, and each column of data in the matrix sequentially represents a certain type of user behavior data of the n users. When each user has only one first preprocessed data, the first preprocessed data of the n users may also be processed into an expression form of an n-dimensional column vector.

[0130] Among them, n, k and r are all positive integers. In one embodiment, n, k and r are all integers greater than 1, that is, multiple historical behavior data of multiple users and multiple types of historical user portrait data of these multiple users can be obtained, and more reliable analysis results can be obtained by performing regression analysis on as much data as possible.

[0131] In this implementation, in order to avoid the problem of excessive memory usage caused by generating high-dimensional categorical variables, the r types of historical user portrait data of the n users can be processed into an n×r fixed effect variable matrix, and each column of data in the matrix represents a type of historical user portrait data of the n users. For example, the user portrait data has three categories: age, gender, and region, then 3 columns of data can be generated, representing the age, gender, and region of the n users, respectively.

[0132] Then, the historical behavior data, the target behavior data, and the first pre-processed data may be decentralized based on the historical user portrait data. Specifically, the historical behavior data, the target behavior data, and each column of data in the first pre-processed data may be grouped according to the first column of data in the historical user portrait data, and the data with the same value in the first column of data corresponding to the historical user portrait data may be grouped into one group (for example, if the first column of historical user portrait data is age, the historical behavior data of users of the same age may be grouped into one group, the target behavior data of users of the same age may be grouped into one group, and the first pre-processed data of users of the same age may be grouped into one group). The processed data is divided into one group), and the mean of each group is calculated, and then each data in each column of data is subtracted from the mean of the group to obtain a processed data set; then according to the second column of data in the historical user portrait data, the same grouping, averaging and mean subtraction steps are repeated until each column of data in the processed data set converges and no longer changes, that is, the mean of each column of data after grouping according to any column of data in the historical user portrait data is 0, and finally a data set is obtained after removing the influence of the historical user portrait data on the target behavior data, that is, the third sub-behavior data, the fourth sub-behavior data, the target behavior processing data and the second preprocessing data.

[0133] For example, for each column of the historical behavior data, the data can be grouped by age first, and the behavior data of the same age can be grouped into one group. The mean of each column of behavior data after grouping is calculated, and then the mean of the corresponding group is subtracted from each behavior data in each column of behavior data, that is, the historical behavior data after removing the influence of the age variable on the target behavior data. For example, three users of the same age purchase 100 yuan, 200 yuan and 300 yuan respectively in the live broadcast room of a live broadcast platform on a certain day, then the purchase amount of these three users can be averaged, specifically (100+200+300) / 3=200, and the mean is the average purchase amount of users of this age. Subtract the mean of 200 yuan from the purchase amount of each user, and the purchase amounts of these three users after processing are -100 yuan, 0 yuan, and 100 yuan respectively. At this time, the mean of the group is 0, and the purchase amount after processing is the purchase amount after removing the influence of the user's age; then, it can be grouped by gender, and the behavior data of the same gender can be grouped together, and similar averaging and mean subtraction processing can be performed to obtain the historical behavior data after removing the influence of the gender variable on the target behavior data. Repeat similar steps until the influence of each type of historical user portrait data on the target behavior data is removed, and the processed historical behavior data does not change relative to the previous historical behavior data. Corresponding processing is also performed on the target behavior data and each column of data in the first pre-processed data.

[0134] In this way, by performing the above-mentioned processing on the historical behavior data, the target behavior data and the first preprocessed data according to the historical user portrait data, the fixed effect influence of the historical user portrait data on the target behavior data can be eliminated, thereby ensuring that by performing corresponding regression processing on the processed data set, estimated parameters that can more accurately represent the causal relationship between the historical behavior data and the target behavior data can be obtained.

[0135] In step S14, regression processing is performed on the second preprocessed data and the third sub-behavior data to obtain an estimated value of the third sub-behavior data.

[0136] In the embodiment of the present disclosure, in order to overcome the endogeneity problem of the first sub-behavior data, the first pre-processed data can be used to replace the first sub-behavior data to perform regression analysis on the target behavior data. Therefore, after the user's data set is decentralized, the second pre-processed data can be used to perform a first-stage regression processing on the third sub-behavior data, establish a regression model between the second pre-processed data and the third sub-behavior data, and determine the corresponding model coefficients. Then, based on the known model and the second pre-processed data, the estimated value of the third sub-behavior data can be calculated, that is, the estimated value of the third sub-behavior data is obtained.

[0137] Optionally, the step S14 includes:

[0138] Performing regression processing on the second preprocessed data, the fourth sub-behavior data, and the third sub-behavior data to obtain coefficient estimates of the second preprocessed data and the fourth sub-behavior data;

[0139] Based on the coefficient estimate, the second pre-processed data and the fourth sub-behavior data, a third sub-behavior data estimate is determined.

[0140] In one embodiment, when analyzing the impact of the first preprocessed data on the first sub-behavior data, the correlation between the historical behavior data can also be considered, that is, the first sub-behavior data may be correlated with the second sub-behavior data. Therefore, the second preprocessed data and the fourth sub-behavior data after decentralized processing can be used to perform the first stage regression processing on the third sub-behavior data. Specifically, a multivariate regression model between the second preprocessed data, the fourth sub-behavior data and the third sub-behavior data can be established, and then the corresponding model coefficients can be determined by the least squares method, etc. Then, based on the determined model coefficients, the second preprocessed data and the fourth sub-behavior data can be substituted into the model for calculation to obtain the corresponding third sub-behavior data estimate.

[0141] For example, the multivariate regression model of the first stage is X′1=γ1Q′+γ2X′2, where X′1 represents the third sub-behavior data obtained after the first sub-behavior data X1 is decentralized, Q′ represents the second pre-processed data obtained after the first pre-processed data Q is decentralized, Q is the instrumental variable of X1, X′2 represents the fourth sub-behavior data obtained after the second sub-behavior data X2 is decentralized, γ1 and γ2 are model coefficients, and the third sub-behavior data is subjected to least squares regression using the second pre-processed data and the fourth sub-behavior data to determine the coefficient estimates of γ1 and γ2, where γ1 should be significant, and if the obtained γ2 is not significant, such as close to or equal to 0, it means that there is almost no correlation between the first sub-behavior data and the second sub-behavior data, and if γ2 is more significant, it means that there is a certain correlation between the first sub-behavior data and the second sub-behavior data. Then, by substituting the second pre-processed data and the fourth sub-behavior data into the determined model for calculation, the estimated value of the third sub-behavior data can be obtained.

[0142] In this way, by comprehensively considering the correlation between the first sub-behavior data and the second sub-behavior data, the accuracy of the final analysis result can be further improved.

[0143] In step S15, regression processing is performed on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain estimated parameters, wherein the estimated parameters are used to represent the causal relationship between the historical behavior data and the target behavior data.

[0144] In this step, the third sub-behavior data estimate and the fourth sub-behavior data can be used to perform the second stage regression processing on the target behavior processing data, establish a multivariate regression model between the third sub-behavior data estimate, the fourth sub-behavior data and the target behavior processing data, and determine the corresponding model coefficients, that is, the estimated parameters, which include the coefficient estimate of the third sub-behavior data estimate and the coefficient estimate of the fourth sub-behavior data.

[0145] Specifically, the target behavior processing data can be used as the dependent variable, and the estimated value of the third sub-behavior data and the fourth sub-behavior data can be used as independent variables to perform multivariate linear regression. For example, a multivariate linear equation can be used to establish the relationship between the dependent variable and the independent variable, and the coefficients in the multivariate linear equation can be obtained through curve fitting. The coefficients are the required estimated parameters, and the target behavior processing data is approximately equal to the sum of the product of the estimated value of the third sub-behavior data multiplied by the first coefficient and the product of the fourth sub-behavior data multiplied by the second coefficient.

[0146] In this way, the influence coefficient of the user's historical behavior data on the user's target behavior data can be determined through the estimated parameters. For example, the influence coefficient of the user's live broadcast room purchase amount on the subsequent reward amount can be determined.

[0147] Optionally, the step S15 includes:

[0148] A multivariate linear regression model is used to perform regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain model parameters of the multivariate linear regression model, and the model parameters of the multivariate linear regression model are used as the estimated parameters.

[0149] In one embodiment, the third sub-behavior data estimate obtained in the first-stage regression and the fourth sub-behavior data after decentralized processing can be used to perform a second-stage multivariate regression on the target behavior processing data. Specifically, a multivariate linear regression model can be established between the third sub-behavior data estimate, the fourth sub-behavior data, and the target behavior processing data, and then the corresponding model parameters can be determined by the least squares method, etc. The model parameters are the required estimated parameters.

[0150] For example, the multivariate linear regression model of the second stage is Y′=β1X′1+β2X′2+∈, where Y′ represents the target behavior processing data obtained after the target behavior data Y is decentralized, X′1 represents the third sub-behavior data obtained after the first sub-behavior data X1 is decentralized, X′2 represents the fourth sub-behavior data obtained after the second sub-behavior data X2 is decentralized, β1 and β2 are model parameters, and ∈ is the residual term. In regression, the estimated value of the third sub-behavior data is used. Instead of the third sub-behavior data X′1, the third behavior data estimate and the fourth sub-behavior data are used to perform least squares regression on the target behavior processing data to determine the parameter estimates of β1 and β2, where β1 and β2 are the estimated parameters representing the causal relationship between the historical behavior data and the target behavior data.

[0151] Thus, through this implementation, the estimated parameters can be calculated quickly and accurately. The model parameters of the multivariate linear regression model obtained in the embodiment of the present disclosure are the required estimated parameters.

[0152] It should be noted that, in the embodiment of the present disclosure, in order to process the historical behavior data and the target behavior data more quickly and efficiently, a preset fixed effect model can also be constructed to determine the causal relationship between the historical behavior data and the target behavior data. Specifically, the preset fixed effect model can be Y=β1X1+β2X2+Dα+∈, wherein Y is the dependent variable, X1 is the endogenous independent variable, X2 is the exogenous independent variable, β1 and β2 are model parameters, D is the fixed effect variable matrix, α is the fixed effect parameter, and ∈ is the residual term. It can be seen from the model formula that the model parameters β1 and β2 can reflect the causal relationship between the independent variable and the dependent variable, that is, they can reflect how the independent variable X affects the dependent variable Y.

[0153] Therefore, in order to obtain the causal relationship between the user's historical behavior data and the target behavior data, we can Figure 2 As shown, the user's first sub-behavior data (such as the purchase amount in the live broadcast room) is taken as the endogenous independent variable X1, the user's second sub-behavior data (such as the viewing time in the live broadcast room) is taken as the exogenous independent variable X2, the target behavior data (such as the reward amount in the live broadcast room) is taken as the dependent variable Y, and the user's first pre-processed data (such as whether to block the live broadcast of the follow-up page) is taken as the instrumental variable Q for the endogenous independent variable X1. At the same time, considering that the user portrait data is fixed user individual characteristic data, in order to avoid omitting the impact of this variable on the target behavior data, the appropriate user portrait data can be selected as needed to be converted into a fixed effect variable matrix D.

[0154] Then, based on the fixed effect variable matrix, the endogenous independent variables, exogenous independent variables and dependent variables can be decentralized (de-mean) to eliminate the influence of the fixed effect variable matrix D in the preset fixed effect model on the dependent variable Y, that is, to eliminate the Dα part in the preset fixed effect model, so that the preset fixed effect model can be converted into a multivariate linear regression model Y′=β1X′1+β2X′2+∈, where Y′, X′1 and X′2 are the decentralized dependent variable, endogenous independent variable and exogenous independent variable respectively. In addition, the instrumental variable Q can also be decentralized in the same way to obtain Q′, and the model X′1=γ1Q′+γ2X′2 is used to perform the first-stage regression processing on X′1, X′2 and Q′ to determine the model coefficients γ1 and γ2, and then use the model to output the estimated value of X′1 Then, you can use Substitute X′1, and combine X′2 and Y′, input the multivariate linear regression model Y′=β1X′1+β2X′2+∈, perform the second stage regression processing, and determine the estimated values ​​of the model parameters β1 and β2. Finally, the causal relationship between the user's historical behavior data and the target behavior data can be determined based on the estimated values ​​of the model parameters β1 and β2, and by using instrumental variables to replace the endogenous independent variables for regression, the problem of biased parameter estimates caused by the endogeneity of the dependent variable in the related technology can be solved, and the problem of omitted variables and selection bias in the related technology can be solved by fixing the individual user effects through the preset fixed effect model.

[0155] In step S16, the current behavior data and current user portrait data of the user at the current moment are obtained, and based on the estimated parameters, the user's future behavior data is estimated.

[0156] After obtaining the estimated parameters, the estimated parameters can be used to estimate the user's future behavior data. Specifically, the current behavior data and current user portrait data of the user at the current moment can be obtained first. The current moment can be the current day, the past week, the past month, etc. For example, when it is necessary to estimate the amount of rewards that the user will give on a live broadcast platform in the next day, week or month, the purchase amount, other relevant behavior data and user portrait data of the user in the live broadcast room on the same day, the past week or the past month can be obtained, and then the purchase amount and other relevant behavior data can be multiplied by the corresponding estimated parameters respectively to get the estimated reward amount.

[0157] It should be noted that after performing regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain the estimated parameters, other parameters in the multivariate regression model, such as residual terms, fixed effect parameters, etc., can be further determined. Therefore, when estimating the user's future behavior data based on the user's current behavior data and current user portrait data at the current moment, the estimated parameters and other parameters can be combined to obtain more accurate and reliable estimated data.

[0158] Optionally, after step S15 and before step S16, the method further includes:

[0159] Based on the estimated parameters, a preset fixed effect model is constructed, wherein the endogenous independent variable of the preset fixed effect model is the first sub-behavior data of the user, the exogenous independent variable of the preset fixed effect model is the second sub-behavior data of the user, the dependent variable of the preset fixed effect model is the target behavior data of the user, and the model parameters of the preset fixed effect model are the estimated parameters;

[0160] The step S16 comprises:

[0161] The current behavior data and current user portrait data of the user at the current moment are obtained, and the user's future behavior data is estimated using the preset fixed effect model.

[0162] In this implementation, in order to more accurately analyze the impact of user historical behavior data on target behavior data, and to more reliably predict user future target behavior data based on current user historical behavior data and user portrait data, this can be achieved by constructing a suitable fixed effect model.

[0163] Specifically, the user's historical behavior data can be used as the independent variable of the preset fixed effect model to be constructed, wherein the first sub-behavior data is an endogenous independent variable, and the second sub-behavior data is an exogenous independent variable. The user's target behavior data is used as the dependent variable of the preset fixed effect model, and the estimated parameters are the model parameters of the preset fixed effect model, that is, the coefficients of the independent variables of the preset fixed effect model. Specifically, the preset fixed effect model can be Y=β1X1+β2X2+Dα+∈, wherein Y is the dependent variable, X1 is the endogenous independent variable, X2 is the exogenous independent variable, β1 and β2 are model parameters, α is the fixed effect parameter, ∈ is the residual term, and D is the fixed effect variable matrix. In the embodiment of the present disclosure, the user portrait data can be converted into a fixed effect variable matrix.

[0164] In this way, the user's future behavior data can be estimated by constructing a preset fixed effect model. Specifically, when the model parameters β1 and β2 of the preset fixed effect model are determined, the current behavior data of the user at the current moment obtained can be decentralized according to the current user portrait data obtained, and then the processed current behavior data is input into the preset fixed effect model as an independent variable. By multiplying the independent variable and the model parameters through the preset fixed effect model, the estimated user's future behavior data output by the preset fixed effect model can be obtained.

[0165] Of course, a more complete preset fixed effect model can be constructed by further determining the residual term and fixed effect parameters in the preset fixed effect model, and the user portrait data can be converted into a fixed effect variable matrix, so that the current behavior data of the user at the current moment and the converted fixed effect variable matrix can be input into the preset fixed effect model, and the user's future behavior data estimated by the preset fixed effect model can be quickly obtained. The determination of the residual term and fixed effect parameters in the preset fixed effect model is described in detail below.

[0166] Optionally, constructing a preset fixed effect model based on the estimated parameters includes:

[0167] Determine a target behavior processing data estimation value based on the third sub-behavior data, the fourth sub-behavior data and the estimated parameter;

[0168] Determining a residual term in the preset fixed effect model based on a difference between the target behavior processed data and an estimated value of the target behavior processed data;

[0169] Based on the estimated parameters and the residual terms, the preset fixed effect model is constructed.

[0170] That is to say, in order to obtain more reliable analysis results and more accurate estimated data, after determining the estimated parameters, the residual term of the preset fixed effect model can be further determined.

[0171] Specifically, after the data set is decentralized, the preset fixed effect model can be converted to Y′=β1X′1+β2X′2+∈, where Y′ represents the target behavior processing data, X′1 represents the third sub-behavior data, and X′2 represents the fourth sub-behavior data. In this way, when β1 and β2 are determined, the estimated value of the target behavior processing data can be obtained by substituting the third sub-behavior data and the fourth sub-behavior data into the model for calculation. Then, the residual term ∈ can be determined by subtracting the estimated value of the target behavior processing data from the true value of the target behavior processing data, i.e., Y′.

[0172] In this way, by determining the residual term in the preset fixed effect model, it can help to better analyze the causal relationship between the user's historical behavior data and the target behavior data, and obtain a more complete preset fixed effect model, and then based on the preset fixed effect model, more accurately predict the user's future behavior data.

[0173] Furthermore, after determining the residual term in the preset fixed effect model, the method further includes:

[0174] Obtain the correlation between each type of historical user portrait data, and cluster each column of historical user portrait data based on the correlation;

[0175] Determine the covariance matrix of the model parameters according to the residual items corresponding to the historical user portrait data after clustering;

[0176] Based on the diagonal elements in the covariance matrix, estimated standard errors of the model parameters are determined.

[0177] That is, after determining the residual term in the preset fixed effect model, the estimated standard error of the model parameters in the preset fixed effect model can be further determined to measure the significance of the model parameters, that is, to determine the credibility of the model parameters.

[0178] In actual applications, some user portrait data may be correlated, for example, the purchase amounts of users in the same age group are correlated, or the live broadcast viewing time of a certain user today is correlated with the live broadcast viewing time of yesterday, etc. Therefore, based on the correlation between user portrait data in actual situations, the correlation between each type of historical user portrait data can be determined. Specifically, the correlation between each type of historical user portrait data can be pre-set, so that the pre-set correlation between each type of historical user portrait data can be obtained during use, and the historical user portrait data with correlation are clustered based on the correlation, and the estimated standard error of the model parameters in the clustering case is calculated.

[0179] Specifically, for r types of historical user portrait data, each column of historical user portrait data can be clustered based on the correlation between each type of historical user portrait data. For example, for a column of historical user portrait data representing age, the column of historical user portrait data can be clustered according to age groups. The ages of n users can be divided into 5 categories according to 5 different age groups. In this way, the residual terms corresponding to users in the same age group obey the same distribution.

[0180] Next, the covariance matrix of the model parameters can be calculated based on the residual items corresponding to the clustered historical user portrait data. For example, assuming that the age group can be divided into 5 categories, the covariance matrix can be calculated based on the formula V(β)=(X T X) -1 B(X T X) -1 , to calculate the covariance matrix V(β) of the model parameters, where X is the independent variable, X T is the transformation rank of X, (X T X) -1 For X T The inverse matrix of X, in the case of clustering, ∈ j =(∈ 11 …∈ t1 ) T ,∈ 11 …∈ t1 The residual items corresponding to users belonging to the same age group, such as among n users, user 1, user 2 and user 3 belong to the same age group, then the residual items corresponding to n users, the residual item ∈1 corresponding to user 1, the residual item ∈2 corresponding to user 2 and the residual item ∈3 corresponding to user 3 can be grouped together to calculate the correlation between user 1, user 2 and user 3. It should be understood that these three residual items are not correlated with the residual items corresponding to users of other age groups, that is, the correlation with users of other age groups is 0. The calculation process can be as follows Figure 2 shown.

[0181] It should be noted that in the case of no clustering, that is, each type of historical user portrait data is not clustered, it can be divided into different cases of independent homoscedasticity and heteroscedasticity, and different B is used to calculate the covariance matrix of the model parameters. For example, for the case of independent homoscedasticity, that is, each type of historical user portrait data is independent and uncorrelated, then For the case of heteroscedasticity, that is, there is correlation between each type of user portrait data, then

[0182] After determining the covariance matrix of the model parameters, the variance of the model parameters can be determined based on the diagonal elements in the covariance matrix, and the estimated standard error of the model parameters can be obtained after taking the square root of the variance.

[0183] In this way, by calculating the estimated standard error of the model parameters in the preset fixed effect model, it can be used to evaluate the credibility of the calculated model parameters. For example, when the estimated standard error of the model parameter is small, it can be known that the model parameter has a greater credibility and is closer to the true value. When the estimated standard error of the model parameter is large, it can be known that the credibility of the model parameter is not high and the gap with the true value is large. Based on this, the sample data (i.e., historical behavior data and target behavior data) can be further analyzed to investigate whether there are problems such as multicollinearity, and the reasons for the large estimated standard error can be analyzed one by one. After determining the reason, the data can be corrected and the model parameters and estimated standard error can be recalculated.

[0184] Optionally, the fixed effect variable of the preset fixed effect model is user portrait data;

[0185] After determining the residual term in the preset fixed effect model and before constructing the preset fixed effect model based on the estimated parameters and the residual term, the method further includes:

[0186] Determining fixed effect parameters in the preset fixed effect model based on the historical behavior data, the target behavior data, the historical user portrait data, the model parameters and the residual term;

[0187] The step of constructing a preset fixed effect model based on the estimated parameters and the residual term includes:

[0188] The preset fixed effect model is constructed based on the estimated parameters, the residual terms and the fixed effect parameters.

[0189] In certain embodiments, if it is necessary to determine the size of the fixed effect of the user's individual characteristic data, that is, the user portrait data, on the user's target behavior data, the fixed effect parameters in the preset fixed effect model can be further solved based on the known historical behavior data (including the first sub-behavior data and the second sub-behavior data), the target behavior data and the historical user portrait data, and the model parameters and residual terms of the preset fixed effect model. That is, according to the preset fixed effect model formula Y=β1X1+β2X2+Dα+∈, when the parameters Y, X1, X2, β1, β2, D and ∈ are all known, the fixed effect parameter α can be determined, which can reflect the size of the impact of various types of user portrait data on the user's target behavior data. For example, the fixed effect parameter corresponding to the user's age is greater than the fixed effect parameter corresponding to the user's region, indicating that the impact of age on the user's target behavior data is greater than the impact of region on the user's target behavior data.

[0190] Among them, in order to avoid the need to invert D in the calculation, which results in a large amount of calculation, the mapping method can be used to solve the fixed effect parameter α, that is, after inputting the parameters Y, X1, X2, β1, β2, D and ∈ of the preset fixed effect model, the estimated value of α is continuously mapped so that the estimated value of α is continuously close to the true value, and the result of α is output.

[0191] In this way, by determining the fixed effect parameters in the preset fixed effect model, the fixed effect influence of the user portrait data on the target behavior data can be determined. After the fixed effect parameters are determined, each parameter in the preset fixed effect model is determined, and then based on the preset fixed effect model, the user target behavior data can be predicted more accurately. For example, the current behavior data of the user at the current moment obtained can be used as an independent variable, and the current user portrait data obtained can be used as a fixed effect variable, and input into the preset fixed effect model (each parameter of the model has been determined), and a more accurate estimate of the user target behavior data can be output through the preset fixed effect model.

[0192] Furthermore, the residual term is an n-dimensional column vector; after determining the fixed effect parameters in the preset fixed effect model, the method further includes:

[0193] Sampling n data in the residual term n times to obtain a target residual term consisting of n sampled data, and repeating the process of sampling n data in the residual term n times until m target residual terms are obtained, where n is a positive integer and m is a positive integer. In an example, m is an integer greater than 1;

[0194] Determining m fixed effect parameter estimates based on the historical behavior data, the target behavior data, the historical user profile data, the model parameters, and the m target residual terms;

[0195] Based on the m fixed effect parameter estimates, the estimated standard errors of the fixed effect parameters are determined.

[0196] That is, after determining the fixed effect parameters in the preset fixed effect model, the estimated standard error of the preset fixed effect parameters can be further calculated to measure the degree of deviation between the calculated fixed effect parameters and their true values, that is, the credibility of the fixed effect parameters.

[0197] Specifically, for n users, n residual items can be calculated, that is, the residual item can be an n-dimensional column vector composed of n data. First, the n data in the residual item can be sampled n times, specifically, n times of sampling with replacement can be performed to obtain a target residual item composed of n sampled data, that is, each time a data is extracted from the n data, the extracted data is used as one of the target residual items, and the extracted data is put back into the n data for the next extraction, and the sampling is repeated n times to obtain n sampled data, and there may be multiple identical data in the n sampled data. And the process of sampling n times for n data in the residual item can be repeated m times until m target residual items are obtained.

[0198] Then, based on the historical behavior data (including the first sub-behavior data and the second sub-behavior data), the target behavior data, the historical user profile data, the model parameters and the m target residual terms, based on the preset fixed effect model formula Y=β1X1+β2X2+Dα+∈, the estimated values ​​α1, α2, ..., α of the m fixed effect parameters can be calculated accordingly. m .

[0199] Finally, the estimated values ​​α1, α2, ..., α of the m fixed effect parameters can be used. m , to calculate the estimated standard error of the fixed effect parameter α, specifically, the formula can be used To calculate the estimated standard error σ of α α ,in, are α1, α2, ... and α m The mean of .

[0200] In this way, by calculating the estimated standard error of the fixed effect parameter in the preset fixed effect model, the credibility of the calculated fixed effect parameter can be determined.

[0201] The data estimation method in the embodiment of the present disclosure obtains the historical behavior data and historical user portrait data of the user in a first historical time period, and determines the first pre-processed data related to the first sub-behavior data, wherein the historical behavior data includes the first sub-behavior data and the second sub-behavior data; obtains the target behavior data of the user in a second historical time period, wherein the second historical time period is the time period after the first historical time period; based on the historical user portrait data, the first sub-behavior data, the second sub-behavior data, the target behavior data and the first pre-processed data are decentralized to obtain the corresponding third sub-behavior data, fourth sub-behavior data, target behavior processing data and second pre-processed data; the second pre-processed data and the third sub-behavior data are regressed to obtain the estimated value of the third sub-behavior data; the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data are regressed to obtain the estimated parameters, wherein the estimated parameters are used to represent the causal relationship between the historical behavior data and the target behavior data; the current behavior data and the current user portrait data of the user at the current moment are obtained, and based on the estimated parameters, the user's future behavior data is estimated. In this way, by using the first preprocessed data related to the first sub-behavior data as an instrumental variable to replace the first sub-behavior data to regress the target behavior data, the problem of variable endogeneity can be solved, thereby ensuring that the estimated parameters obtained by regression are consistent, and then ensuring that based on the estimated parameters and the user's current behavior data and portrait data, the user's future behavior data can be more accurately predicted.

[0202] Figure 3 is a block diagram of a data estimation device according to an exemplary embodiment. Figure 3 The data estimation device 300 includes a first acquisition module 301, a second acquisition module 302, a first processing module 303, a second processing module 304, a third processing module 305 and an estimation module 306.

[0203] The first acquisition module 301 is configured to execute acquisition of historical behavior data and historical user portrait data of a user in a first historical time period, and determine first preprocessing data related to first sub-behavior data, wherein the historical behavior data includes the first sub-behavior data and the second sub-behavior data;

[0204] The second acquisition module 302 is configured to execute acquisition of target behavior data of the user in a second historical time period, wherein the second historical time period is a time period after the first historical time period;

[0205] The first processing module 303 is configured to perform decentralized processing on the first sub-behavior data, the second sub-behavior data, the target behavior data and the first pre-processed data based on the historical user portrait data to obtain corresponding third sub-behavior data, fourth sub-behavior data, target behavior processing data and second pre-processed data;

[0206] The second processing module 304 is configured to perform regression processing on the second pre-processed data and the third sub-behavior data to obtain an estimated value of the third sub-behavior data;

[0207] The third processing module 305 is configured to perform regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain an estimated parameter, wherein the estimated parameter is used to represent the causal relationship between the historical behavior data and the target behavior data;

[0208] The estimation module 306 is configured to obtain the current behavior data and the current user portrait data of the user at the current moment, and estimate the user's future behavior data based on the estimation parameters.

[0209] Optionally, the first acquisition module 301 includes:

[0210] A first processing unit is configured to perform preprocessing on a service function of the service to be analyzed within the first historical time period, and obtain first preprocessing data of the user;

[0211] The acquisition unit is configured to execute acquisition of the preprocessed historical behavior data of the user in the first historical time period, and to acquire historical user portrait data, wherein the first preprocessed data is an instrumental variable of the first sub-behavior data.

[0212] Optionally, the second processing module 304 includes:

[0213] A second processing unit is configured to perform regression processing on the second preprocessed data, the fourth sub-behavior data and the third sub-behavior data to obtain coefficient estimates of the second preprocessed data and the fourth sub-behavior data;

[0214] The first determining unit is configured to determine a third sub-behavior data estimation value based on the coefficient estimation, the second pre-processed data and the fourth sub-behavior data.

[0215] Optionally, the third processing module 305 is configured to perform regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data using a multivariate linear regression model to obtain model parameters of the multivariate linear regression model, and use the model parameters of the multivariate linear regression model as the estimated parameters.

[0216] Optionally, the target behavior data is an n-dimensional column vector, the historical behavior data is an n×k matrix, and the historical user portrait data is an n×r matrix, where n is the number of users, k is the number of historical behavior data, and r is the number of categories of historical user portrait data, and n, k, and r are all positive integers;

[0217] The first processing module 303 is configured to perform grouping and averaging of each column of the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data according to a column of data in the historical user portrait data, and to subtract the mean of the group to which each data in each column of data belongs from each data until each column of the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data no longer changes, wherein each column of data corresponding to the same historical user portrait data is a group.

[0218] Optionally, the data estimation device 300 further includes:

[0219] A construction module is configured to execute construction of a preset fixed effect model based on the estimated parameters, wherein the endogenous independent variable of the preset fixed effect model is the first sub-behavior data of the user, the exogenous independent variable of the preset fixed effect model is the second sub-behavior data of the user, the dependent variable of the preset fixed effect model is the target behavior data of the user, and the model parameters of the preset fixed effect model are the estimated parameters;

[0220] The estimation module 306 is configured to obtain the current behavior data and current user portrait data of the user at the current moment, and use the preset fixed effect model to estimate the user's future behavior data.

[0221] Optionally, the building blocks include:

[0222] A second determining unit is configured to determine a target behavior processing data estimation value based on the third sub-behavior data, the fourth sub-behavior data and the estimated parameter;

[0223] A third determining unit is configured to determine a residual term in the preset fixed effect model based on a difference between the target behavior processed data and the target behavior processed data estimate;

[0224] A construction unit is configured to construct the preset fixed effect model based on the estimated parameters and the residual term.

[0225] Optionally, the data estimation device 300 further includes:

[0226] A clustering module is configured to obtain the correlation between each type of historical user portrait data, and cluster each column of historical user portrait data based on the correlation;

[0227] A first determination module is configured to determine the covariance matrix of the model parameters according to the residual term corresponding to the clustered historical user portrait data;

[0228] The second determination module is configured to determine the estimated standard errors of the model parameters based on the diagonal elements in the covariance matrix.

[0229] Optionally, the fixed effect variable of the preset fixed effect model is user portrait data;

[0230] The data estimation device 300 further includes:

[0231] A third determination module is configured to determine the fixed effect parameters in the preset fixed effect model based on the historical behavior data, the target behavior data, the historical user portrait data, the model parameters and the residual term;

[0232] The construction module is configured to construct the preset fixed effect model based on the estimated parameters, the residual terms and the fixed effect parameters.

[0233] Optionally, the residual term is an n-dimensional column vector;

[0234] The data estimation device 300 further includes:

[0235] a sampling module configured to perform n sampling on the n data in the residual term n times to obtain a target residual term consisting of the n sampled data, and repeat the process of performing n sampling on the n data in the residual term n times until m target residual terms are obtained, where n is a positive integer and m is a positive integer;

[0236] A fourth determination module is configured to determine m fixed effect parameter estimates based on the historical behavior data, the target behavior data, the historical user portrait data, the model parameters and the m target residual terms;

[0237] The fifth determination module is configured to determine the estimated standard error of the fixed effect parameter based on the m fixed effect parameter estimates.

[0238] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0239] The data estimation device in the embodiment of the present disclosure obtains the historical behavior data and historical user portrait data of the user in a first historical time period, and determines the first pre-processed data related to the first sub-behavior data, wherein the historical behavior data includes the first sub-behavior data and the second sub-behavior data; obtains the target behavior data of the user in a second historical time period, wherein the second historical time period is the time period after the first historical time period; based on the historical user portrait data, the first sub-behavior data, the second sub-behavior data, the target behavior data and the first pre-processed data are decentralized to obtain the corresponding third sub-behavior data, fourth sub-behavior data, target behavior processing data and second pre-processed data; the second pre-processed data and the third sub-behavior data are regressed to obtain the estimated value of the third sub-behavior data; the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data are regressed to obtain the estimated parameter, wherein the estimated parameter is used to represent the causal relationship between the historical behavior data and the target behavior data; the current behavior data and the current user portrait data of the user at the current moment are obtained, and based on the estimated parameter, the user's future behavior data is estimated. In this way, by using the first preprocessed data related to the first sub-behavior data as an instrumental variable to replace the first sub-behavior data to regress the target behavior data, the problem of variable endogeneity can be solved, thereby ensuring that the estimated parameters obtained by regression are consistent, and then ensuring that based on the estimated parameters and the user's current behavior data and portrait data, the user's future behavior data can be more accurately predicted.

[0240] Figure 4 is a block diagram of a data estimation device 400 according to an exemplary embodiment.

[0241] Reference Figure 4 The data estimation device 400 includes: a processor 401, a memory 402, and a bus interface 403.

[0242] The processor 401 is used to read the program in the memory 402 and execute the following process:

[0243] Acquire historical behavior data and historical user portrait data of a user in a first historical time period, and determine first preprocessing data related to first sub-behavior data, wherein the historical behavior data includes the first sub-behavior data and the second sub-behavior data;

[0244] Acquire target behavior data of the user in a second historical time period, wherein the second historical time period is a time period after the first historical time period;

[0245] Based on the historical user portrait data, the first sub-behavior data, the second sub-behavior data, the target behavior data and the first pre-processed data are decentralized to obtain corresponding third sub-behavior data, fourth sub-behavior data, target behavior processed data and second pre-processed data;

[0246] Performing regression processing on the second preprocessed data and the third sub-behavior data to obtain an estimated value of the third sub-behavior data;

[0247] Performing regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain an estimated parameter, wherein the estimated parameter is used to represent the causal relationship between the historical behavior data and the target behavior data;

[0248] The current behavior data and current user portrait data of the user at the current moment are obtained, and based on the estimated parameters, the user's future behavior data is estimated.

[0249] exist Figure 4 In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically linking various circuits of one or more processors represented by processor 401 and memory represented by memory 402. The bus architecture may also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. Bus interface 403 provides an interface.

[0250] The processor 401 is responsible for managing the bus architecture and general processing, and the memory 402 can store data used by the processor 401 when performing operations.

[0251] Optionally, the processor 401 is further configured to:

[0252] Preprocessing the service function of the service to be analyzed within the first historical time period to obtain first preprocessed data of the user;

[0253] Obtain the preprocessed historical behavior data of the user in the first historical time period, and obtain historical user portrait data, wherein the first preprocessed data is an instrumental variable of the first sub-behavior data.

[0254] Optionally, the processor 401 is further configured to:

[0255] Performing regression processing on the second preprocessed data, the fourth sub-behavior data, and the third sub-behavior data to obtain coefficient estimates of the second preprocessed data and the fourth sub-behavior data;

[0256] Based on the coefficient estimate, the second pre-processed data and the fourth sub-behavior data, a third sub-behavior data estimate is determined.

[0257] Optionally, the processor 401 is further configured to:

[0258] A multivariate linear regression model is used to perform regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain model parameters of the multivariate linear regression model, and the model parameters of the multivariate linear regression model are used as the estimated parameters.

[0259] Optionally, the target behavior data is an n-dimensional column vector, the historical behavior data is an n×k matrix, and the historical user portrait data is an n×r matrix, where n is the number of users, k is the number of historical behavior data, and r is the number of categories of historical user portrait data, and n, k, and r are all positive integers;

[0260] The processor 401 is further configured to:

[0261] For each column of the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data, the data are grouped according to a column of the historical user portrait data and the average is calculated each time, and the average of the group to which the data belongs is subtracted from each data in each column of the data until the data in each column of the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data no longer changes, wherein each column of data corresponding to the same historical user portrait data is a group.

[0262] Optionally, the processor 401 is further configured to:

[0263] Based on the estimated parameters, a preset fixed effect model is constructed, wherein the endogenous independent variable of the preset fixed effect model is the first sub-behavior data of the user, the exogenous independent variable of the preset fixed effect model is the second sub-behavior data of the user, the dependent variable of the preset fixed effect model is the target behavior data of the user, and the model parameters of the preset fixed effect model are the estimated parameters;

[0264] The current behavior data and current user portrait data of the user at the current moment are obtained, and the user's future behavior data is estimated using the preset fixed effect model.

[0265] Optionally, the processor 401 is further configured to:

[0266] Determine a target behavior processing data estimation value based on the third sub-behavior data, the fourth sub-behavior data and the estimated parameter;

[0267] Determining a residual term in the preset fixed effect model based on a difference between the target behavior processed data and an estimated value of the target behavior processed data;

[0268] Based on the estimated parameters and the residual terms, the preset fixed effect model is constructed.

[0269] Optionally, the processor 401 is further configured to:

[0270] Obtain the correlation between each type of historical user portrait data, and cluster each column of historical user portrait data based on the correlation;

[0271] Determine the covariance matrix of the model parameters according to the residual items corresponding to the historical user portrait data after clustering;

[0272] Based on the diagonal elements in the covariance matrix, estimated standard errors of the model parameters are determined.

[0273] Optionally, the fixed effect variable of the preset fixed effect model is user portrait data;

[0274] The processor 401 is further configured to:

[0275] Determining fixed effect parameters in the preset fixed effect model based on the historical behavior data, the target behavior data, the historical user portrait data, the model parameters and the residual term;

[0276] The preset fixed effect model is constructed based on the estimated parameters, the residual terms and the fixed effect parameters.

[0277] Optionally, the residual term is an n-dimensional column vector; the processor 401 is further configured to:

[0278] Sampling n data in the residual term n times to obtain a target residual term consisting of n sampled data, and repeating the process of sampling n data in the residual term n times until m target residual terms are obtained, where n is a positive integer and m is a positive integer;

[0279] Determining m fixed effect parameter estimates based on the historical behavior data, the target behavior data, the historical user profile data, the model parameters, and the m target residual terms;

[0280] Based on the m fixed effect parameter estimates, the estimated standard errors of the fixed effect parameters are determined.

[0281] The data estimation device 400 can implement each process in the aforementioned embodiments, which will not be described again here to avoid repetition.

[0282] In an exemplary embodiment, a storage medium including instructions is also provided, such as a memory including instructions, and the instructions can be executed by a processor of the data estimation device to complete the above method. Optionally, the storage medium can be a non-temporary computer-readable storage medium, for example, the non-temporary computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0283] In an exemplary embodiment, a computer program product is also provided, comprising executable instructions, and when the executable instructions are executed on a computer, the computer is enabled to execute the above data estimation method.

[0284] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0285] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A data estimation method, characterized in that: include: Acquire historical behavior data and historical user portrait data of a user in a first historical time period, and determine first preprocessing data related to first sub-behavior data, wherein the historical behavior data includes the first sub-behavior data and second sub-behavior data, and one of the first sub-behavior data and the second sub-behavior data is an endogenous independent variable, and the other is an exogenous independent variable; Acquire target behavior data of the user in a second historical time period, wherein the second historical time period is a time period after the first historical time period; Based on the historical user portrait data, the first sub-behavior data, the second sub-behavior data, the target behavior data and the first pre-processed data are decentralized to obtain corresponding third sub-behavior data, fourth sub-behavior data, target behavior processed data and second pre-processed data; Performing regression processing on the second preprocessed data and the third sub-behavior data to obtain an estimated value of the third sub-behavior data; Performing regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain an estimated parameter, wherein the estimated parameter is used to represent the causal relationship between the historical behavior data and the target behavior data; The current behavior data and current user portrait data of the user at the current moment are obtained, and based on the estimated parameters, the user's future behavior data is estimated.

2. The method according to claim 1, characterized in that The obtaining of historical behavior data and historical user portrait data of the user in the first historical time period, and determining first pre-processed data related to the first sub-behavior data, includes: Preprocessing the service function of the service to be analyzed within the first historical time period to obtain first preprocessed data of the user; Obtain the preprocessed historical behavior data of the user in the first historical time period, and obtain historical user portrait data, wherein the first preprocessed data is an instrumental variable of the first sub-behavior data.

3. The method according to claim 1, characterized in that The performing regression processing on the second preprocessed data and the third sub-behavior data to obtain an estimated value of the third sub-behavior data includes: Performing regression processing on the second preprocessed data, the fourth sub-behavior data, and the third sub-behavior data to obtain coefficient estimates of the second preprocessed data and the fourth sub-behavior data; Based on the coefficient estimate, the second pre-processed data and the fourth sub-behavior data, a third sub-behavior data estimate is determined.

4. The method according to claim 1, characterized in that: The step of performing regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processed data to obtain an estimated parameter includes: A multivariate linear regression model is used to perform regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain model parameters of the multivariate linear regression model, and the model parameters of the multivariate linear regression model are used as the estimated parameters.

5. The method according to claim 1, characterized in that The target behavior data is an n-dimensional column vector, the historical behavior data is an n×k matrix, and the historical user portrait data is an n×r matrix, where n is the number of users, k is the number of historical behavior data, and r is the number of categories of historical user portrait data, and n, k, and r are all positive integers; The decentralizing the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data based on the historical user portrait data includes: For each column of the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data, the data are grouped according to a column of data in the historical user portrait data and the average is calculated, and the average of the group to which the data belongs is subtracted from each data in each column of data until the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data no longer change, wherein each column of data corresponding to the same historical user portrait data is a group.

6. The method according to claim 1, characterized in that After performing regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain the estimated parameters, before obtaining the current behavior data and the current user portrait data of the user at the current moment and estimating the user's future behavior data based on the estimated parameters, the method further includes: Based on the estimated parameters, a preset fixed effect model is constructed, wherein the endogenous independent variable of the preset fixed effect model is the first sub-behavior data of the user, the exogenous independent variable of the preset fixed effect model is the second sub-behavior data of the user, the dependent variable of the preset fixed effect model is the target behavior data of the user, and the model parameters of the preset fixed effect model are the estimated parameters; The obtaining of the current behavior data and the current user portrait data of the user at the current moment, and estimating the user's future behavior data based on the estimation parameters, includes: The current behavior data and current user portrait data of the user at the current moment are obtained, and the user's future behavior data is estimated using the preset fixed effect model.

7. The method according to claim 6, characterized in that The step of constructing a preset fixed effect model based on the estimated parameters includes: Determine a target behavior processing data estimation value based on the third sub-behavior data, the fourth sub-behavior data and the estimated parameter; Determining a residual term in the preset fixed effect model based on a difference between the target behavior processed data and an estimated value of the target behavior processed data; Based on the estimated parameters and the residual terms, the preset fixed effect model is constructed.

8. The method according to claim 7, characterized in that After determining the residual term in the preset fixed effect model, the method further includes: Obtain the correlation between each type of historical user portrait data, and cluster each column of historical user portrait data based on the correlation; Determine the covariance matrix of the model parameters according to the residual items corresponding to the historical user portrait data after clustering; Based on the diagonal elements in the covariance matrix, estimated standard errors of the model parameters are determined.

9. The method according to claim 7, characterized in that: The fixed effect variable of the preset fixed effect model is user portrait data; After determining the residual term in the preset fixed effect model and before constructing the preset fixed effect model based on the estimated parameters and the residual term, the method further includes: Determining fixed effect parameters in the preset fixed effect model based on the historical behavior data, the target behavior data, the historical user portrait data, the model parameters and the residual term; The step of constructing a preset fixed effect model based on the estimated parameters and the residual term includes: The preset fixed effect model is constructed based on the estimated parameters, the residual terms and the fixed effect parameters.

10. The method according to claim 9, characterized in that The residual term is an n-dimensional column vector; after determining the fixed effect parameters in the preset fixed effect model, the method further includes: Sampling n data in the residual term n times to obtain a target residual term consisting of n sampled data, and repeating the process of sampling n data in the residual term n times until m target residual terms are obtained, where n is a positive integer and m is a positive integer; Determining m fixed effect parameter estimates based on the historical behavior data, the target behavior data, the historical user profile data, the model parameters, and the m target residual terms; Based on the m fixed effect parameter estimates, determine the estimated standard errors of the fixed effect parameters.

11. A data estimation device, characterized in that: include: A first acquisition module is configured to execute acquisition of historical behavior data and historical user portrait data of a user in a first historical time period, and determine first preprocessing data related to first sub-behavior data, wherein the historical behavior data includes the first sub-behavior data and the second sub-behavior data, and one of the first sub-behavior data and the second sub-behavior data is an endogenous independent variable, and the other is an exogenous independent variable; A second acquisition module is configured to execute acquisition of target behavior data of users in a second historical time period, wherein the second historical time period is a time period after the first historical time period; A first processing module is configured to perform decentralized processing on the first sub-behavior data, the second sub-behavior data, the target behavior data and the first pre-processed data based on the historical user portrait data to obtain corresponding third sub-behavior data, fourth sub-behavior data, target behavior processing data and second pre-processed data; A second processing module is configured to perform regression processing on the second pre-processed data and the third sub-behavior data to obtain an estimated value of the third sub-behavior data; a third processing module, configured to perform regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data to obtain an estimated parameter, wherein the estimated parameter is used to represent a causal relationship between the historical behavior data and the target behavior data; The estimation module is configured to obtain the current behavior data and current user portrait data of the user at the current moment, and estimate the user's future behavior data based on the estimation parameters.

12. The data estimation device according to claim 11, characterized in that: The first acquisition module includes: A first processing unit is configured to perform preprocessing on a service function of the service to be analyzed within the first historical time period, and obtain first preprocessing data of the user; The acquisition unit is configured to execute acquisition of the preprocessed historical behavior data of the user in the first historical time period, and to acquire historical user portrait data, wherein the first preprocessed data is an instrumental variable of the first sub-behavior data.

13. The data estimation device according to claim 11, characterized in that: The second processing module comprises: A second processing unit is configured to perform regression processing on the second preprocessed data, the fourth sub-behavior data and the third sub-behavior data to obtain coefficient estimates of the second preprocessed data and the fourth sub-behavior data; The first determining unit is configured to determine a third sub-behavior data estimation value based on the coefficient estimation, the second pre-processed data and the fourth sub-behavior data.

14. The data estimation device according to claim 11, characterized in that: The third processing module is configured to perform regression processing on the estimated value of the third sub-behavior data, the fourth sub-behavior data and the target behavior processing data using a multivariate linear regression model to obtain model parameters of the multivariate linear regression model, and use the model parameters of the multivariate linear regression model as the estimated parameters.

15. The data estimation device according to claim 11, characterized in that: The target behavior data is an n-dimensional column vector, the historical behavior data is an n×k matrix, and the historical user portrait data is an n×r matrix, where n is the number of users, k is the number of historical behavior data, and r is the number of categories of historical user portrait data, and n, k, and r are all positive integers; The first processing module is configured to perform grouping and averaging of each column of the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data according to a column of data in the historical user portrait data, and to subtract the mean of the group to which the data belongs from each data in each column of data until each column of the first sub-behavior data, the second sub-behavior data, the target behavior data, and the first pre-processed data no longer changes, wherein each column of data corresponding to the same historical user portrait data is a group.

16. The data estimation device according to claim 11, characterized in that: The data estimation device also includes: A construction module is configured to execute construction of a preset fixed effect model based on the estimated parameters, wherein the endogenous independent variable of the preset fixed effect model is the first sub-behavior data of the user, the exogenous independent variable of the preset fixed effect model is the second sub-behavior data of the user, the dependent variable of the preset fixed effect model is the target behavior data of the user, and the model parameters of the preset fixed effect model are the estimated parameters; The estimation module is configured to obtain the current behavior data and current user portrait data of the user at the current moment, and use the preset fixed effect model to estimate the user's future behavior data.

17. The data estimation device according to claim 16, characterized in that: The building blocks include: A second determining unit is configured to determine a target behavior processing data estimation value based on the third sub-behavior data, the fourth sub-behavior data and the estimated parameter; A third determining unit is configured to determine a residual term in the preset fixed effect model based on a difference between the target behavior processed data and the target behavior processed data estimate; A construction unit is configured to construct the preset fixed effect model based on the estimated parameters and the residual term.

18. The data estimation device according to claim 17, characterized in that: The data estimation device also includes: A clustering module is configured to obtain the correlation between each type of historical user portrait data, and cluster each column of historical user portrait data based on the correlation; A first determination module is configured to determine the covariance matrix of the model parameters according to the residual term corresponding to the clustered historical user portrait data; The second determination module is configured to determine the estimated standard errors of the model parameters based on the diagonal elements in the covariance matrix.

19. The data estimation device according to claim 17, characterized in that: The fixed effect variable of the preset fixed effect model is user portrait data; The data estimation device also includes: A third determination module is configured to determine the fixed effect parameters in the preset fixed effect model based on the historical behavior data, the target behavior data, the historical user portrait data, the model parameters and the residual term; The construction module is configured to construct the preset fixed effect model based on the estimated parameters, the residual terms and the fixed effect parameters.

20. The data estimation device according to claim 19, characterized in that: The residual term is an n-dimensional column vector; The data estimation device also includes: a sampling module configured to perform n sampling on the n data in the residual term n times to obtain a target residual term consisting of the n sampled data, and repeat the process of performing n sampling on the n data in the residual term n times until m target residual terms are obtained, where n is a positive integer and m is a positive integer; A fourth determination module is configured to determine m fixed effect parameter estimates based on the historical behavior data, the target behavior data, the historical user portrait data, the model parameters and the m target residual terms; The fifth determination module is configured to determine the estimated standard error of the fixed effect parameter based on the m fixed effect parameter estimates.

21. A data estimation device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the data estimation method according to any one of claims 1 to 10.

22. A storage medium, when instructions in the storage medium are executed by a processor of a data estimation device, the data estimation device is enabled to perform the data estimation method according to any one of claims 1 to 10.