Method, device, electronic device and storage medium for determining heterogeneous causal effects

Abstract

The present disclosure is about a method, device, electronic device and storage medium for determining heterogeneous causal effects, which relates to the field of computer technology and at least solves the problem that the heterogeneous causal effects determined in the related technology are not accurate enough. The method includes: generating a causal forest including at least one causal tree according to the behavior data of the user account; determining a test effect relationship set corresponding to the causal tree including the test effect relationship of each leaf node according to a preset function; determining a heterogeneous causal effect used to represent the response degree of each user account to the functional test task according to the test effect relationship set corresponding to all causal trees in the causal forest.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a method, device, electronic device and storage medium for determining heterogeneous causal effects. Background Art

[0002] In related technologies, when determining which function of a content community application can affect the usage behavior of a user account, causal forests are usually used to perform functional test estimation. Causal forests are a method for estimating heterogeneous causal effects. Compared with traditional methods for estimating causal effects, the core idea of ​​causal forests is to calculate the causal effect of each user account for this function, and then determine the user account that can have the greatest behavioral response to this function, and adopt corresponding strategies for the user account. Specifically, the heterogeneous causal effect estimation for each function can be regarded as a test.

[0003] Considering that there are many functions in a content community application, the heterogeneous causal effect estimation of a content community application will correspond to multiple test tasks. In this case, the heterogeneous causal effect determined by the current heterogeneous causal effect estimation method will be inaccurate due to large deviations, which will affect subsequent decision-making and strategy execution. Summary of the invention

[0004] The present disclosure provides a method, device, electronic device and storage medium for determining heterogeneous causal effects, so as to at least solve the problem that the heterogeneous causal effects determined in the related art are not accurate enough.

[0005] The technical solution of the present disclosure is as follows:

[0006] According to a first aspect of an embodiment of the present disclosure, a method for determining heterogeneous causal effects is provided, the method comprising: generating a causal forest including at least one causal tree based on behavioral data of multiple user accounts; each causal tree includes at least one leaf node, one leaf node corresponds to a category of a user account, and the leaf node includes user accounts belonging to the corresponding category among multiple user accounts; according to a preset function, determining a test effect relationship set corresponding to each causal tree, including the test effect relationship of each leaf node of the causal tree; the preset function includes one or more test variables corresponding to one or more functional test tasks; the test effect relationship of the leaf node is used to represent the nonlinear superposition relationship between the degree to which multiple user accounts on the leaf node are respectively affected by all test variables and all test variables corresponding to the leaf node; determining the heterogeneous causal effect of each user account based on the test effect relationship set corresponding to all causal trees in the causal forest; the heterogeneous causal effect of each user account is used to represent the response degree of each user account to one or more functional test tasks.

[0007] Optionally, the above-mentioned “generating a causal forest including at least one causal tree based on the behavioral data of multiple user accounts” includes: obtaining the behavioral data of multiple user accounts and preset causal parameters; the preset causal parameters are used to represent the number of causal trees in the causal forest and the hierarchical depth of the multiple causal trees; based on the behavioral data of multiple user accounts and the preset causal parameters, for each causal tree, classifying the multiple user accounts, and storing the user accounts belonging to the same category in the corresponding leaf nodes.

[0008] Optionally, the above-mentioned "determining a set of test effect relationships corresponding to each causal tree and including the test effect relationship of each leaf node of the causal tree according to a preset function" includes: constructing a preset function according to one or more test variables corresponding to one or more functional test tasks and parameter variables of the causal forest; for each leaf node in each causal tree, determining the change relationship corresponding to the preset function when one or more test variables change, and determining all change relationships of the preset function as the test effect relationship of the leaf node.

[0009] Optionally, in the above method for determining heterogeneous causal effects, the heterogeneous causal effects include the weighted average test effect and the standard error of the test effect for each user account, and the standard error of the test effect is used to screen the weighted average test effect.

[0010] Optionally, the above-mentioned “determining the heterogeneous causal effect of each user account based on the test effect relationship set corresponding to all causal trees in the causal forest” includes: determining the weighted average test effect of each user account based on the test effect relationship set corresponding to all causal trees in the causal forest; determining the standard error of the test effect of each user account based on the weighted average test effect of each user account and the value of the parameter variable of the causal forest.

[0011] Optionally, the above-mentioned "determining the weighted average test effect of each user account based on the test effect relationship set corresponding to all causal trees in the causal forest" includes: inputting one or more test variables of one or more functional testing tasks into a preset algorithm model to obtain the value of the parameter variable of the causal forest output by the preset algorithm model; determining the test effect of each user account in different causal trees according to the value of the parameter variable and the test effect relationship set corresponding to each causal tree; determining the weighted average test effect of each user account based on all the test effects of each user account.

[0012] According to a second aspect of an embodiment of the present disclosure, a device for determining heterogeneous causal effects is provided, the device comprising a causal forest generation unit, a test effect relationship acquisition unit and a heterogeneous causal effect acquisition unit; the causal forest generation unit is used to generate a causal forest including at least one causal tree according to the behavior data of multiple user accounts; each causal tree includes at least one leaf node, one leaf node corresponds to a category of a user account, and the leaf node includes user accounts belonging to the corresponding category among multiple user accounts; the test effect relationship acquisition unit is used to determine, according to a preset function, a test effect relationship set including the test effect relationship of each leaf node of the causal tree corresponding to each causal tree; the preset function includes one or more test variables corresponding to one or more functional test tasks; the test effect relationship of the leaf node is used to represent the nonlinear superposition relationship between the degree to which multiple user accounts on the leaf node are respectively affected by all test variables and all test variables corresponding to the leaf node; the heterogeneous causal effect acquisition unit is used to determine the heterogeneous causal effect of each user account according to the test effect relationship set corresponding to all causal trees in the causal forest; the heterogeneous causal effect of each user account is used to represent the response degree of each user account to one or more functional test tasks.

[0013] Optionally, the above-mentioned causal forest generation unit is specifically used to: obtain behavioral data and preset causal parameters of multiple user accounts; the preset causal parameters are used to represent the number of causal trees in the causal forest and the hierarchical depth of the multiple causal trees; according to the behavioral data and preset causal parameters of multiple user accounts, for each causal tree, classify and process multiple user accounts, and save user accounts belonging to the same category in corresponding leaf nodes.

[0014] Optionally, the above-mentioned test effect relationship acquisition unit is specifically used to: construct a preset function based on one or more test variables corresponding to one or more functional test tasks and the parameter variables of the causal forest; for each leaf node in each causal tree, determine the change relationship corresponding to the preset function when one or more test variables change, and determine all change relationships of the preset function as the test effect relationship of the leaf node.

[0015] Optionally, the above-mentioned heterogeneous causal effect includes a weighted average test effect and a standard error of the test effect for each user account, and the standard error of the test effect is used to screen the weighted average test effect.

[0016] Optionally, the above-mentioned heterogeneous causal effect acquisition unit is specifically used to: determine the weighted average test effect of each user account based on the test effect relationship set corresponding to all causal trees in the causal forest; determine the standard error of the test effect of each user account based on the weighted average test effect of each user account and the value of the parameter variable of the causal forest.

[0017] Optionally, the above-mentioned heterogeneous causal effect acquisition unit is specifically used to: input one or more test variables of one or more functional testing tasks into a preset algorithm model to obtain the values ​​of parameter variables of the causal forest output by the preset algorithm model; determine the test effects of each user account in different causal trees according to the values ​​of the parameter variables and the test effect relationship set corresponding to each causal tree; and determine the weighted average test effect of each user account based on all the test effects of each user account.

[0018] According to a third aspect of an embodiment of the present disclosure, there is provided an electronic device, comprising a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to execute instructions to implement the method for determining the heterogeneous causal effect described in any optional implementation method of the first aspect above.

[0019] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method for determining the heterogeneous causal effect described in any optional implementation method of the first aspect above.

[0020] According to a fifth aspect of an embodiment of the present disclosure, there is provided a computer program product, comprising computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the method for determining the heterogeneous causal effect described in any optional implementation of the first aspect.

[0021] The technical solution provided by the embodiment of the present disclosure brings at least the following beneficial effects: Since each leaf node includes user accounts belonging to the same category, the test effect relationship of each leaf node on the causal tree is determined by using a preset function, and the test effect relationship of each user account of this category on the leaf node is determined. Each user account has a test effect relationship on each causal tree, and finally all the test effect relationships of each user account are integrated to further determine the heterogeneous causal effect of each user account. Compared with the related art that only considers the linear influence of the test variable on the user account behavior when the test variables do not affect each other, the solution of the embodiment of the present disclosure can take into account the nonlinear superposition influence of the test variable on the user account behavior when the test variables have a mutually influential relationship. And the nonlinear superposition influence of the test variable on the user account behavior can be determined by a preset function, and the determined nonlinear superposition influence can be used as the heterogeneous causal effect of the user account, so that when the test variables have a mutually influential relationship, the determined heterogeneous causal effect can be more accurate and comprehensive.

[0022] 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

[0023] 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.

[0024] Figure 1 is a structural diagram of a causal effect determination system according to an exemplary embodiment of the present disclosure;

[0025] Figure 2 is a flowchart of a method for determining heterogeneous causal effects according to an exemplary embodiment of the present disclosure;

[0026] Figure 3 is a structural schematic diagram of a causal forest according to an exemplary embodiment of the present disclosure;

[0027] Figure 4 is a flowchart of another method for determining heterogeneous causal effects according to an exemplary embodiment of the present disclosure;

[0028] Figure 5 is a flowchart of another method for determining heterogeneous causal effects according to an exemplary embodiment of the present disclosure;

[0029] Figure 6 is a flowchart of another method for determining heterogeneous causal effects according to an exemplary embodiment of the present disclosure;

[0030] Figure 7 is a flowchart of another method for determining heterogeneous causal effects according to an exemplary embodiment of the present disclosure;

[0031] Figure 8 is a structural schematic diagram of a device for determining heterogeneous causal effects according to an exemplary embodiment of the present disclosure;

[0032] Fig. 9 The figure is a schematic diagram showing the structure of a server according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0033] 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.

[0034] It should be noted that, in the embodiments of the present disclosure, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present disclosure should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0035] 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.

[0036] In the embodiments of the present disclosure, "at least one" means one or more, and "plurality" means two or more.

[0037] It should also be understood that the term “comprising” indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements and / or components.

[0038] The method for determining heterogeneous causal effects provided by the embodiments of the present disclosure may be applicable to a causal effect determination system. Figure 1 FIG. 1 is a schematic diagram of a causal effect determination system according to an exemplary embodiment of the present disclosure. Figure 1 As shown, the causal effect determination system includes at least one terminal device 100 and at least one server 110. The terminal device 100 is connected to the server 110. The terminal device 100 communicates with the server 110 via a wired communication method or a wireless communication method. The terminal device 100 can also be used to interact with the server 110 for data.

[0039] The terminal device 100 may be any computer device, including but not limited to mobile phones, tablet computers, desktop computers, laptop computers, vehicle-mounted terminals, handheld terminals, augmented reality (AR) devices, virtual reality (VR) devices, and other devices that can install and use content community applications (such as Kuaishou). The embodiments of the present disclosure do not impose any special restrictions on the specific form of the terminal device 100. It can interact with the user account through one or more methods such as keyboard, touchpad, touch screen, remote control, voice interaction or handwriting device, thereby generating behavioral data of the user account.

[0040] The behavioral data of the user account includes, but is not limited to, the length of time the user account uses the content community application, the number of comments the user account makes on different types of videos in the content community application, the length of time the user account browses different types of videos in the content community application, etc.

[0041] The server 110 may be a single server, or a server cluster composed of multiple servers or a cloud computing service center, which is not limited in the embodiments of the present disclosure. The terminal device 100 may send the collected behavior data of the user account on a certain content community application to the server 110. After receiving the behavior data of all user accounts, the server 110 classifies the user accounts and applies at least one functional testing task on the content community application to the user accounts, determines the degree of response of each user account to the functional testing task, and thereby obtains a heterogeneous causal effect. The server 110 may also further determine the decision to recommend specific functions on the content community application to the user account based on the heterogeneous causal effect, and send the decision data to the terminal device 100.

[0042] It is understandable that the terminal device 100 and the server 110 may be independently configured or integrated into one device, and this is not limited in the embodiments of the present disclosure.

[0043] In the following embodiments provided in the present disclosure, the terminal device 100 and the server 110 are independently configured for explanation.

[0044] In practical applications, the method for determining heterogeneous causal effects provided by the embodiments of the present disclosure can be applied to electronic devices, which can specifically be terminal devices or servers.

[0045] In the related art, when determining which function of a content community application can affect the usage behavior of a user account, the server 110 usually uses a causal forest to perform function test estimation. Causal forest is a method for estimating heterogeneous causal effects. Compared with traditional methods for estimating causal effects, the core idea of ​​causal forest is to calculate the causal effect of each user account for this function, and then determine the user account that can have the greatest behavioral response to this function, and adopt corresponding strategies for the user account. Specifically, the heterogeneous causal effect estimation for each function can be regarded as a test task.

[0046] Considering that there will be many functions in a content community application, the heterogeneous causal effect estimation of a content community application will correspond to multiple test tasks. In the current heterogeneous causal effect estimation method, it is usually assumed that various functional test tasks are independent of each other. For example, in the heterogeneous causal effect estimation of a short video software, the user account receives two test tasks: multiple exposure live videos and multiple exposure face value videos. If the functional test task determines that the user account has increased the time spent using the short video software, it will be considered that the user account has a 1+1=2 effect. However, in actual application scenarios, it is not difficult to find that the various test tasks actually have a mutually influential relationship. For example, user accounts who like to browse short videos may produce a 1+1>2 effect when they receive two test tasks: multiple exposure face value videos and multiple exposure friend videos; while for user accounts who like to watch live broadcasts, if they receive these two test tasks, then they may produce a 1+1<2 effect.

[0047] Therefore, in the process of estimating heterogeneous causal effects, if there is a mutual influence relationship between the test tasks, the heterogeneous causal effects determined by the current method of estimating heterogeneous causal effects will be inaccurate due to large deviations, which will affect subsequent decision-making and strategy execution.

[0048] In order to avoid the estimation results with large deviations in the current estimation method of heterogeneous causal effects, the embodiment of the present disclosure provides a method for determining heterogeneous causal effects. The following describes the method for determining heterogeneous causal effects provided by the embodiment of the present disclosure by taking the application of the method for determining heterogeneous causal effects to a server as an example in conjunction with the accompanying drawings.

[0049] Figure 2 FIG. 1 is a flow chart of a method for determining heterogeneous causal effects according to an exemplary embodiment of the present disclosure. Figure 2 As shown, the method for determining heterogeneous causal effects provided by the embodiment of the present disclosure may include the following steps S101-S103.

[0050] S101. Generate a causal forest including at least one causal tree according to behavior data of multiple user accounts.

[0051] The causal forest consists of one or more causal trees, which are machine learning methods based on regression trees. Each causal tree in the causal forest has its own hierarchical depth and classification criteria. When performing heterogeneous causal effect analysis on multiple user accounts based on the causal forest, multiple user accounts can be classified based on each causal tree first. Because the classification criteria of each causal tree are different, the classification of multiple user accounts by each causal tree is also different.

[0052] The causal tree includes at least one level and at least one leaf node, and each leaf node includes a user account of a category determined based on the current causal tree classification standard. All user accounts exist in each causal tree respectively. Since the causal trees have different level depths and classification standards, the same user account may exist in leaf nodes of different levels in different causal trees, that is, the categories to which the same user account belongs in different causal trees may be different.

[0053] For example, Figure 3 As shown, the causal forest may include causal tree A, causal tree B and causal tree C, and the user accounts that need to be classified may be user account 1, user account 2, user account 3, user account 4 and user account 5. Causal tree A has two levels and three leaf nodes, causal tree B has three levels and five leaf nodes, and causal tree C has one level and one leaf node. Through the classification methods or classification standards of the causal trees, causal tree A divides the above five user accounts into three categories, where user account 1 and user account 3 exist on the same leaf node, user account 2 and user account 4 exist on the same leaf node, and user account 5 exists on a leaf node; causal tree B divides the above five user accounts into five categories, and the five user accounts exist on a leaf node respectively; causal tree C divides all the above five user accounts into one category, so the five user accounts all exist on a leaf node.

[0054] Taking user account 5 as an example, user account 5 is located at different leaf nodes in causal tree A, causal tree B and causal tree C respectively, and thus the categories to which user account 5 belongs in different causal trees may be different.

[0055] S102. Determine, according to a preset function, a test effect relationship set corresponding to each causal tree, the test effect relationship set including the test effect relationship of each leaf node of the causal tree.

[0056] The preset function may be in any function form, such as basic elementary functions such as linear function, exponential function, power function, logarithmic function, polynomial function, or a composite function composed of them.

[0057] Exemplarily, in the embodiments of the present disclosure, y=f(W1, W2, ...)+∈ can be used as the expression of the preset function, where f(W1, W2, ...) can represent the above-mentioned linear function, exponential function, power function, logarithmic function, polynomial function and other basic elementary functions or composite functions composed of them, W1, W2, ... represent test variables corresponding to different functional testing tasks, ∈ represents a constant, and y represents the degree to which the user account is affected by the superposition of one or more test variables after the above-mentioned W1, W2, ... and other test variables are applied to the user account on a leaf node.

[0058] If multiple functional testing tasks are independent of each other and do not affect each other, in the above preset function, f(W1, W2, ...) can be a linear function, then the degree to which multiple user accounts on a leaf node are affected by the superposition has a certain linear superposition relationship with the test variable.

[0059] If multiple test variables have a mutual influence relationship, in the above preset function, f(W1, W2, ...) can be other functions or composite functions other than the above linear functions. Then the degree to which multiple user accounts on a leaf node are respectively affected by the superposition has a certain nonlinear superposition relationship with the test variables. In the embodiment of the present disclosure, this relationship is called the test effect relationship of the leaf node.

[0060] After obtaining the test effect relationship of each leaf node on a causal tree, all the test effect relationships are integrated to obtain the test effect relationship set corresponding to the causal tree.

[0061] S103. Determine the heterogeneous causal effect of each user account according to the test effect relationship set corresponding to all causal trees in the causal forest.

[0062] Exemplarily, the heterogeneous causal effect may include a weighted average test effect and a standard error of the test effect for each user account, wherein the standard error of the test effect is used to screen the weighted average test effect.

[0063] Still Figure 3Taking the causal forest shown as an example, the test effect relationship set corresponding to causal tree A includes test effect relationships corresponding to three leaf nodes, the test effect relationship set corresponding to causal tree B includes test effect relationships corresponding to five leaf nodes, and the test effect relationship set corresponding to causal tree C includes a test effect relationship corresponding to one leaf node. Among them, user account 5 has test effect relationships on causal tree A, causal tree B and causal tree C respectively. In order to make a more accurate causal effect analysis of user account 5, all test effect relationships of user account 5 can be integrated to determine the weighted average test effect and test effect standard error of user account 5. In the same way, the weighted average test effect and test effect standard error of user account 4 of user account 1, user account 2, and user account 3 can be determined respectively.

[0064] When adopting corresponding strategies for user accounts according to heterogeneous causal effects in the future, the weighted average test effect of the user account can be compared with the preset effect threshold. If the difference between the two exceeds the standard error of the test effect of the user account, it means that the change between the weighted average test effect and the preset effect threshold is large. At this time, in the subsequent decision-making process, there is no need to consider the weighted average test effect. When the difference between the two does not exceed the standard error of the test effect of the user account, it means that the change between the weighted average test effect and the preset effect threshold meets the standard error requirements, and the impact of the changing test variables in the weighted average test effect on the user account can be considered in the future. It can be seen that the weighted average test effect can be screened through the standard error of the test effect, and the weighted average test effect that meets the standard error requirements can be retained, so that the subsequent decision-making process has a more accurate basis.

[0065] According to the above method, the weighted average test effect and the standard error of the test effect of each user account can be obtained, and then the heterogeneous causal effect used to represent the response degree of each user account to the functional test task can be obtained.

[0066] The technical solution provided by the above embodiment can at least bring the following beneficial effects: Since each leaf node includes user accounts belonging to the same category, the test effect relationship of each leaf node on the causal tree is determined by using a preset function, and the test effect relationship of each user account of this category on the leaf node is determined. Each user account has a test effect relationship on each causal tree, and finally all the test effect relationships of each user account are integrated to further determine the heterogeneous causal effect of each user account. Compared with the related art that only considers the linear influence of the test variable on the user account behavior when the test variables do not affect each other, the solution of the embodiment of the present disclosure can take into account the nonlinear superposition influence of the test variable on the user account behavior when the test variables have a mutually influential relationship. And the nonlinear superposition influence of the test variable on the user account behavior can be determined by a preset function, and the determined nonlinear superposition influence can be used as the heterogeneous causal effect of the user account, so that when the test variables have a mutually influential relationship, the determined heterogeneous causal effect can be more accurate and comprehensive.

[0067] In one embodiment, if Figure 4 As shown, the method for generating a causal forest including at least one causal tree based on the behavioral data of multiple user accounts may further specifically include the following steps S201-S202.

[0068] S201. Obtain behavioral data of multiple user accounts and preset causal parameters.

[0069] The behavior data of the user account is generated by the user account using the content community application on the terminal device 100 and is collected by the terminal device 100. It is understandable that one content community application can usually be installed on one terminal device 100, but at least one user account can be logged in to one content community application. For example, if three user accounts have been logged in to one content community application, then the terminal device 100 needs to collect the behavior data of these three user accounts.

[0070] The preset causal parameters are parameters for generating a causal forest, including the number of causal trees in the causal forest and the hierarchical depth of multiple causal trees. They may also include classification methods or classification standards for generating different leaf nodes on the causal tree. The causal trees are independent of each other. The different hierarchical depths and classification standards of the causal trees mean that the distribution and number of leaf nodes in the causal trees may be different, and different causal trees will have different classifications for user accounts.

[0071] It should be noted that the behavioral data and other information of the user account involved in the embodiments of the present disclosure (including but not limited to device information, device data, user's personal information, user's personal data, etc. of the device where the user account is located) are all information authorized by the user or fully authorized by all parties.

[0072] S202. Classify the multiple user accounts according to the behavior data of the multiple user accounts and the preset causal parameters for each causal tree, and store the user accounts belonging to the same category in the corresponding leaf nodes.

[0073] The number of causal trees in the causal forest and the structure of each causal tree can be determined according to the preset causal parameters. Then, according to the classification parameters or classification requirements of each causal tree, all user accounts are classified for each causal tree.

[0074] Among them, the classification standard or classification requirement of the causal tree can be determined based on the behavioral characteristics of the user account, and the behavioral characteristics can be reflected based on the behavioral data. For example, if in the behavioral data of the user account, the browsing time of the user account browsing type a video in the content community application exceeds the preset browsing time, then the time for the user account to browse type a video can be used as a behavioral feature to classify the user account into a category; or, in the behavioral data of the user account, the number of comments on type b videos by the user account in the content community application exceeds the preset number of comments, then the number of comments on type b videos by the user account can be used as a behavioral feature to classify the user account into a category.

[0075] For example, the causal tree finally generated can be found in the above Figure 3 In causal tree A, user account 1 and user account 3 belong to the same category, user account 2 and user account 4 belong to the same category, and user account 5 belongs to a category alone; in causal tree B, the five user accounts each belong to a category; in causal tree C, the five user accounts all belong to the same category.

[0076] The technical solution provided by the present disclosure can at least bring the following beneficial effects: in the process of generating a causal tree, based on the behavioral data of the user account, the user account can be classified based on different behavioral characteristics; based on the preset causal parameters, a preset number of causal trees can be generated, and the hierarchy of each causal tree will also be different. During the classification recursion, user accounts in different categories are placed on different leaf nodes of the causal tree, so that the generated causal tree has hierarchical relationships and logical relationships between user accounts of different categories. When using the classified causal tree to determine the heterogeneous causal effect, all user accounts belonging to the same category have the same test effect relationship, and there is no need to obtain the test effect relationship for each user account separately, which reduces a large number of calculation operations, making the method for determining the heterogeneous causal effect faster and more convenient.

[0077] In one embodiment, if Figure 5 As shown, the method for determining the test effect relationship set corresponding to each causal tree and including the test effect relationship of each leaf node of the causal tree according to a preset function can also specifically include the following steps S301-S302.

[0078] S301. Construct a preset function according to one or more test variables corresponding to one or more functional test tasks and parameter variables of the causal forest.

[0079] In order to determine the responsiveness of user accounts to different functions in content community applications, it is necessary to perform functional testing tasks on user accounts of different categories, that is, to provide the same functions to user accounts of different categories and obtain the behaviors or reactions of user accounts to these functions. These behaviors or reactions are quantified to determine the responsiveness of user accounts.

[0080] As can be seen from the foregoing, in the embodiments of the present disclosure, the responsiveness of the user account can be determined by using a preset function. Before constructing the preset function, it is necessary to first determine the functional test task imposed on the user account and the test variables corresponding to the functional test task.

[0081] For example, if the functional test tasks imposed on the user account are to provide type a videos and type b videos respectively, then the test variables W1 and W2 corresponding to the two functional test tasks are the number of type a videos provided and the number of type b videos provided respectively.

[0082] In addition, the preset function also includes at least one parameter variable of the causal forest. Exemplarily, the parameter variable can be represented by β. If the parameter variable is greater than one, the parameter variable can be represented by β1, β2, β3, ...

[0083] Referring to the above embodiment, the preset function can be expressed as y=f(W1, W2, ...)+∈. Exemplarily, if there are only two functional test tasks, the preset function is expanded into a polynomial form, that is, y=β1W1+β2W2+β3W1×W2+∈.

[0084] When W1 and W2 have an independent relationship without affecting each other, the value of parameter variable β3 is zero. When W1 and W2 have a mutual influence relationship, the value of parameter variable β3 is not zero. At this time, the determined y represents the nonlinear superposition relationship between the degree to which multiple user accounts are affected by all test variables and all test variables corresponding to the leaf nodes.

[0085] S302. For each leaf node in each causal tree, determine the change relationship corresponding to the preset function when one or more test variables change, and determine all the change relationships of the preset function as the test effect relationship of the leaf node.

[0086] Since each leaf node includes at least one user account belonging to the same category, applying a functional test task to each leaf node can be regarded as applying a functional test task to all user accounts belonging to a category. The behaviors or reactions of all user accounts to the functional test task are processed by quantification, superposition, averaging, etc., and finally the test effect relationship of each leaf node is obtained. Since each leaf node corresponds to a user account of a category, after determining the test effect relationship of a leaf node, this test effect relationship can be used as the test effect relationship of each user account of this category.

[0087] When applying functional test tasks to user accounts, in order to determine which one or which functional test tasks will have a greater response or be more affected by the user account, the test variables will change. The test effect relationship refers to the change relationship corresponding to the preset function when one or more test variables change. That is, when one or more test variables that affect each other change, one or more variables in W1, W2, ... in the above preset function will change, and the function of y finally determined will also have different representations.

[0088] For example, if there are only two mutually affecting functional test tasks, that is, y=f(W1, W2)+∈. When determining the test effect relationship, three forms of y, y1, y2 and y3, can be considered when W1 and W2 change by one unit each and both change by one unit.

[0089] y1=f(W1+1, W2)-f(W1, W2),

[0090] y2=f(W1, W2+1)-f(W1, W2),

[0091] y3=f(W1+1, W2+1)-f(W1, W2).

[0092] y1 represents the change in y when one unit of W1 is added to the user account but W2 remains unchanged, that is, the change in the degree of impact on the user account; y2 represents the change in y when one unit of W2 is added to the user but W1 remains unchanged; y3 represents the change in y when one unit of W1 and W2 are added to the user account at the same time.

[0093] When the test variables change, the values ​​of the parameter variables β1, β2, and β3 of the causal forest are unknown. Then, through the binomial expansion calculation and other processes, we can further obtain the linear relationship between y1 and β1, β2, and β3, the linear relationship between y2 and β1, β2, and β3, and the linear relationship between y3 and β1, β2, and β3.

[0094] The change relationship of the preset function in the embodiment of the present disclosure is the linear relationship between y1, y2, y3 and β1, β2 and β3 respectively. The test effect relationship of each leaf node is the set of y1, y2, y3.

[0095] It is worth noting that when the test variables are changed, the change units between different test variables are the same. The above embodiment is only described by changing one unit as an example. In other achievable embodiments, according to the specific requirements of the functional test task or the behavioral characteristics of the user account, the test variables can also be controlled to change other units, and the embodiments of the present disclosure are not limited to this.

[0096] The technical solution provided by the present disclosure can at least bring the following beneficial effects: in the method for determining heterogeneous causal effects, several functional test tasks are implemented on the user account, and then there will be several test variables in the constructed preset function. Since there are multiple test variables in the preset function, it is necessary to determine the change of the preset function when different test variables change respectively, so as to ensure a more comprehensive estimation and analysis of the heterogeneous causal effect based on each test variable.

[0097] In one embodiment, if Figure 6 As shown, the method for determining the heterogeneous causal effect of each user account according to the test effect relationship set corresponding to all causal trees in the causal forest can also specifically include the following steps S401-S402.

[0098] S401. Determine the weighted average test effect of each user account according to the test effect relationship set corresponding to all causal trees in the causal forest.

[0099] For example, Figure 3 As shown, if there are three leaf nodes in causal tree A, and each leaf node corresponds to a test effect relationship, then the test effect relationships corresponding to the three leaf nodes can together constitute the test effect relationship set corresponding to causal tree A. If the causal forest includes causal tree A, causal tree B, and causal tree C, then in the test effect relationship sets corresponding to all causal trees, user account 5 can have test effect relationship E1 in causal tree A, test effect relationship E2 in causal tree B, and test effect relationship E3 in causal tree C, respectively.

[0100] In the causal forest, each test effect relationship of a user account can represent the nonlinear superposition response degree of the user account to the functional test task. However, in order to determine a more accurate response degree, in S401, the specific numerical value of the parameter variable can also be obtained, and the value of the parameter variable can be substituted into each test effect relationship of the user account, so as to obtain the test effect represented by the specific numerical value corresponding to each test effect relationship, and finally obtain the weighted average test effect of all test effects of the user account.

[0101] In one embodiment, if Figure 7 As shown, the method for determining the weighted average test effect of each user account according to the test effect relationship set corresponding to all causal trees in the causal forest may further specifically include the following steps S501-S503.

[0102] S501. Input one or more test variables of one or more functional test tasks into a preset algorithm model to obtain values ​​of parameter variables of the causal forest output by the preset algorithm model.

[0103] The preset algorithm model can be some commonly used models for obtaining the linear superposition effect of the test variables of the functional test task on the user account behavior. When using the preset algorithm model, the established function is a linear function that assumes that the test variables do not affect each other, the input content is the test variables of different functional test tasks, and the output content is the response degree of the user account to the linear superposition of these test variables based on this linear function. The value of the parameter variable of the causal forest mentioned in the embodiment of the present disclosure is the quantitative value of the response degree of the user account to the linear superposition of the test variables output by the preset algorithm model.

[0104] S502: Determine the test effect of each user account in different causal trees according to the value of the parameter variable and the test effect relationship set corresponding to each causal tree.

[0105] For example, if user account 5 can have test effect relationship E1 in causal tree A, test effect relationship E2 in causal tree B, and test effect relationship E3 in causal tree C, then after obtaining the value of the parameter variable in S502, substituting it into E1, E2, and E3, the test effects e1, e2, and e3 of user account 5 can be obtained respectively.

[0106] For example, taking two mutually influencing test variables as an example, the test effect of the user account in a causal tree is calculated. When the test variable W1 changes by Δ1, but the test variable W2 remains unchanged, a test effect relationship of the user account in a causal tree can be expressed as:

[0107] y1=f(W1+Δ1,W2)-f(W1,W2)=β1Δ1+β3W2Δ1=(Δ1,0,W2Δ1)β T .

[0108] Among them, β represents the matrix of parameter variables, and each element in the matrix represents a parameter variable.

[0109] After obtaining the values ​​of parameter variables β1, β2 and β3 through S501, they are substituted into the test effect relationship y1 to obtain the test effect of y1.

[0110] For example, taking two mutually influencing test variables applied to a user account as an example, the test effect of the user account in a causal tree is calculated. When the amount of change of the test variable W1 is Δ1 and the amount of change of the test variable W2 is Δ2, a test effect of the user account on a causal tree can be expressed as:

[0111] y3=β1Δ1+β2Δ2+β3[W1Δ2+W2Δ1+Δ1Δ2]=(Δ1,Δ2,W1Δ2+W2Δ1+Δ1Δ2)βT=A(Δ)β T .

[0112] Where A represents a constant matrix, and Δ represents the matrices of Δ1 and Δ2.

[0113] Similarly, after obtaining the values ​​of parameter variables β1, β2 and β3 through S501, they are substituted into the test effect relationship y3 to obtain the test effect of y3.

[0114] S503: Determine a weighted average test effect of each user account based on all test effects of each user account.

[0115] The test effect relationship corresponding to a user account in a causal tree is the test effect relationship corresponding to the leaf node where the user account is located. Each user account exists in all causal trees. Since the levels or classification methods of each causal tree are different, the test effect relationships corresponding to the leaf nodes of the user account on different causal trees are different, so the test effect relationships corresponding to the user accounts in different causal trees are also different. Calculating the weighted average of all test effects corresponding to the user account in all causal trees can make an overall and accurate estimate of the test effect of the user account, thereby obtaining a more accurate estimate and analysis result of the heterogeneous causal effect.

[0116] S402. Determine the standard error of the test effect of each user account based on the weighted average test effect of each user account and the value of the parameter variable of the causal forest.

[0117] Since the weighted average test effect is a method for calculating the average level of each test effect of a user account, the weighted average effect can also be expressed by the formula in the above embodiment, for example:

[0118]

[0119] in, represents the weighted average test effect of all test effects y1 of the user account, represents the mean value of the test variable W1, represents the mean value of the test variable W2.

[0120] or,

[0121]

[0122] In S402, the variance of the weighted average test effect of the user account may be determined first, and the variance of the weighted average test effect may be determined as the standard error of the test effect of the user account.

[0123] Exemplarily, the weighted average test effect is determined as The variance of is:

[0124]

[0125] Wherein, Var(β) represents the variance of the parameter variable matrix β, and the specific values ​​of the parameter variables in the matrix variance can be obtained through the above step S501.

[0126] Exemplarily, the weighted average test effect is determined as The variance of is:

[0127]

[0128] Since the above y1, y3 have linear relationships with β1, β2 and β3 respectively, it can be understood that: and They also have a linear relationship with β1, β2 and β3 respectively, and the specific values ​​of β1, β2 and β3 can be obtained according to the above step S501. Then, the specific values ​​of the linear relationship and parameter variables are substituted into the above variance formula to obtain the weighted average test effect of the user account. and The variance of .

[0129] Through the above process, each weighted average test effect of the user account and the test effect standard error of this weighted average test effect can be determined, and then the heterogeneous causal effect of the user account's response degree to the functional test task can be obtained.

[0130] When adopting corresponding strategies for user accounts based on heterogeneous causal effects in the future, the weighted average test effect of the user account can be compared with the preset effect threshold. If the difference between the two exceeds the standard error of the test effect of the user account, it means that the change between the weighted average test effect and the preset effect threshold is large. At this time, in the subsequent decision-making process, there is no need to consider the weighted average test effect. If the difference between the two does not exceed the standard error of the test effect of the user account, it means that the change between the weighted average test effect and the preset effect threshold meets the standard error requirements, and the impact of the changed test variables in the weighted average test effect on the user can be considered in the future.

[0131] For example, the weighted average test effects of user accounts are and The preset effect threshold is Y, according to and The standard error of the test effect for each user account is and In the subsequent decision-making process, if The difference between and Y exceeds The difference between and Y exceeds and The difference between and Y does not exceed Then when making decisions about user accounts, only consider The impact of changing test variables on user accounts.

[0132] The technical solution provided by the present disclosure can at least bring the following beneficial effects: after obtaining the standard error of the test effect of the user account, the weighted average test effect of the user account is compared with the preset effect threshold, and it is determined whether the difference between the weighted average test effect and the preset effect threshold exceeds the standard error of the test effect. If not, the test variable that changes in the weighted average test effect is determined as a functional test task that has a greater impact on the user. In this way, in the subsequent decision-making process for the user account, more or less functional content related to the functional test task can be provided to such user accounts, and the accuracy of the provided content can also be higher.

[0133] The above mainly introduces the solution provided by the embodiment of the present disclosure from the perspective of the method. In order to achieve the above functions, it includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should easily realize that, in combination with the method steps of each example described in the embodiment disclosed in this article, the present disclosure can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present disclosure.

[0134] Based on such understanding, the embodiments of the present disclosure also provide a device for determining heterogeneous causal effects, which can be applied to electronic devices. Figure 8 FIG. 1 is a schematic diagram of a device for determining heterogeneous causal effects according to an exemplary embodiment of the present disclosure. Figure 8 As shown, the device may include: a causal forest generation unit 601, a test effect relationship acquisition unit 602 and a heterogeneous causal effect acquisition unit 603.

[0135] The causal forest generation unit 601 is used to generate a causal forest including at least one causal tree based on the behavioral data of multiple user accounts; each causal tree includes at least one leaf node, a leaf node corresponds to a category of a user account, and the leaf node includes user accounts belonging to the corresponding category among the multiple user accounts.

[0136] The test effect relationship acquisition unit 602 is used to determine the test effect relationship set corresponding to each causal tree and including the test effect relationship of each leaf node of the causal tree according to a preset function; the preset function includes one or more test variables corresponding to one or more functional test tasks; the test effect relationship of the leaf node is used to represent the nonlinear superposition relationship between the degree to which multiple user accounts on the leaf node are respectively affected by all the test variables and all the test variables corresponding to the leaf node;

[0137] The heterogeneous causal effect acquisition unit 603 is used to determine the heterogeneous causal effect of each user account according to the test effect relationship set corresponding to all causal trees in the causal forest; the heterogeneous causal effect of each user account is used to represent the response degree of each user account to one or more functional test tasks.

[0138] Optionally, the causal forest generation unit 601 is specifically used to: obtain behavioral data and preset causal parameters of multiple user accounts; the preset causal parameters are used to represent the number of causal trees in the causal forest and the hierarchical depth of the multiple causal trees; according to the behavioral data and preset causal parameters of multiple user accounts, for each causal tree, classify and process multiple user accounts, and save user accounts belonging to the same category in corresponding leaf nodes.

[0139] Optionally, the test effect relationship acquisition unit 602 is specifically used to: construct a preset function based on one or more test variables corresponding to one or more functional test tasks and the parameter variables of the causal forest; for each leaf node in each causal tree, determine the change relationship corresponding to the preset function when one or more test variables change, and determine all change relationships of the preset function as the test effect relationship of the leaf node.

[0140] Optionally, the heterogeneous causal effect acquisition unit 603 is specifically used to: determine the weighted average test effect of each user account based on the test effect relationship set corresponding to all causal trees in the causal forest; determine the standard error of the test effect of each user account based on the weighted average test effect of each user account and the value of the parameter variable of the causal forest.

[0141] Optionally, the heterogeneous causal effect acquisition unit 603 is specifically used to: input one or more test variables of one or more functional testing tasks into a preset algorithm model to obtain the values ​​of parameter variables of the causal forest output by the preset algorithm model; determine the test effects of each user account in different causal trees according to the values ​​of the parameter variables and the test effect relationship set corresponding to each causal tree; and determine the weighted average test effect of each user account according to all the test effects of each user account.

[0142] 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.

[0143] Fig. 9 1 is a schematic diagram of the structure of a server 110 provided by the present disclosure. Fig. 9As shown, the server 110 may include at least one processor 1101 and a memory 1102 for storing executable instructions of the processor 1101. The processor 1101 is configured to execute the instructions in the memory 1102 to implement the method for determining the heterogeneous causal effect in the above embodiment.

[0144] In addition, the server 110 may further include a communication bus 1103 and at least one communication interface 1104 .

[0145] The processor 1101 may be a central processing unit (CPU), a microprocessing unit, an ASIC, or one or more integrated circuits for controlling the execution of the program of the disclosed solution.

[0146] The communication bus 1103 may include a pathway to transmit information between the above-mentioned components.

[0147] The communication interface 1104 uses any transceiver or other device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0148] The memory 1102 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may exist independently and be connected to the processing unit through a bus. The memory may also be integrated with the processing unit.

[0149] The memory 1102 is used to store instructions for executing the solution of the present disclosure, and the execution is controlled by the processor 1101. The processor 1101 is used to execute the instructions stored in the memory 1102, so as to realize the functions in the method of the present disclosure.

[0150] As an example, combining Figure 8 The functions implemented by the causal forest generation unit 601, the test effect relationship acquisition unit 602 and the heterogeneous causal effect acquisition unit 603 in the heterogeneous causal effect determination device are similar to those Fig. 9 The function of processor 1101 in is the same.

[0151] In a specific implementation, as an embodiment, the processor 1101 may include one or more CPUs, such as Fig. 9 CPU0 and CPU1 in.

[0152] In a specific implementation, as an embodiment, the server 110 may include multiple processors, such as Fig. 9 1 and 1105. Each of these processors may be a single-CPU processor or a multi-CPU processor. A processor herein may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0153] In a specific implementation, as an embodiment, the server 110 may further include an output device 1106 and an input device 1107. The output device 1106 communicates with the processor 1101 and may display information in a variety of ways. For example, the output device 1106 may be a liquid crystal display (LCD), a light emitting diode (LED) display device, a cathode ray tube (CRT) display device, or a projector. The input device 1107 communicates with the processor 1101 and may receive user input in a variety of ways. For example, the input device 1107 may be a mouse, a keyboard, a touch screen device, or a sensor device.

[0154] Those skilled in the art will understand that Fig. 9 The structure shown in the figure does not constitute a limitation on the server 110, and the server 110 may include more or less components than those shown in the figure, or combine some components, or adopt a different component arrangement.

[0155] In addition, a computer-readable storage medium including computer instructions is also provided in the embodiments of the present disclosure. When the instructions in the computer-readable storage medium are executed by a processor of an electronic device such as the above-mentioned server, the electronic device can perform the method for determining the heterogeneous causal effect provided in the above-mentioned embodiments. Optionally, the 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.

[0156] In addition, a computer program product is also provided in an embodiment of the present disclosure, including computer instructions. When the computer instructions are executed on an electronic device such as the above-mentioned server, the electronic device executes a method for determining heterogeneous causal effects as provided in the above-mentioned embodiment.

[0157] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure 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 description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0158] 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 method for determining heterogeneous causal effects, characterized in that: include: generating a causal forest including at least one causal tree according to the behavioral data of the plurality of user accounts; Each of the causal trees includes at least one leaf node, one leaf node corresponds to a category of a user account, and the leaf node includes user accounts belonging to the corresponding category among the multiple user accounts; According to a preset function, a test effect relationship set corresponding to each of the causal trees, including the test effect relationship of each leaf node of the causal tree, is determined; the preset function includes one or more test variables corresponding to one or more functional test tasks; the test effect relationship of the leaf node is used to represent the nonlinear superposition relationship between the degree to which the multiple user accounts on the leaf node are respectively affected by all the test variables and all the test variables corresponding to the leaf node; The heterogeneous causal effect of each user account is determined based on the test effect relationship set corresponding to all causal trees in the causal forest; the heterogeneous causal effect of each user account is used to represent the response degree of each user account to the one or more functional test tasks.

2. The method for determining heterogeneous causal effects according to claim 1, characterized in that: The step of generating a causal forest including at least one causal tree according to the behavior data of the plurality of user accounts comprises: Obtaining behavioral data of multiple user accounts and preset causal parameters; the preset causal parameters are used to represent the number of the causal trees in the causal forest and the hierarchical depth of the multiple causal trees; According to the behavior data of the multiple user accounts and the preset causal parameters, for each causal tree, the multiple user accounts are classified and processed, and user accounts belonging to the same category are stored in the corresponding leaf nodes.

3. The method for determining heterogeneous causal effects according to claim 1, characterized in that: The step of determining, according to a preset function, a test effect relationship set corresponding to each causal tree and including a test effect relationship of each leaf node of the causal tree comprises: Constructing the preset function according to the one or more test variables corresponding to the one or more functional test tasks and the parameter variables of the causal forest; For each leaf node in each of the causal trees, determine the change relationship corresponding to the preset function when the one or more test variables change, and determine all the change relationships of the preset function as the test effect relationship of the leaf node.

4. The method for determining heterogeneous causal effects according to claim 1, characterized in that: The heterogeneous causal effect includes the weighted average test effect and the standard error of the test effect of each user account, and the standard error of the test effect is used to screen the weighted average test effect.

5. The method for determining heterogeneous causal effects according to claim 4, characterized in that: The determining of the heterogeneous causal effect of each user account according to the test effect relationship set corresponding to all causal trees in the causal forest includes: Determine the weighted average test effect of each user account according to the test effect relationship set corresponding to all the causal trees in the causal forest; The standard error of the test effect of each user account is determined based on the weighted average test effect of each user account and the value of the parameter variable of the causal forest.

6. The method for determining heterogeneous causal effects according to claim 5, characterized in that: The determining the weighted average test effect of each user account according to the test effect relationship set corresponding to all the causal trees in the causal forest includes: Inputting the one or more test variables of the one or more functional test tasks into a preset algorithm model to obtain the values ​​of the parameter variables of the causal forest output by the preset algorithm model; Determine the test effect of each user account in different causal trees according to the value of the parameter variable and the test effect relationship set corresponding to each causal tree; The weighted average test effect of each user account is determined based on all the test effects of each user account.

7. A device for determining heterogeneous causal effects, characterized in that: It includes a causal forest generation unit, a test effect relationship acquisition unit and a heterogeneous causal effect acquisition unit; The causal forest generation unit is used to generate a causal forest including at least one causal tree according to the behavior data of multiple user accounts; each of the causal trees includes at least one leaf node, one leaf node corresponds to a category of a user account, and the leaf node includes a user account belonging to a corresponding category among the multiple user accounts; The test effect relationship acquisition unit is used to determine the test effect relationship set corresponding to each of the causal trees, including the test effect relationship of each leaf node of the causal tree, according to a preset function; the preset function includes one or more test variables corresponding to one or more functional test tasks; the test effect relationship of the leaf node is used to represent the nonlinear superposition relationship between the degree to which the multiple user accounts on the leaf node are respectively affected by all the test variables and all the test variables corresponding to the leaf node; The heterogeneous causal effect acquisition unit is used to determine the heterogeneous causal effect of each user account based on the test effect relationship set corresponding to all causal trees in the causal forest; the heterogeneous causal effect of each user account is used to represent the response degree of each user account to the one or more functional test tasks.

8. The device for determining heterogeneous causal effects according to claim 7, characterized in that: The causal forest generation unit is specifically used for: Obtaining behavioral data of multiple user accounts and preset causal parameters; the preset causal parameters are used to represent the number of the causal trees in the causal forest and the hierarchical depth of the multiple causal trees; According to the behavior data of the multiple user accounts and the preset causal parameters, for each causal tree, the multiple user accounts are classified and processed, and user accounts belonging to the same category are stored in the corresponding leaf nodes.

9. The device for determining heterogeneous causal effects according to claim 7, characterized in that: The test effect relationship acquisition unit is specifically used for: Constructing the preset function according to the one or more test variables corresponding to the one or more functional test tasks and the parameter variables of the causal forest; For each leaf node in each of the causal trees, determine the change relationship corresponding to the preset function when the one or more test variables change, and determine all the change relationships of the preset function as the test effect relationship of the leaf node.

10. The device for determining heterogeneous causal effects according to claim 7, characterized in that: The heterogeneous causal effect includes the weighted average test effect and the standard error of the test effect of each user account, and the standard error of the test effect is used to screen the weighted average test effect.

11. The device for determining heterogeneous causal effects according to claim 10, characterized in that: The heterogeneous causal effect acquisition unit is specifically used for: Determine the weighted average test effect of each user account according to the test effect relationship set corresponding to all the causal trees in the causal forest; The standard error of the test effect of each user account is determined based on the weighted average test effect of each user account and the value of the parameter variable of the causal forest.

12. The device for determining heterogeneous causal effects according to claim 11, characterized in that: The heterogeneous causal effect acquisition unit is specifically used for: Inputting the one or more test variables of the one or more functional test tasks into a preset algorithm model to obtain the values ​​of the parameter variables of the causal forest output by the preset algorithm model; Determine the test effect of each user account in different causal trees according to the value of the parameter variable and the test effect relationship set corresponding to each causal tree; The weighted average test effect of each user account is determined based on all the test effects of each user account.

13. An electronic device, characterized in that: include: processor; A memory for storing the processor executable instructions; wherein the processor is configured to execute the instructions to implement the method for determining the heterogeneous causal effect as described in any one of claims 1-6.

14. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method for determining the heterogeneous causal effect as described in any one of claims 1 to 6.

15. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed on an electronic device, the electronic device executes the method for determining heterogeneous causal effects according to any one of claims 1 to 6.

Images