CN113112311A - Method for training causal inference model, information prompting method and device - Google Patents

Method for training causal inference model, information prompting method and device Download PDF

Info

Publication number
CN113112311A
CN113112311A CN202110519634.8A CN202110519634A CN113112311A CN 113112311 A CN113112311 A CN 113112311A CN 202110519634 A CN202110519634 A CN 202110519634A CN 113112311 A CN113112311 A CN 113112311A
Authority
CN
China
Prior art keywords
value
sample data
target
fluctuation
causal
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110519634.8A
Other languages
Chinese (zh)
Other versions
CN113112311B (en
Inventor
钱丽华
熊健
王浩
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Baidu Netcom Science and Technology Co Ltd
Original Assignee
Beijing Baidu Netcom Science and Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Baidu Netcom Science and Technology Co Ltd filed Critical Beijing Baidu Netcom Science and Technology Co Ltd
Priority to CN202110519634.8A priority Critical patent/CN113112311B/en
Publication of CN113112311A publication Critical patent/CN113112311A/en
Application granted granted Critical
Publication of CN113112311B publication Critical patent/CN113112311B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0242—Determining effectiveness of advertisements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/24323—Tree-organised classifiers
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0277—Online advertisement

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Game Theory and Decision Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Economics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The disclosure provides a method for training a causal inference model and an information prompting method and device, relates to the technical field of computers, and particularly relates to the technical field of artificial intelligence, big data and deep learning. The specific implementation scheme is as follows: training a fluctuation prediction model and an operation prediction model using the first sample dataset; determining a plurality of data fluctuation prediction values based on the second sample data set using the trained fluctuation prediction model; determining a plurality of operation prediction values based on the second sample data set using the trained operation prediction model; and training a causal inference model using the second sample data set, the plurality of data fluctuation predicted values, and the plurality of operational predicted values.

Description

Method for training causal inference model, information prompting method and device
Technical Field
The present disclosure relates to the field of computer technology, and more particularly to the field of artificial intelligence, big data, and deep learning technology.
Background
In the internet advertisement delivery process, data such as advertisement presentation amount, click rate, consumption amount and the like often fluctuate, and these fluctuations are very sensitive to an advertisement delivery party (hereinafter referred to as a user). These data fluctuations are typically due to actions made by the user in the management system. Since a user may perform various types of operations before data fluctuation, which may not be all the causes of data fluctuation, it is difficult for the user to determine which operations are the main causes of data fluctuation.
Disclosure of Invention
The present disclosure provides a method, apparatus, device, and storage medium for training a causal inference model.
According to an aspect of the present disclosure, there is provided a method of training a causal inference model, comprising: training a fluctuation prediction model and an operation prediction model using the first sample dataset; determining a plurality of data fluctuation prediction values based on the second sample data set using the trained fluctuation prediction model; determining, with the trained operation prediction model, a plurality of operation prediction values based on the second sample data set; and training a causal inference model using the second sample data set, the plurality of data fluctuation predicted values, and the plurality of operational predicted values.
According to another aspect of the present disclosure, there is provided an information prompting method, including: acquiring a target data fluctuation value, and a target user characteristic and a target operation value corresponding to the target data fluctuation value, wherein the target operation value corresponds to at least one target operation; inputting the target user characteristics, the target data fluctuation value and the target operation value into a trained first causal inference model to obtain a first causal parameter prediction value; inputting the target user characteristic, the target data fluctuation value and the target operation value into a trained second causal inference model to obtain a second causal parameter prediction value; determining a target cause and effect parameter predicted value according to the target operation value, the first cause and effect parameter predicted value and the second cause and effect parameter predicted value; and generating hint information for the at least one target operation if the target causal parameter prediction value is greater than a causal parameter threshold, wherein the first and second causal inference models are trained based on a method of training a causal inference model of an embodiment of the present disclosure.
Another aspect of the disclosure provides a computing device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of the embodiments of the present disclosure.
According to another aspect of the disclosed embodiments, there is provided a non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method shown in the disclosed embodiments.
According to another aspect of the embodiments of the present disclosure, there is provided a computer program product, a computer program, which when executed by a processor implements the method shown in the embodiments of the present disclosure.
It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.
Drawings
The drawings are included to provide a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:
FIG. 1 schematically illustrates a flow chart of a method of training a causal inference model according to an embodiment of the present disclosure;
FIG. 2 schematically illustrates a flow diagram of a method of training a surge prediction model according to another embodiment of the present disclosure;
FIG. 3 schematically illustrates a flow diagram of a method of training an operational prediction model according to another embodiment of the present disclosure;
FIG. 4 schematically illustrates a flow chart of a method of training a causal inference model according to another embodiment of the present disclosure;
FIG. 5 schematically illustrates a flow chart of an information prompting method according to an embodiment of the present disclosure;
FIG. 6A schematically illustrates a first stage of an information prompting method according to another embodiment of the present disclosure;
FIG. 6B schematically illustrates a diagram of a second stage of an information prompting method according to another embodiment of the present disclosure;
FIG. 7 schematically illustrates a block diagram of an apparatus for training a model according to an embodiment of the present disclosure;
FIG. 8 schematically illustrates a block diagram of an information prompting device according to an embodiment of the present disclosure;
fig. 9 schematically shows a schematic block diagram of an example electronic device according to an embodiment of the present disclosure.
Detailed Description
Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of the embodiments of the disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
The method of training the causal inference model will be described in detail below with reference to fig. 1.
FIG. 1 schematically illustrates a flow chart of a method of training a causal inference model according to an embodiment of the present disclosure.
As shown in fig. 1, the method 100 includes the following operations S110 to S150.
In operation S110, a fluctuation prediction model is trained using the first sample data set.
According to embodiments of the present disclosure, the first sample data set may include a plurality of sample data, each of which may include, for example, a user characteristic, a data fluctuation value, and an operation value.
Wherein the user characteristics are used to represent characteristic data of the user. The user characteristics may include, for example, at least one of a base attribute, an industry attribute, an operating frequency, a projected budget, and an effective material size. The user basic attributes may include features such as a last day exposure amount, a last day consumption amount, a last day click amount, and the like. Industry attributes may include, for example, characteristics such as industry ids at various levels. The user operation frequency may include, for example, characteristics such as an average number of operations in the user's last seven days, a number of days in which the user has operated in the last seven days, and the like. The planned budget may include, for example, characteristics of the budget planned by the user, whether it is a particular delivery mode, and the like. The effective material size can comprise the characteristics of effective unit quantity, effective word quantity, effective creative quantity and the like.
The data fluctuation value may be used to indicate the fluctuation size of the particular data over a particular period of time. Illustratively, in the present embodiment, the specific data may include one or more of a display amount, a click amount, and a consumption amount of the advertisement, and the specific time period may be, for example, a last day, a last week, and the like.
An operation value may be used to represent a set of operations in the form of a numerical value, which may include one or more operations that are of the same operation type. For example, in the present embodiment, the operation type may include, for example, adjusting an account budget, adjusting an online time, adjusting a delivery area, and the like.
According to an embodiment of the present disclosure, the operation value may be calculated according to the following formula:
Figure BDA0003063067370000041
where vbid represents an operation value. Illustratively, in this embodiment, if vbid > 0, it represents a positive operation, otherwise, it represents a negative operation. bid represents the amount of change in the parameter after adjustment for each operation in the set of operations,
Figure BDA0003063067370000042
the operating moment at which the parameter is adjusted to bid for the operation,
Figure BDA0003063067370000043
for the moment when the parameter is adjusted from bid to another value,
Figure BDA0003063067370000044
indicating the duration that the parameter remains at bid after it is adjusted to bid by this operation.
According to the embodiment of the disclosure, the fluctuation prediction model can be used for predicting to obtain a corresponding fluctuation predicted value according to the user characteristics, wherein the fluctuation predicted value can be used for representing the data fluctuation amount possibly caused by the user with the user characteristics. For example, in this embodiment, the data fluctuation value in each first sample data may be used as a label, and the user feature in each first sample data may be used as input data to train the fluctuation prediction model.
Then, in operation S120, a plurality of data fluctuation prediction values are determined based on the second sample data set using the trained fluctuation prediction model.
According to an embodiment of the present disclosure, user characteristics in each first sample in the first sample dataset may be input into the trained fluctuation prediction model to derive a plurality of data fluctuation prediction values.
In operation S130, an operational prediction model is trained using the first sample data set.
According to the embodiment of the disclosure, the operation prediction model may be used to predict a corresponding operation prediction value according to the user characteristic, where the operation prediction value may be used to represent an operation that may be performed by a user having the user characteristic. For example, in this embodiment, the operation prediction model may be trained by using the operation value in each first sample data as a label and using the user feature in each first sample data as input data.
In operation S140, a plurality of operation prediction values are determined based on the second sample data set using the trained operation prediction model.
According to an embodiment of the present disclosure, user features in each first sample in the first sample dataset may be input into the trained operation prediction model to derive a plurality of operation prediction values.
It should be noted that the present disclosure does not specifically limit the execution sequence between operations S110 to S120 and S130 to S140, and the operations S110 to S120 and S130 to S140 may be executed in any sequence. For example, operations S110 to S120 may be performed first, and then operations S130 to S140 may be performed, operations S130 to S140 may be performed first, and then operations S110 to S120 may be performed, or operations S110 to S120 and S130 to S140 may be performed simultaneously.
In operation S150, a causal inference model is trained using the second set of sample data, the plurality of data fluctuation predictors, and the plurality of operation predictors.
According to an embodiment of the present disclosure, the second sample data set may include a plurality of sample data, each of which may include, for example, a user characteristic, a data fluctuation value, and an operation value. It should be noted that the sample data in the second sample data set may be different from the sample data in the first sample data set. The causal inference model can be used for predicting to obtain a corresponding causal parameter prediction value according to the user characteristics. The causal parameter predictors may be used, among other things, to determine a degree of correlation between a particular operation and data fluctuations.
Illustratively, in the present embodiment, the fluctuation prediction model and the operation prediction model may include, for example, a gradient lifting tree (GBDT) model, respectively. The causal inference model may include, for example, a random forest model. The GBDT model is used for modeling the operation value, so that the GBDT model can fit a continuous T value, and therefore the method can be applied to a discrete T value scene and has good support for the continuous T value.
In accordance with embodiments of the present disclosure, a trained causal inference model may be used to diagnose fluctuations in data to determine operations that affect the fluctuations in data.
In addition, the related art adopts a rule base established manually, and determines the operation influencing the data fluctuation based on the rule base, and the process is complicated. Compared with a method for manually establishing a rule base, the method for training the causal inference model according to the embodiment of the disclosure is simple and does not need rich prior knowledge.
The method of training the fluctuation prediction model will be described in detail below with reference to fig. 2.
FIG. 2 schematically illustrates a flow diagram of a method of training a surge prediction model according to another embodiment of the present disclosure.
As shown in FIG. 2, the method 210 of training the volatility prediction model may include the following operations S211-S215.
In operation S211, the user characteristics in each of the first sample data are input into the fluctuation prediction model to obtain a fluctuation prediction value corresponding to each of the first sample data.
In operation S212, a first objective function is calculated.
According to an embodiment of the present disclosure, the first objective function may be calculated according to the following formula:
Figure BDA0003063067370000061
wherein, Obj1For the first objective function, Yi is a data fluctuation value in the ith first sample data in the first sample data set,
Figure BDA0003063067370000062
is the fluctuation predicted value corresponding to the ith first sample data, and N is the total number of the first sample data in the first sample data set.
In operation S213, it is determined whether the first objective function converges. In the case where the first objective function does not converge, operation S214 is performed, and in the case where the first objective function converges, operation S215 is performed.
According to an embodiment of the present disclosure, it may be determined that the first objective function converges in a case where a value of the first objective function reaches a minimum. Illustratively, in this embodiment, if the value of the first objective function is k in the nearest place1If the amount of change in the training of the round is less than the first threshold, it may be determined that the first objective function converges. Wherein k is1Is a positive integer. k is a radical of1And the value of the first threshold can be set according to actual needs respectively, and the k is set according to the actual needs1And the value of the first threshold is not particularly limited.
In operation S214, parameters of the fluctuation prediction model are adjusted according to the value of the first objective function, and return is made to operation S211
In operation S215, the training for the fluctuation prediction model is ended.
The method of training the operation prediction model will be described in detail below with reference to fig. 3.
FIG. 3 schematically illustrates a flow diagram of a method of training an operational prediction model according to another embodiment of the present disclosure.
As shown in FIG. 3, the method 320 of training the operational prediction model may include the following operations S321-S325.
In operation S321, the user characteristic in each first sample data is input into the operation prediction model to obtain an operation prediction value corresponding to each first sample data.
In operation S322, a second objective function is calculated.
According to an embodiment of the present disclosure, the second objective function may be calculated according to the following formula:
Figure BDA0003063067370000063
wherein, Obj2Ti is an operation value in the ith first sample data in the first sample data set,
Figure BDA0003063067370000071
is the operation prediction value corresponding to the ith first sample data, and N is the total number of the first sample data in the first sample data set.
In operation S323, it is determined whether the second objective function converges. In the case where the second objective function does not converge, operation S324 is performed, and in the case where the second objective function converges, operation S325 is performed.
According to the embodiment of the present disclosure, it may be determined that the second objective function converges in a case where the value of the second objective function reaches a minimum. Illustratively, in this embodiment, if the value of the second objective function is k in the nearest place2If the amount of change in the training of the round is less than the second threshold, then it may be determined that the second objective function converges. Wherein k is2Is a positive integer. k is a radical of2And the value of the second threshold can be set according to actual needs respectively, and the k is set according to the actual needs2And the value of the second threshold is not particularly limited.
In operation S324, parameters of the operation prediction model are adjusted according to the value of the second objective function, and return is made to operation S321.
In operation S325, the training for the operation prediction model is ended.
The method of training the causal inference model will be described in detail below with reference to FIG. 4.
FIG. 4 schematically illustrates a flow chart of a method of training a causal inference model according to another embodiment of the present disclosure.
As shown in FIG. 4, the method 450 of training a causal inference model may include the following operations S451-S458.
The following operations S451 to S453 may be performed for each second sample data in the second sample data set.
In operation S451, a fluctuation value residual is determined according to a data fluctuation value in the second sample data and a data fluctuation prediction value obtained based on the second sample data.
According to embodiments of the present disclosure, the fluctuation value residual may be calculated according to the following formula:
Figure BDA0003063067370000072
wherein, YiIs the true value of the click volume fluctuation value,
Figure BDA0003063067370000073
the predicted value of the click volume fluctuation value obtained by the fluctuation prediction model is obtained.
In operation S452, an operation value residual is determined according to the operation value in the second sample data and the operation prediction value obtained based on the second sample data.
According to an embodiment of the present disclosure, the operation value residual may be calculated according to the following formula:
Figure BDA0003063067370000074
wherein, TiIs the true value of the operational value or values,
Figure BDA0003063067370000081
is a predicted value of the operating value obtained by the GBDT model 620,
Figure BDA0003063067370000082
is the residual error of the click volume fluctuation value,
Figure BDA0003063067370000083
is the residual of the operation value.
It should be noted that the present disclosure does not specifically limit the execution sequence between operations S451 and S452, and the operations S451 and S452 may be executed in any sequence. For example, operation S451 may be performed first, and then operation S452 may be performed, operation S452 may be performed first, and then operation S451 may be performed, and operations S451 and S452 may be performed at the same time.
In operation S453, causal parameter samples are determined from the fluctuation value residuals and the operation value residuals.
According to an embodiment of the present disclosure, a causal parameter is a parameter with respect to a fluctuating value residual and an operational value residual. Causal parameter samples are causal parameters used as samples for training a causal inference model. For example, in the present implementation, the ratio of the fluctuation value residual to the operation value residual can be used as a causal parameter.
I.e. causal parameter samples equal to
Figure BDA0003063067370000084
Then, the causal inference model is trained by using the causal parameter samples as labels and the user features in the second sample data as input data. Namely, the following operations S454 to S458.
In operation S454, the user characteristics in each second sample data are input into the causal inference model to obtain a causal parameter prediction value corresponding to each second sample data.
In operation S455, a third objective function is calculated.
According to an embodiment of the present disclosure, the third objective function may be calculated according to the following formula:
Figure BDA0003063067370000085
wherein, Obj3In order to be the third objective function,
Figure BDA0003063067370000086
is the operation value residual error corresponding to the ith second sample data in the second sample data set,
Figure BDA0003063067370000087
is the fluctuation value residual corresponding to the ith second sample data, θ (x)i) M is the total number of second sample data in the second sample data set for the causal parameter prediction value corresponding to the ith second sample data.
In operation S456, it is determined whether the third objective function converges. In the case where the third objective function does not converge, operation S457 is performed. In the case where the third objective function converges, operation S458 is performed.
According to the embodiment of the present disclosure, it may be determined that the third objective function converges in a case where the value of the third objective function reaches a minimum. Illustratively, in this embodiment, if the value of the third objective function is k in the nearest place3If the amount of change in the round of training is less than the third threshold, it may be determined that the third objective function converges. Wherein k is3Is a positive integer. k is a radical of3And the value of the third threshold can be set according to actual needs respectively, and the k is set according to the actual needs3And the value of the third threshold is not particularly limited.
In operation S457, parameters of the causal inference model are adjusted according to the value of the third objective function, and operation S454 is returned.
In operation S458, the training for the operation prediction model is ended.
According to the embodiment of the disclosure, the causal inference model is trained by using the residuals of the operation predicted value and the fluctuation predicted value instead of directly using the operation predicted value and the fluctuation predicted value for modeling, so that the over-fitting phenomenon is at least partially avoided on one hand, and the deviation caused by direct fitting is reduced on the other hand.
The information presentation method will be described in detail below with reference to fig. 5.
FIG. 5 schematically shows a flow chart of an information prompting method according to an embodiment of the disclosure.
As shown in fig. 5, the information prompting method 500 may include the following operations S510 to S550.
In operation S510, a target data fluctuation value, and a target user characteristic and a target operation value corresponding to the target data fluctuation value are acquired.
Wherein the target operation value corresponds to at least one target operation.
According to the embodiment of the disclosure, the target data may be monitored in advance, and the fluctuation amount of the target data is determined as the fluctuation value of the target data when the fluctuation amount of the target data in the preset time period is monitored to be larger than the preset fluctuation amount. The preset time period and the preset fluctuation amount can be set according to actual needs. Illustratively, the preset time period may be, for example, the last day, the last week, and the like, and the preset fluctuation amount may be, for example, 20% of the target data value before fluctuation.
For example, current target data may be acquired as raw data, and target data before a preset time period may be acquired as comparison data. The difference between the original data and the comparison data is then calculated. And determining the difference value as a target data fluctuation value under the condition that the difference value is larger than a preset fluctuation threshold value.
According to an embodiment of the present disclosure, at least one target operation related to a target data fluctuation value may be determined, where the target operation may include, for example, an operation of adjusting a parameter. And then determining a target operation value according to the variation of the parameter after each target operation in the at least one target operation is adjusted and the duration of the adjustment. Where an operation value is used to represent one or more operations in the form of a numerical value. For example, the target operation value may be used to represent the at least one target operation in the form of a numerical value.
In operation S520, the target user characteristic, the target data fluctuation value, and the target operation value are input into the trained first causal inference model to obtain a first causal parameter prediction value.
In operation S530, the target user characteristic, the target data fluctuation value, and the target operation value are input into the trained second causal inference model to obtain a second causal parameter prediction value.
It should be noted that the present disclosure does not specifically limit the execution sequence between operations S520 and S530, and the operations S520 and S530 may be executed in any sequence. For example, operation S520 may be performed first, and then operation S530 may be performed, operation S530 may be performed first, and then operation S520 may be performed, or operations S520 and S530 may be performed simultaneously.
According to an embodiment of the disclosure, the first causal inference model and the second causal inference model are trained based on a method of training the causal inference model as illustrated in an embodiment of the disclosure. The first sample data set used for training the first causal inference model is the same as the second sample data set used for training the second causal inference model, and the second sample data set used for training the first causal inference model is the same as the first sample data set used for training the second causal inference model. The data set is divided into a first sample data set and a second sample data set, the two parts are subjected to cross fitting, and a first causal inference model and a second causal inference model are trained respectively, so that the over-fitting phenomenon is at least partially avoided, and the fitting deviation is reduced.
In operation S540, a target causal parameter prediction value is determined according to the target operation value, the first causal parameter prediction value, and the second causal parameter prediction value.
According to an embodiment of the present disclosure, the target causal parameter prediction value may be calculated according to the following formula:
Figure BDA0003063067370000101
wherein ATT is a target causal parameter predicted value theta1(x) For the first causal parameter prediction value, θ2(x) And T is a target operation value.
In operation S550, in the case that the target causal parameter prediction value is greater than the causal parameter threshold value, prompt information for at least one target operation is generated.
According to the embodiment of the disclosure, if the target cause and effect parameter predicted value is greater than the cause and effect parameter threshold value, it indicates that at least one target operation corresponding to the target cause and effect parameter predicted value is a main cause of data fluctuation, and therefore, corresponding prompt information may be generated to prompt a user for the at least one target operation.
According to the embodiment of the disclosure, the operation value of different operations is predicted through a causal inference model, and the operation influencing data fluctuation can be determined. On the basis of determining the operation influencing data fluctuation, prompting the operation influencing data fluctuation can make the user know the reason of the data fluctuation.
The following further describes the method of the information prompting method with reference to specific embodiments. Those skilled in the art will appreciate that the following example embodiments are only for the understanding of the present disclosure, and the present disclosure is not limited thereto.
According to the embodiment of the disclosure, the method of the information prompting method can be applied to the field of internet advertisements. In this embodiment, the fluctuation entities include, for example, two types of placement packages of advertisements and plans of advertisements, and the fluctuation dimensions may include, for example, three types of exposure, click rate, and consumption. An application scenario for the planned exposure amount diagnosis will be described below as an example. If any dimension of the daily/ring ratio, the weekly/ring ratio and the daily/ring ratio of a certain plan is increased or decreased by more than 20%, and the user operates the corresponding plan or the delivery package in the management system from the comparison day to the current day, it can be determined that the operation performed by the user may cause the fluctuation of the presentation amount of the plan, and the reason for the fluctuation needs to be further analyzed.
The operations associated with the plan need to be extracted first and screened, a process also referred to as an operation recall. In the operation of the original extraction, it may happen that a type of operation is performed several times in different directions and at different values, in which case the type of operation may be aggregated so that the operations are represented by operation values. Taking the operation for unit bid as an example, assuming that the user's operation of making multiple changes to the unit bid, the operations for the unit bid dimension can be aggregated according to the following formula:
Figure BDA0003063067370000111
where vbid represents a virtual operation value after aggregation (hereinafter referred to as an operation value), and if vbid > 0, it is a positive operation, and otherwise, it is a negative operation. bid represents the unit bid for the current time period,
Figure BDA0003063067370000112
representing the duration for which the unit bid at bid value,
Figure BDA0003063067370000113
for the operating moment with the modification unit bid,
Figure BDA0003063067370000114
time to modify the unit bid from bid to other values.
After aggregating the operations to obtain the operation value, the relationship between the operation value symbol and the fluctuation direction can be judged, if the two directions are consistent, for example, vbid > 0, and the fluctuation is also positive fluctuation, the operation value is retained, and if the two directions are inconsistent, the operation value is discarded.
In addition, for each operation type, in addition to positive and negative calculation rules defining the operation, custom check rules may also be used. Taking the operation of budget down regulation as an example, a wire collision threshold value can be preset, under the condition that the operation meets the positive and negative calculation rules, whether the value of the budget exceeds the wire collision threshold value after the operation is performed with the budget down regulation is judged, if the value of the budget exceeds the wire collision threshold value, the operation of budget down regulation is indicated to be effective operation, the operation is reserved, and if the value of the budget down regulation does not exceed the wire collision threshold value, the operation is discarded.
After the operation recall phase, a set of candidate operations that may cause a fluctuation may be obtained. Subsequently, causal parameters of the operations in the candidate operation set can be determined, and the operations are sorted according to the values of the causal parameters.
According to the embodiment of the disclosure, the fluctuating dimension may include, for example, three of exposure amount, click amount, and consumption amount, and the operation types that may cause data fluctuation may include, for example: adding a plan, pausing a plan, increasing a budget, decreasing a budget, etc. Different models can be built for different fluctuation dimensions and different types of operations. Exemplarily, the determination process of the causal parameter is specifically described below by taking the determination of the causal effect of the budget increase operation on the click volume as an example. The budget increasing operation is a positive operation, and the corresponding negative operation is to reduce the account budget. In this embodiment, for example, the data set may include a plurality of sets of data, each set of data including the user characteristic X, the operation value T, and the fluctuation value Y corresponding to each other. The user characteristics X may include, for example, characteristic data of five dimensions of user basic attributes, industry attributes, operation frequency, plan budget, and effective material size. The value of the operation value T may be, for example, the absolute value of the operation value obtained in the recall phase. The fluctuation value Y may include, for example, the fluctuation amplitude of the exposure amount, the click amount, and the consumption amount, and may be expressed by the following equation:
Y＝Curpv/clk/pay-Cmppv/clk/pay
wherein Curpv/clk/payCurrent day value, Cmp, indicating exposure, click, and consumptionpv/clk/payThe values refer to the comparative days of the presentation amount, click amount, and consumption amount.
Fig. 6A schematically illustrates a schematic diagram of a first stage of an information prompting method according to another embodiment of the present disclosure. As shown in fig. 6A, two data sets, sample data set a 601 and sample data set B602, may be set in advance.
A fluctuation prediction model (hereinafter, referred to as a Y model) 610 and an operation prediction model (hereinafter, referred to as a T model) 620 may be trained based on the sample data set a 601.
For example, the click volume fluctuation Y in the sample data set a 601 may be used as a label, the user characteristic X may be used as an input characteristic, and the GBDT model may be used to fit the click volume fluctuation Y to obtain the Y model 610. Wherein the fitted objective function is as follows:
Figure BDA0003063067370000121
where Yi is the true value of the click volume fluctuation value,
Figure BDA0003063067370000122
is the predicted value of the click rate fluctuation value obtained by the GBDT model 620, and N is the total number of samples in the sample data set a 601.
The operation value T may be fitted using the GBDT model 620 using the operation value, namely vbid as a label and the user feature X as an input feature to obtain an X model. Wherein the fitted objective function is as follows:
Figure BDA0003063067370000131
where Ti is the true value of the operating value,
Figure BDA0003063067370000132
is a predicted value of the operating value obtained by the GBDT model 620 and N is the total number of samples.
Then, each sample in the sample data set B602 is predicted by using the trained Y model 610 and T model 620, and let the Y model 610 predict the user feature XiThe Y predicted value 603 obtained as an input is
Figure BDA0003063067370000133
T model 620 combines user features XiThe Y predicted value 604 obtained as an input is
Figure BDA00030630673700001313
Where i is the serial number of the sample in the sample data set B602. Then, the differences between the Y predicted value 603 and the T predicted value 604 and the corresponding real values are calculated to obtain the residual 605 of Y and T,606, which can be expressed as:
Figure BDA00030630673700001314
Figure BDA0003063067370000134
wherein, YiIs the true value of the click volume fluctuation value,
Figure BDA0003063067370000135
is the predicted value of the fluctuation value of click rate, T, obtained by the Y model 610iIs the true value of the operational value or values,
Figure BDA0003063067370000136
is a predicted value of the operating value from T model 620,
Figure BDA0003063067370000137
is the residual error of the click volume fluctuation value,
Figure BDA0003063067370000138
is the residual of the operation value. After the residuals of Y and T are obtained, θ (X) can be modeled.
According to an embodiment of the present disclosure, the following equation holds:
Figure BDA0003063067370000139
where epsilon is the error term. If a value for e (x) is desired by the above formula, one can be fit
Figure BDA00030630673700001310
About
Figure BDA00030630673700001311
The parameter model of (2) may be, and the value of the parameter is the value of θ (x). Based on this, the optimization goal may be set as:
Figure BDA00030630673700001312
but most parametric models have weak non-linear expression capability and are easily over-fitted. Therefore, in this embodiment, a random forest may be selected to fit θ (x). Based on this, the above objective function can be converted into:
Figure BDA0003063067370000141
according to the above-mentioned objective function, can
Figure BDA0003063067370000142
As a label, the label values are fitted using a random forest model, using the user feature x as a feature input, to obtain a corresponding e (x) model (i.e., causal inference model) 630. After the causal inference model is obtained, θ (x) can be obtained by predicting the input x607 which needs to be actually predicted by using the causal inference model 630i) The predicted value 608 of (a).
FIG. 6B schematically shows a diagram of a second stage of an information prompting method according to another embodiment of the present disclosure.
As shown in FIG. 6B, let M be the e (x) model 631 obtained by the above operations1To further prevent overfitting, sample data set a 601 and sample data set B602 may be exchanged, i.e., Y model 610 and T model 620 are fitted using sample data set B602, and θ (x) model 632 is fitted using sample data set a 601, where θ (x) model 632 is set to M2。
To obtain a model M 1631 and M 2632, M may be used respectively1631 and M 2632 are predicted to obtain corresponding predicted values 608, 609 by predicting the input x607 which actually needs to be predicted. The causal parameter (ATT)6010 for the corresponding operation T may be calculated by the following equation:
Figure BDA0003063067370000143
wherein, theta1(x) As model M 1631 the predicted value of θ (x) 608, θ2(x) As model M 2632 is obtained as a predicted value 609 of theta (x)
For each operation in the candidate operation set, the causal parameters can be calculated according to the method, and then the operations with higher causal parameters are output according to the ranking of the causal parameters.
The apparatus for training the model will be described in detail below with reference to fig. 7.
FIG. 7 schematically shows a block diagram of an apparatus for training a model according to an embodiment of the present disclosure.
As shown in fig. 7, the apparatus 700 for training a model may include a first training module 710, a fluctuation prediction module 720, an operation prediction module 730, and a second training module 740.
A first training module 710 may be used to train the fluctuation prediction model and the operation prediction model using the first sample data set.
A fluctuation prediction module 720 may be used to determine a plurality of data fluctuation prediction values based on the second set of sample data using the trained fluctuation prediction model.
An operation prediction module 730, which may be configured to determine a plurality of operation prediction values based on the second set of sample data using the trained operation prediction model.
The second training module 740 may be configured to train the causal inference model using the second set of sample data, the plurality of data fluctuation predictors, and the plurality of operation predictors.
The information presentation apparatus will be described in detail below with reference to fig. 8.
Fig. 8 schematically shows a block diagram of an information presentation apparatus according to an embodiment of the present disclosure.
As shown in fig. 8, the information prompting device 800 may include an acquisition module 810, a first cause and effect prediction module 820, a second cause and effect prediction module 830, a target cause and effect parameter determination module 840, and a prompting information generation module 850.
The obtaining module 810 may be configured to obtain a target data fluctuation value, and a target user characteristic and a target operation value corresponding to the target data fluctuation value, where the target operation value corresponds to at least one target operation.
The first causal prediction module 820 may be configured to input the target user characteristic, the target data fluctuation value, and the target operation value into the trained first causal inference model to obtain a first causal parameter prediction value.
The second causal prediction module 830 may be configured to input the target user characteristic, the target data fluctuation value, and the target operation value into the trained second causal inference model to obtain a second causal parameter prediction value.
The target cause and effect parameter determination module 840 may be configured to determine a target cause and effect parameter predicted value based on the target operating value, the first cause and effect parameter predicted value, and the second cause and effect parameter predicted value.
The hint information generation module 850 may be configured to generate hint information for at least one target operation if the target causal parameter prediction value is greater than the causal parameter threshold, wherein the first causal inference model and the second causal inference model are trained based on a method of training a causal inference model of an embodiment of the present disclosure.
It should be noted that, in the technical solution of the present disclosure, the acquisition, storage, application, and the like of the personal information of the related user all conform to the regulations of the relevant laws and regulations, and do not violate the common customs of the public order.
The present disclosure also provides an electronic device, a readable storage medium, and a computer program product according to embodiments of the present disclosure.
FIG. 9 illustrates a schematic block diagram of an example electronic device 900 that can be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the disclosure described and/or claimed herein.
As shown in fig. 9, the apparatus 900 includes a computing unit 901, which can perform various appropriate actions and processes in accordance with a computer program stored in a Read Only Memory (ROM)902 or a computer program loaded from a storage unit 908 into a Random Access Memory (RAM) 903. In the RAM 903, various programs and data required for the operation of the device 900 can also be stored. The calculation unit 901, ROM 902, and RAM 903 are connected to each other via a bus 904. An input/output (I/O) interface 905 is also connected to bus 904.
A number of components in the device 900 are connected to the I/O interface 905, including: an input unit 906 such as a keyboard, a mouse, and the like; an output unit 907 such as various types of displays, speakers, and the like; a storage unit 908 such as a magnetic disk, optical disk, or the like; and a communication unit 909 such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 909 allows the device 900 to exchange information/data with other devices through a computer network such as the internet and/or various telecommunication networks.
The computing unit 901 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The computing unit 901 performs the various methods and processes described above, such as the method of training the model and/or the information prompting method. For example, in some embodiments, the method information prompting method of training the model may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as storage unit 908. In some embodiments, part or all of the computer program may be loaded and/or installed onto device 900 via ROM 902 and/or communications unit 909. When the computer program is loaded into RAM 903 and executed by computing unit 901, one or more steps of the method information prompting method of training a model described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to perform the method information prompting method of the training model by any other suitable means (e.g., by means of firmware).
Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), Application Specific Standard Products (ASSPs), system on a chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.
Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions/operations specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.
In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic, speech, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.
The computer system may include a client and a server. A user terminal and a server are generally remote from each other and typically interact through a communication network. The relationship of user side and server arises by virtue of computer programs running on the respective computers and having a user side-server relationship to each other.
It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present disclosure may be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, and the present disclosure is not limited herein.
The above detailed description should not be construed as limiting the scope of the disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present disclosure should be included in the scope of protection of the present disclosure.

Claims (18)

1. A method of training a causal inference model, comprising:
training a fluctuation prediction model and an operation prediction model using the first sample dataset;
determining a plurality of data fluctuation prediction values based on the second sample data set using the trained fluctuation prediction model;
determining, with the trained operation prediction model, a plurality of operation prediction values based on the second sample data set; and
training a causal inference model using the second sample data set, the plurality of data fluctuation predicted values, and the plurality of operational predicted values.
2. The method of claim 1, wherein the first sample data set comprises a plurality of first sample data, each of the plurality of first sample data comprising a user characteristic, a data fluctuation value, and an operation value;
the training of the fluctuation prediction model and the operation prediction model using the first sample dataset comprises:
taking a data fluctuation value in each first sample data as a label, and taking a user characteristic in each first sample data as input data, and training the fluctuation prediction model; and
and training an operation prediction model by taking the operation value in each first sample data as a label and taking the user characteristic in each first sample data as input data.
3. The method of claim 2, wherein the training a surge prediction model comprises:
inputting the user characteristics in each first sample data into a fluctuation prediction model to obtain a fluctuation prediction value corresponding to each first sample data;
calculating a first objective function according to the following formula:
Figure FDA0003063067360000011
wherein the Obj1Is the first objective function, the YiIs a data fluctuation value in the ith first sample data in the first sample data set
Figure FDA0003063067360000012
Is a fluctuation predicted value corresponding to the ith first sample data, and N is the total number of first sample data in the first sample data set; and
and under the condition that the first objective function is not converged, adjusting parameters of the fluctuation prediction model according to the value of the first objective function, and returning to the step of inputting the user characteristics in each first sample data into the fluctuation prediction model.
4. The method of claim 2, wherein the training an operational predictive model comprises:
inputting the user characteristics in each first sample data into an operation prediction model to obtain an operation prediction value corresponding to each first sample data;
calculating a second objective function according to the following formula:
Figure FDA0003063067360000021
wherein the Obj2Is the second objective function, the TiIs an operation value in the ith first sample data in the first sample data set
Figure FDA0003063067360000022
Is an operation prediction value corresponding to the ith first sample data, and N is the total number of first sample data in the first sample data set; and
and under the condition that the second objective function does not converge, adjusting parameters of the operation prediction model according to the value of the second objective function, and returning to the step of inputting the user characteristics in each first sample data into the operation prediction model.
5. The method of claim 1, wherein said training a causal inference model using said second sample data set, said plurality of data fluctuation predictors, and said plurality of operation predictors comprises:
for each second sample data of the second set of sample data,
determining a fluctuation value residual error according to the data fluctuation value in the second sample data and a data fluctuation predicted value obtained based on the second sample data;
determining an operation value residual error according to the operation value in the second sample data and an operation predicted value obtained based on the second sample data;
determining a causal parameter sample according to the fluctuation value residual error and the operation value residual error; and
and training the causal inference model by taking the causal parameter sample as a label and taking the user characteristics in the second sample data as input data.
6. The method of claim 5, wherein the training the causal inference model with the causal parameter sample as a label and user features in the second sample data as input data comprises:
inputting the user characteristics in each second sample data into a causal inference model to obtain a causal parameter predicted value corresponding to each second sample data;
calculating a third objective function according to the following formula:
Figure FDA0003063067360000031
wherein the Obj3Is the third objective function, the
Figure FDA0003063067360000032
Is the operation value residual error corresponding to the ith second sample data in the second sample data set
Figure FDA0003063067360000033
Is the fluctuation value residual corresponding to the ith second sample data, said θ (x)i) For a causal parameter prediction value corresponding to the ith second sample data, M is the total number of second sample data in the second sample data set; and
and under the condition that the third objective function is not converged, adjusting parameters of the causal inference model according to the value of the third objective function, and returning to the step of inputting the user characteristics in each second sample data into the causal inference model.
7. The method of claim 1, wherein the user characteristics include at least one of a base attribute, an industry attribute, an operating frequency, a projected budget, and an effective material size.
8. The method of claim 1, wherein the fluctuation prediction model and the operation prediction model comprise gradient lifting tree models and the causal inference model comprises a random forest model.
9. An information prompting method comprises the following steps:
acquiring a target data fluctuation value, and a target user characteristic and a target operation value corresponding to the target data fluctuation value, wherein the target operation value corresponds to at least one target operation;
inputting the target user characteristics, the target data fluctuation value and the target operation value into a trained first causal inference model to obtain a first causal parameter prediction value;
inputting the target user characteristic, the target data fluctuation value and the target operation value into a trained second causal inference model to obtain a second causal parameter prediction value;
determining a target cause and effect parameter predicted value according to the target operation value, the first cause and effect parameter predicted value and the second cause and effect parameter predicted value; and
generating hint information for the at least one target operation if the target causal parameter prediction value is greater than a causal parameter threshold,
wherein the first causal inference model and the second causal inference model are trained based on the method of any of claims 1-8.
10. The method of claim 9, wherein a first sample data set used in training the first causal inference model is the same as a second sample data set used in training the second causal inference model, and wherein the second sample data set used in training the first causal inference model is the same as the first sample data set used in training the second causal inference model.
11. The method of claim 9, wherein the determining a target causal parameter predictor from the target operational value, the first causal parameter predictor and a second causal parameter predictor comprises:
calculating the target causal parameter prediction value according to the following formula:
Figure FDA0003063067360000041
wherein the ATT is the target causal parameter predicted value, theta1(x) For a first causal parameter prediction value, said θ2(x) And T is a second causal parameter predicted value and is the target operation value.
12. The method of claim 9, wherein the obtaining a target data fluctuation value comprises:
acquiring original data and comparison data;
calculating a difference between the original data and the comparison data; and
and under the condition that the difference value is larger than a preset fluctuation threshold value, determining the difference value as the target data fluctuation value.
13. The method of claim 9, wherein the obtaining a target operational value corresponding to the target data fluctuation value comprises:
determining at least one target operation related to the target data fluctuation value, wherein the target operation is used for adjusting parameters; and
and determining the target operation value according to the variation of the parameter after each target operation in the at least one target operation is adjusted and the duration of the adjustment.
14. An apparatus for training a model, comprising:
a first training module to train a fluctuation prediction model and an operation prediction model using the first sample dataset;
a fluctuation prediction module to determine a plurality of data fluctuation prediction values based on the second sample data set using the trained fluctuation prediction model;
an operation prediction module to determine a plurality of operation prediction values based on the second set of sample data using a trained operation prediction model; and
a second training module to train a causal inference model using the second set of sample data, the plurality of data fluctuation predicted values, and the plurality of operational predicted values.
15. An information presentation device comprising:
the device comprises an acquisition module, a storage module and a processing module, wherein the acquisition module is used for acquiring a target data fluctuation value, and a target user characteristic and a target operation value which correspond to the target data fluctuation value, and the target operation value corresponds to at least one target operation;
a first cause and effect prediction module, configured to input the target user characteristic, the target data fluctuation value, and the target operation value into a trained first cause and effect inference model to obtain a first cause and effect parameter prediction value;
the second cause and effect prediction module is used for inputting the target user characteristics, the target data fluctuation value and the target operation value into a trained second cause and effect inference model to obtain a second cause and effect parameter prediction value;
a target cause and effect parameter determination module for determining a target cause and effect parameter predicted value according to the target operation value, the first cause and effect parameter predicted value and the second cause and effect parameter predicted value; and
a hint information generation module to generate hint information for the at least one target operation if the target causal parameter prediction value is greater than a causal parameter threshold,
wherein the first causal inference model and the second causal inference model are trained based on the method of any of claims 1-8.
16. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-13.
17. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1-13.
18. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-13.
CN202110519634.8A 2021-05-12 2021-05-12 Method for training causal inference model and information prompting method and device Active CN113112311B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110519634.8A CN113112311B (en) 2021-05-12 2021-05-12 Method for training causal inference model and information prompting method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110519634.8A CN113112311B (en) 2021-05-12 2021-05-12 Method for training causal inference model and information prompting method and device

Publications (2)

Publication Number Publication Date
CN113112311A true CN113112311A (en) 2021-07-13
CN113112311B CN113112311B (en) 2023-07-25

Family

ID=76722068

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110519634.8A Active CN113112311B (en) 2021-05-12 2021-05-12 Method for training causal inference model and information prompting method and device

Country Status (1)

Country Link
CN (1) CN113112311B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114664452A (en) * 2022-05-20 2022-06-24 之江实验室 General multi-disease prediction system based on causal verification data generation

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160148223A1 (en) * 2014-11-21 2016-05-26 Adobe Systems Incorporated Forecasting web metrics using statistical causality based feature selection
CN110245984A (en) * 2019-06-09 2019-09-17 广东工业大学 A kind of shopping at network behavior analysis method and system based on causal inference
CN110674019A (en) * 2019-08-30 2020-01-10 中国人民财产保险股份有限公司 Method and device for predicting system fault and electronic equipment
WO2020078098A1 (en) * 2018-10-17 2020-04-23 阿里巴巴集团控股有限公司 Gradient boosting decision tree-based method and device for model training
CN111428008A (en) * 2020-06-11 2020-07-17 北京百度网讯科技有限公司 Method, apparatus, device and storage medium for training a model
CN112016689A (en) * 2019-05-28 2020-12-01 株式会社日立制作所 Information processing apparatus, prediction discrimination system, and prediction discrimination method
CN112685674A (en) * 2020-12-30 2021-04-20 百果园技术(新加坡)有限公司 Feature evaluation method and device influencing user retention
WO2021081962A1 (en) * 2019-10-31 2021-05-06 华为技术有限公司 Recommendation model training method, recommendation method, device, and computer-readable medium

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160148223A1 (en) * 2014-11-21 2016-05-26 Adobe Systems Incorporated Forecasting web metrics using statistical causality based feature selection
WO2020078098A1 (en) * 2018-10-17 2020-04-23 阿里巴巴集团控股有限公司 Gradient boosting decision tree-based method and device for model training
CN112016689A (en) * 2019-05-28 2020-12-01 株式会社日立制作所 Information processing apparatus, prediction discrimination system, and prediction discrimination method
CN110245984A (en) * 2019-06-09 2019-09-17 广东工业大学 A kind of shopping at network behavior analysis method and system based on causal inference
CN110674019A (en) * 2019-08-30 2020-01-10 中国人民财产保险股份有限公司 Method and device for predicting system fault and electronic equipment
WO2021081962A1 (en) * 2019-10-31 2021-05-06 华为技术有限公司 Recommendation model training method, recommendation method, device, and computer-readable medium
CN111428008A (en) * 2020-06-11 2020-07-17 北京百度网讯科技有限公司 Method, apparatus, device and storage medium for training a model
CN112685674A (en) * 2020-12-30 2021-04-20 百果园技术(新加坡)有限公司 Feature evaluation method and device influencing user retention

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
李国;江晓东;: "基于提升回归树与随机森林的风电功率集成预测方法", 电力系统及其自动化学报, no. 11 *
范程岸;刘博文;: "基于GBDT和神经网络的光伏发电组合预测", 无线互联科技, no. 01 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114664452A (en) * 2022-05-20 2022-06-24 之江实验室 General multi-disease prediction system based on causal verification data generation
CN114664452B (en) * 2022-05-20 2022-09-23 之江实验室 General multi-disease prediction system based on causal verification data generation

Also Published As

Publication number Publication date
CN113112311B (en) 2023-07-25

Similar Documents

Publication Publication Date Title
CN112148987A (en) Message pushing method based on target object activity and related equipment
JP2022033695A (en) Method, device for generating model, electronic apparatus, storage medium and computer program product
CN114494784A (en) Deep learning model training method, image processing method and object recognition method
CN114202076A (en) Deep learning model training method, natural language processing method and device
CN114118570A (en) Service data prediction method and device, electronic equipment and storage medium
CN114065864A (en) Federal learning method, federal learning device, electronic device, and storage medium
CN113112311A (en) Method for training causal inference model, information prompting method and device
US20200410296A1 (en) Selective Data Rejection for Computationally Efficient Distributed Analytics Platform
CN115185606A (en) Method, device, equipment and storage medium for obtaining service configuration parameters
CN113408641B (en) Training of resource generation model and generation method and device of service resource
CN114925275A (en) Product recommendation method and device, computer equipment and storage medium
CN115795146A (en) Method, device and equipment for determining resources to be recommended and storage medium
CN115759283A (en) Model interpretation method and device, electronic equipment and storage medium
CN115099875A (en) Data classification method based on decision tree model and related equipment
CN114707638A (en) Model training method, model training device, object recognition method, object recognition device, object recognition medium and product
CN113408632A (en) Method and device for improving image classification accuracy, electronic equipment and storage medium
CN114021642A (en) Data processing method and device, electronic equipment and storage medium
CN114067415A (en) Regression model training method, object evaluation method, device, equipment and medium
CN113806541A (en) Emotion classification method and emotion classification model training method and device
CN113361574A (en) Training method and device of data processing model, electronic equipment and storage medium
CN113553407B (en) Event tracing method and device, electronic equipment and storage medium
US20240144135A1 (en) Systems and methods for risk processing of supply chain management system data
CN113344405B (en) Method, device, equipment, medium and product for generating information based on knowledge graph
CN117744808A (en) Decision method and device based on causal effect estimation model and related equipment
CN117892008A (en) Tool item recommendation method, device, electronic equipment, storage medium and computer program product

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant