Method, device, equipment and storage medium for determining causal influence

Abstract

The present application discloses a method, device, equipment and storage medium for determining causal influence, which belongs to the field of computer and Internet technology. The method comprises: obtaining at least one set of user information corresponding to the external characteristics of the first user from the user information corresponding to at least one user account logged in in the application; determining the probability of occurrence of each type of user intrinsic attribute based on the user intrinsic attributes included in at least one set of user information; determining the mathematical expectation of user activity when each type of user intrinsic attribute is given based on the user intrinsic attributes and user activity included in at least one set of user information; taking the probability of occurrence of each type of user intrinsic attribute as the weight, performing weighted summation processing on the mathematical expectation of user activity when each type of user intrinsic attribute is given, and obtaining a first causal effect estimate. The embodiment of the present application improves the accuracy of user activity correlation analysis, which helps to effectively improve user activity.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer and Internet technology, and in particular, to a method, device, equipment and storage medium for determining causal influence. Background Art

[0002] Watching live broadcasts and / or short videos has become a common form of entertainment and leisure in people's daily lives, and various live broadcast and / or short video applications have emerged one after another. For developers and operators of live broadcast and / or short video applications, user activity is a very important analysis indicator.

[0003] User activity refers to the degree of activity of users in live broadcast and / or short video applications. There are many factors that affect user activity, such as the type of terminal used by the user, the operating system equipped with the live broadcast and / or short video application, the age of the user, the occupation of the user, etc. Among these influencing factors, the influencing factors that are relatively easy to collect, such as the type of terminal used by the user and the operating system equipped with the live broadcast and / or short video application, are the influencing factors that the developers and operators of the live broadcast and / or short video application pay special attention to in the process of analyzing user activity. Here, we call these influencing factors that are relatively easy to collect user external features. In the related art, the analysis method for user activity mainly includes: estimating the impact of user external features on user activity by estimating the correlation between user external features and user activity. For example, if the user activity corresponding to the operating system A carried is higher than the user activity corresponding to the operating system B carried, it can be analyzed that the user experience of operating system A is better than the user experience of operating system B.

[0004] However, there is a possibility that the user's intrinsic attributes, such as the user's age and occupation, not only affect the user's external features, but also affect the user's activity. Therefore, according to the Bayesian network, in the presence of the user's intrinsic attributes, the user's external features may be correlated with the user's activity, but not necessarily causally related. Therefore, the relevant technology estimates the impact of the user's external features on the user's activity based only on the correlation between the user's external features and the user's activity, which is not accurate enough. Summary of the invention

[0005] The embodiments of the present application provide a method, device, equipment and storage medium for determining causal influence, which can be used to improve the accuracy of user activity correlation analysis. The technical solution is as follows:

[0006] On the one hand, an embodiment of the present application provides a method for determining causal influence, the method comprising:

[0007] Acquire at least one set of user information corresponding to the first user's external features from the user information corresponding to at least one user account logged in in the application, each set of user information including a user's internal attributes, user's external features, and user activity corresponding to a user account;

[0008] Determining the occurrence probability of each type of user intrinsic attribute based on the user intrinsic attributes included in the at least one set of user information;

[0009] Based on the user intrinsic attributes and user activity included in the at least one set of user information, determining the mathematical expectation of the user activity when each type of user intrinsic attribute is given;

[0010] The occurrence probability of each type of user intrinsic attribute is used as a weight, and a weighted summation process is performed on the mathematical expectation of the user activity when each type of user intrinsic attribute is given to obtain a first causal effect estimator, where the first causal effect estimator is used to indicate the causal influence of the first user extrinsic feature on the user activity.

[0011] On the other hand, an embodiment of the present application provides a device for determining causal influence, the device comprising:

[0012] An information acquisition module, configured to acquire at least one set of user information corresponding to the first user's external features from the user information corresponding to at least one user account logged in in the application, wherein each set of user information includes a user's internal attributes, user's external features, and user activity corresponding to a user account;

[0013] A probability determination module, configured to determine the occurrence probability of each type of user intrinsic attribute based on the user intrinsic attributes included in the at least one set of user information;

[0014] an expectation determination module, configured to determine the mathematical expectation of the user activity when each type of user intrinsic attribute is given, based on the user intrinsic attributes and user activity included in the at least one set of user information;

[0015] The estimation calculation module is used to perform weighted summation processing on the mathematical expectations of user activity when each type of user intrinsic attribute is given, using the occurrence probability of each type of user intrinsic attribute as a weight, to obtain a first causal effect estimation, wherein the first causal effect estimation is used to indicate the causal influence of the first user extrinsic feature on the user activity.

[0016] On the other hand, an embodiment of the present application provides a computer device, which includes a processor and a memory, wherein the memory stores a computer program, and the computer program is loaded and executed by the processor to implement the method for determining causal influence as described above.

[0017] On the other hand, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for determining causal influence as described above is implemented.

[0018] On the other hand, an embodiment of the present application provides a computer program product, which, when executed on a computer device, enables the computer device to execute the method for determining causal influence as described above.

[0019] The technical solution provided in the embodiments of the present application can bring the following beneficial effects:

[0020] By taking into account the correlation introduced by user intrinsic attributes between user extrinsic features and user activity when estimating the impact of user extrinsic features on user activity, and marginalizing user intrinsic attributes, we can estimate the causal impact of user extrinsic features on user activity, avoid misjudging the correlation between user extrinsic features and user activity as the impact of user extrinsic features on user activity, and improve the accuracy of user activity-related analysis, which will help to accurately operate applications and guide user operations based on the analysis results in the future, so as to effectively improve user activity. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 is a schematic diagram of a Bayesian network provided by an embodiment of the present application;

[0023] Figure 2 is a schematic diagram of the network structure of a Bayesian network provided by an embodiment of the present application;

[0024] Figure 3 is a schematic diagram of an analysis system provided by an embodiment of the present application;

[0025] Figure 4 is a flow chart of a method for determining causal influence provided by an embodiment of the present application;

[0026] Figure 5 is a schematic diagram of the network structure of a Bayesian network provided by another embodiment of the present application;

[0027] Figure 6 is a block diagram of a device for determining causal influence provided by one embodiment of the present application;

[0028] Figure 7 It is a block diagram of a device for determining causal influence provided by another embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0030] Before introducing and explaining the embodiments of the present application, some terms and related technologies involved in the embodiments of the present application are first introduced and explained.

[0031] 1. Bayesian Network (BN)

[0032] The Bayesian network consists of the following two parts:

[0033] 1. A network structure

[0034] That is, an acyclic directed graph Where V is a set of nodes, V = {v1,…,v N}; A is a set of edges, each edge can be represented by the nodes corresponding to its two ends, that is, a ij =(v i ,v j ). One thing that needs to be explained is that for a directed graph, the edges are directed (i.e., they have (v i ,v j )≠(v j ,v i ), (v i ,v j ) indicates v i →v j In addition, since the directed graph has no cycles, it is impossible to return to the node from a node along the direction of the arrow.

[0035] 2. A probability distribution about the feature variable X, with parameter Θ

[0036] Where X={X1,X2,…,X n}, represents the set of all feature variables. The purpose of establishing a Bayesian network is to express a conditional independence relationship between feature variables through its network structure, so as to achieve the purpose of decomposing the original probability distribution. Among them, each node in the acyclic directed graph corresponds to a variable X in the probability distribution. i .

[0037] The decomposition formula is:

[0038] in, Represents an acyclic directed graph Xi Therefore, according to the probability distribution after the decomposition of the acyclic directed graph, any variable is only related to the variable represented by its parent node.

[0039] Please refer to Figure 1 , which shows a schematic diagram of a Bayesian network provided by an embodiment of the present application. Figure 1 As shown in (a), it is a schematic diagram of an acyclic directed graph provided in an embodiment of the present application, which has seven nodes. Figure 1 (b) is shown, which is Figure 1 (a) shows the conditional independence information corresponding to the acyclic directed graph. G represents the independence condition implied by the structure of the acyclic directed graph. For example, A⊥ G D|C can be interpreted as: C is the parent node of D. When C is given, D and C are conditionally independent. Optionally, judge ⊥ G The existence of can be based on D-Separation theory. Figure 1 (c) is shown, which is Figure 1 (b) shows the probability distribution of conditional independence information. For example, when A⊥ G When D|C, A⊥ P D|C, that is, given C, the probability distributions of random variables corresponding to A and D are independent. Optionally, each conditional independence information implied by the acyclic directed graph can derive a corresponding probability distribution conditional independence information.

[0040] One thing that needs to be explained is that in the Bayesian network, Figure 1 As shown in (a), causality can propagate in the direction of the arrow, but correlation can propagate in the direction of the arrow or in the opposite direction. Therefore, if there is a correlation between a variable and another variable, it does not necessarily mean that the variable has a causal effect (or causal influence) on the other variable.

[0041] For example, suppose there is a two-variable Bayesian network, where the two variables represent "city altitude" and "city average temperature" respectively. From a physical point of view, the higher the city altitude, the lower the city average temperature. Therefore, "city altitude" is the cause and "city average temperature" is the effect. The direction of the arrow is from "city altitude" to "city average temperature". Assuming that the "city average temperature" is known to be relatively low, it can be inferred that the "city altitude" may be relatively high. This shows that there is a correlation between "city average temperature" and "city altitude", and it propagates in the opposite direction of the arrow. However, if the city's average temperature is increased in some way, its altitude will not decrease as a result. This shows that there is no causal effect in the opposite direction of the arrow.

[0042] 2. User intrinsic attributes, user external characteristics and user activity

[0043] User intrinsic attributes are variables that reflect the intrinsic characteristics of users, which are usually difficult to obtain directly. In one example, user intrinsic attributes include at least one of the following: user age, user occupation, and user economic income. Since user intrinsic attributes are usually difficult to obtain directly, users can be encouraged to actively fill in questionnaires or other incentives to obtain user intrinsic attributes.

[0044] Compared with the user's intrinsic attributes, the user's external characteristics are usually easier to obtain. Optionally, the user's external characteristics can be obtained from the background data of the application. Optionally, the user's external characteristics include at least one of the following: the type of terminal used by the user, the operating system on which the application is installed, and the method of registering the application (such as telephone registration, email registration, cooperation channel registration, etc.).

[0045] User activity is used to describe the user's activity level in the application. Optionally, the user activity can also be obtained from the background data of the application. The embodiment of the present application does not limit the specific content of the user activity. Optionally, the user activity includes at least one of the following: the frequency of daily use of the application, the duration of daily use of the application, and the number of times the application was logged in in the past N (N is a positive number) days.

[0046] Based on this, we can construct Figure 2 The network structure of the Bayesian network is shown in Figure 2. Figure 2 As shown in the figure, the parent node of user external features is the user's intrinsic attributes; the parent node of user activity is the user's intrinsic attributes and user external features. Therefore, there are two paths from user external features to user activity:

[0047] 1. User external characteristics → user activity;

[0048] 2. User external characteristics → user internal attributes → user activity.

[0049] In the second path above, since the arrow direction of the Bayesian network is from the user's intrinsic attributes to the user's extrinsic features, in the second path, the user's extrinsic features and user activity are not "causal", but can be "correlated". It is precisely because of the existence of the user's intrinsic attributes that the correlation is introduced between the user's extrinsic features and user activity.

[0050] Since the intrinsic attributes of users are uncontrollable, the external characteristics of users can be intervened and modified to a certain extent. Therefore, if we want to improve user activity, the external characteristics of users are an important breakthrough. Based on this, it is necessary to analyze the impact of external characteristics of users on user activity, so as to operate and guide them in a targeted manner to improve user activity. Figure 2 The Bayesian network shown in the figure introduces a correlation between user external features and user activity due to the existence of user intrinsic attributes. In statistics, user intrinsic attributes can be called a confounding variable between user external features and user activity. Then, considering user intrinsic attributes, how to determine the causal effect of user external features on user activity in order to improve the accuracy of user activity correlation analysis is a technical problem that needs to be solved urgently.

[0051] Based on this, the embodiment of the present application provides a method for determining causal influence, which can be used to solve the above technical problems. Below, the method for determining causal influence provided by the present application is introduced and explained in combination with several embodiments.

[0052] Please refer to Figure 3 , which shows a schematic diagram of an analysis system provided by an embodiment of the present application. Figure 3 As shown, the analysis system includes: a computer device 310 and a server 320.

[0053] Server 320 refers to the background server of the application, which provides background services for the application. The embodiment of the present application does not limit the type of application. In one example, the application includes a live broadcast application, a video playback application, a short video application, a social application, a shopping application, a food application, a travel application, a payment application, a game application, and a music playback application. Optionally, server 320 can be a single server, a server cluster consisting of multiple servers, or a cloud computing center.

[0054] In one example, if Figure 3 As shown, the server 320 can obtain the background data of the application from the device 330. Optionally, the background data of the application includes at least one of the following: user activity, user external characteristics, and user internal attributes. Among them, the device 330 refers to a device on which the application is installed and running, such as a mobile phone, a tablet computer, a game console, an e-book reader, a multimedia playback device, a wearable device, a PC (Personal Computer), and other electronic devices. After the server 320 obtains the background data of the application, it can store the background data of the application in a database.

[0055] Computer device 310 refers to a device with data analysis capabilities. In the embodiment of the present application, computer device 310 can obtain the background data of the application from server 320, and use the background data of the application to perform data analysis, such as analyzing the causal influence of the user's external characteristics on the user's activity. Optionally, when the background data of the application includes the user's intrinsic attributes, that is, when the user fills in the information related to the user's intrinsic attributes through device 330, computer device 310 obtains the user's intrinsic attributes from server 320; when the background data of the application does not include the user's intrinsic attributes, computer device 310 can directly obtain the user's intrinsic attributes, such as manually entering the information related to the user's intrinsic attributes through relevant personnel.

[0056] Optionally, the computer device 310 and the server 320, and the server 320 and the device 330 may communicate with each other via a network, which may be a wired network or a wireless network.

[0057] One thing that needs to be explained is that in actual application, the computer device 310 and the server 320 can be implemented as one device, that is, the computer device 310 can also serve as the background server of the application program, or the server 320 can also have the ability of data analysis. Figure 3 The embodiment is only for the convenience of description, and the computer device 310 and the server 320 are implemented as two devices, but this should not cause any limitation to the technical solution provided in the embodiment of the present application.

[0058] Please refer to Figure 4 , which shows a flow chart of a method for determining causal influence provided by an embodiment of the present application. The method can be applied to the above Figure 3 In the computer device 310 described in the embodiment, the method may include the following steps.

[0059] Step 410, obtaining at least one set of user information corresponding to the first user's external characteristics from the user information corresponding to at least one user account logged in in the application, each set of user information including user intrinsic attributes, user external characteristics and user activity corresponding to one user account.

[0060] The user information corresponding to the user account includes: user intrinsic attributes, user extrinsic features, and user activity. In one example, the causal relationship between user intrinsic attributes, user extrinsic features, and user activity is as follows: user intrinsic attributes have a causal impact on user extrinsic features; user intrinsic attributes have a causal impact on user activity; user extrinsic features have a causal impact on user activity. That is, the causal relationship between user intrinsic attributes, user extrinsic features, and user activity can refer to the above Figure 2The network structure of the Bayesian network shown in FIG. For the relevant definitions and acquisition methods of user intrinsic attributes, user external features, and user activity, please refer to the above embodiments, which will not be elaborated here.

[0061] In the embodiment of the present application, the user information corresponding to the user account can be organized into structured data as described in Table 1 below, so that a corresponding relationship can be established between the user information, so as to facilitate obtaining the required data based on a certain dimension of the user information. For example, the computer device obtains at least one set of user information corresponding to the first external feature based on the user external feature in the user information.

[0062] Table 1 User information corresponding to user account

[0063] User Account User intrinsic attribute X User extrinsic features T User activity Y User Account 1 <![CDATA[X1]]> <![CDATA[T1]]> <![CDATA[Y1]]> User Account 2 <![CDATA[X2]]> <![CDATA[T2]]> <![CDATA[Y2]]> …… …… …… ……

[0064] Step 420: Determine the occurrence probability of each type of user intrinsic attribute based on the user intrinsic attributes included in at least one set of user information.

[0065] Based on at least one set of user information obtained in the above step 410, the user external features included in each set of user information are all first user external features, but the user intrinsic attributes and user activity included in each set of user information may be different, or may be partially the same and partially different. Therefore, the computer device can determine the probability of occurrence of each type of user intrinsic attribute based on the user intrinsic attributes included in at least one set of user information. Optionally, the user intrinsic attributes are divided into multiple categories after discretization processing, which can reduce the possible values ​​of the user intrinsic attributes and reduce the processing and analysis overhead of the computer device. For the discretization processing method of the user intrinsic attributes, please refer to the following method embodiment, which will not be repeated here.

[0066] For example, at least one set of user information includes u user intrinsic attributes, where u is a positive integer. Assuming that the u user intrinsic attributes can be discretized into multiple categories of user intrinsic attributes, and one category of user intrinsic attributes has v, where v is a positive integer, then the occurrence probability of the category of user intrinsic attributes is p(x)=v / u.

[0067] Step 430: Determine the mathematical expectation of the user activity when each type of user intrinsic attribute is given, based on the user intrinsic attributes and user activity included in at least one set of user information.

[0068] In addition to determining the occurrence probability of each type of user intrinsic attribute through the above step 420, the computer device can also determine the mathematical expectation of user activity when each type of user intrinsic attribute is given based on the user intrinsic attribute and user activity included in at least one set of user information. Assuming that the first user's external feature is t=1, the user's intrinsic attribute is x, and the user's activity is y, then when each type of user intrinsic attribute is given, the mathematical expectation of user activity is E(y|t=1,x).

[0069] Step 440, taking the occurrence probability of each type of user intrinsic attribute as a weight, weighted summing is performed on the mathematical expectation of the user activity when each type of user intrinsic attribute is given, to obtain a first causal effect estimator, which is used to indicate the causal influence of the first user's external characteristics on the user activity.

[0070] After determining the occurrence probability of each type of user intrinsic attribute and the mathematical expectation of user activity when each type of user intrinsic attribute is given, the computer device can use the occurrence probability of each type of user intrinsic attribute as a weight to perform weighted summation processing on the mathematical expectation of user activity when each type of user intrinsic attribute is given, so as to obtain a first causal effect estimator. Assuming that the first user extrinsic feature is t=1, the user intrinsic attribute is x, and the user activity is y, and the mathematical expectation of user activity when each type of user intrinsic attribute is given is E(y|t=1,x), then the first causal effect estimator is: E(y|do(t=1))=∑p(x)E(y|t=1,x). Among them, do(t) indicates that the user intrinsic attribute is forcibly removed. Since in the process of calculating the first causal effect estimator, the user extrinsic feature is based on the first user extrinsic feature, the first causal effect estimator is used to indicate the causal influence of the first user extrinsic feature on the user activity.

[0071] From the above-mentioned determination process of the first causal effect estimator, it can be seen that the calculation of the causal effect estimator can be obtained by the mathematical formula E(y|do(t))=∑p(x)E(y|t,x). The derivation process of the mathematical formula is introduced below.

[0072] The solution to be implemented in the embodiment of the present application is to estimate the causal effect (causal influence) of the user's external feature T on the user's activity Y in the presence of the user's internal attribute X. Therefore, it is necessary to assume that the user's external feature T is a mandatory setting, rather than a variable associated with the user's internal attribute X. From the perspective of the Bayesian network, Figure 5 As shown, it is necessary to cut off the connection between the user's intrinsic attribute X and the user's extrinsic feature T.

[0073] Without cutting off the connection between the user's intrinsic attribute X and the user's extrinsic feature T, the following expression can be obtained:

[0074] p(Y=y,T=t,X=x)=p(X=x)p(T=t|X=x)p(Y=y|T=t,X=x).

[0075] For ease of description, in the embodiments of the present application, p(y,t,x)=p(x)p(t|x)p(y|t,x) is used to replace the above expression.

[0076] When the connection between the user's intrinsic attribute X and the user's extrinsic feature T is cut off, the expression can be obtained: p(y,x|do(t))=p(x)p(y|t,x). That is, cutting off the connection between the user's intrinsic attribute X and the user's extrinsic feature T is equivalent to removing the term p(t|x) in the expression. In the embodiment of the present application, do(t) means that the confusion factor of the user's intrinsic attribute is forcibly removed.

[0077] Since the solution to be implemented in the embodiment of the present application is to determine the relationship between the user activity Y and the user's external feature T, after marginalizing the user's intrinsic attribute X, the expression can be obtained: p(y|do(t))=∑p(x)p(y|t,x). If what needs to be determined is the mathematical expectation of the user's activity Y, the expression can be obtained: E(y|do(t))=∑p(x)E(y|t,x). Among them, E(y|do(t)) is the mathematical expectation of the user's activity Y when the user's external feature T=t, which is used to indicate the causal influence of the user's external feature T=t on the user's activity Y.

[0078] To sum up, the technical solution provided by the embodiment of the present application takes into account the correlation introduced by the user's intrinsic attributes between the user's extrinsic features and the user's activity when estimating the impact of the user's extrinsic features on the user's activity, and marginalizes the user's intrinsic attributes, thereby estimating the causal impact of the user's extrinsic features on the user's activity, avoiding the erroneous judgment of the correlation between the user's extrinsic features and the user's activity as the impact of the user's extrinsic features on the user's activity, improving the accuracy of the user activity correlation analysis, and facilitating the subsequent precise operation of the application and guiding the user's operation based on the analysis results, so as to effectively improve the user's activity.

[0079] From the above description, it can be seen that the discretization processing of the user's intrinsic attributes can reduce the possible values ​​of the user's intrinsic attributes and reduce the processing and analysis overhead of the computer device. Next, a discretization processing method provided by an embodiment of the present application is introduced.

[0080] In one example, the user intrinsic attribute is expressed in a discrete numerical form; the above method also includes: obtaining the user intrinsic attribute corresponding to at least one user account; discretizing the user intrinsic attribute corresponding to at least one user account to obtain the user intrinsic attribute expressed in the form of a discrete numerical value.

[0081] The computer device may discretize the intrinsic attributes corresponding to at least one user account before the above step 410, or may discretize the intrinsic attributes corresponding to at least one user account after the above step 410, and the embodiments of the present application are not limited to this. Optionally, the user's intrinsic attributes include attributes of at least one dimension, and when the computer device discretizes the user's intrinsic attributes, it may discretize the attributes of all dimensions included in the user's intrinsic attributes, or may discretize the attributes of some dimensions included in the user's intrinsic attributes. Optionally, the user's intrinsic attributes include attributes of at least one of the following dimensions: the user's age corresponding to the user account, the user's economic income corresponding to the user account, and the user's occupation corresponding to the user account.

[0082] Taking the discretization processing of attributes of all dimensions included in at least one user account by a computer device as an example, the above discretization processing of the user intrinsic attributes corresponding to the at least one user account to obtain the user intrinsic attributes expressed in the form of discretized numerical values ​​includes: obtaining quantile information corresponding to the attributes of at least one dimension; determining the discretization intervals corresponding to the attributes of at least one dimension and the discretization numerical values ​​corresponding to each discretization interval based on the quantile information corresponding to the attributes of at least one dimension; discretizing the user intrinsic attributes corresponding to the at least one user account according to the discretization intervals corresponding to the attributes of at least one dimension and the discretization numerical values ​​corresponding to each discretization interval to obtain the user intrinsic attributes expressed in the form of discretized numerical values.

[0083] The quantile information corresponding to the attribute of each dimension includes at least one quantile corresponding to the attribute of the dimension. According to the at least one quantile, the attribute of the dimension can be discretized into several intervals. For example, the quantile information corresponding to the attribute of a certain dimension includes three quantiles corresponding to the attribute of the dimension. According to the three quantiles, the attribute of the dimension can be discretized into four intervals. By discretizing the attributes of each dimension included in the user's intrinsic attributes, the user's intrinsic attributes can be represented as a multidimensional tuple composed of multiple discretized values.

[0084] For example, assume that the user's intrinsic attributes include the attributes of the two dimensions of the user's economic income (E) corresponding to the user account and the user's age (A) corresponding to the user account, and the attributes of these two dimensions are discretized into four discrete values ​​according to the three quantiles (0.25, 0.5, 0.75). For the attribute of the dimension of the user's age (A) of the user account, assume that the three quantiles of the user's age corresponding to the user account are analyzed from the user's intrinsic attributes corresponding to at least one user account, which include the user's age corresponding to the user account: The user age corresponding to the user account represented in discretized numerical form is Possible values ​​are 1, 2, 3, 4. 1 means 2 means 3 means 4 means Similarly, the attribute of the user economic income (E) dimension of the user account can also be discretized as Therefore, the user intrinsic attribute (X) can be expressed as: That is, the user intrinsic attribute (X) is a discretized variable with 16 possible values, each of which represents a pair of user ages and user economic income The possible values ​​of .

[0085] In summary, the technical solution provided in the embodiment of the present application discretizes the user's intrinsic attributes according to the quantile information, thereby limiting the values ​​of the user's intrinsic attributes that were originally difficult to control and predict to a limited possible range, thereby reducing the possible values ​​of the user's intrinsic attributes. This, on the one hand, facilitates standardized storage, and on the other hand, helps to reduce the processing and analysis overhead of computer devices when performing user activity-related analysis.

[0086] From the description of the above embodiments, it can be seen that the causal influence of user external features on user activity is helpful for accurately operating applications and guiding user operations according to the analysis results. Below, an application method of the causal influence of user external features on user activity provided by an embodiment of the present application is introduced.

[0087] In one example, the method further includes: determining at least one causal effect estimator, each causal effect estimator being used to indicate the causal influence of a user external feature on user activity; and based on at least one causal effect estimator, analyzing the difference between the causal influences of at least one user external feature on user activity.

[0088] In the embodiment of the present application, the difference between the causal effects of multiple user extrinsic features on user activity can be expressed by the difference in the mathematical expectation of user activity when multiple user extrinsic features are given. For example, assuming that the first user extrinsic feature is t=1, the second user extrinsic feature is t=2, the user intrinsic attribute is x, and the user activity is y, then the first causal effect estimator is E(y|do(t=1))=∑p(x)E(y|t=1,x), and the second causal effect estimator is E(y|do(t=2))=∑p(x)E(y|t=2,x), so that the difference between the causal effects of the first user extrinsic feature and the second user extrinsic feature on user activity is E(y|do(t=1))-E(y|do(t=2)), or, E(y|do(t=2))-E(y|do(t=1)).

[0089] By analyzing the differences in the causal effects of multiple user extrinsic features on user activity, targeted improvements can be made to the application, or users can be guided to use the application in a targeted manner, thereby improving the user activity of the application. Optionally, at least one user extrinsic feature includes at least one operating system that carries the application, and after analyzing the differences between the causal effects of at least one user extrinsic feature on user activity based on at least one causal effect estimator, the method further includes: determining an operating system to be referenced and an operating system to be optimized based on the analysis; obtaining a program data package to be referenced and a program data package to be optimized, wherein the program data package to be referenced is a program data package of the application in the operating system to be referenced, and the program data package to be optimized is a program data package of the application in the operating system to be optimized; and improving the program data package to be optimized based on the program data package to be referenced.

[0090] The operating system to be referenced is the operating system with the greater causal influence on the user activity among at least one operating system, and the operating system to be optimized is the operating system with the smaller causal influence on the user activity among at least one operating system. By improving the program data packages related to the application in the operating system to be optimized according to the program data packages related to the application in the operating system to be referenced, such as adding the unique functions of the program data packages to be referenced to the program data packages to be optimized, or imitating the running effects of the application corresponding to the program data packages to be referenced, etc., a higher user activity can be achieved for the application in each operating system.

[0091] To sum up, the technical solution provided in the embodiments of the present application analyzes the causal influence of multiple user external characteristics on user activity, and based on the differences obtained from the analysis, improves the functions and displays of the application in a targeted manner, or guides users on how to use the application in a targeted manner, thereby effectively improving the user activity of the application.

[0092] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0093] Please refer to Figure 6 , which shows a block diagram of a causal influence determination device provided by an embodiment of the present application. The device has the function of implementing the above-mentioned causal influence determination method example, and the function can be implemented by hardware, or by hardware executing corresponding software. The device can be a computer device, or it can be set in a computer device. The device 600 may include: an information acquisition module 610, a probability determination module 620, an expectation determination module 630 and an estimation amount calculation module 640.

[0094] The information acquisition module 610 is used to obtain at least one set of user information corresponding to the first user's external characteristics from the user information corresponding to at least one user account logged in in the application, each set of user information includes user intrinsic attributes, user external characteristics and user activity corresponding to a user account.

[0095] The probability determination module 620 is used to determine the occurrence probability of each type of user intrinsic attribute based on the user intrinsic attributes included in the at least one set of user information.

[0096] The expectation determination module 630 is used to determine the mathematical expectation of the user activity when each type of user intrinsic attribute is given, based on the user intrinsic attributes and user activity included in the at least one set of user information.

[0097] The estimation calculation module 640 is used to perform weighted summation processing on the mathematical expectations of user activity when each type of user intrinsic attribute is given, using the probability of occurrence of each type of user intrinsic attribute as a weight, to obtain a first causal effect estimator, wherein the first causal effect estimator is used to indicate the causal influence of the first user external feature on the user activity.

[0098] In one example, the causal influence relationship between the user's intrinsic attributes, the user's extrinsic features, and the user's activity is as follows: the user's intrinsic attributes have a causal influence on the user's extrinsic features; the user's intrinsic attributes have a causal influence on the user's activity; the user's extrinsic features have a causal influence on the user's activity.

[0099] In one example, the user's intrinsic attribute is expressed in the form of a discrete value; Figure 7 As shown, the device 600 also includes: an attribute acquisition module 650, which is used to obtain the user intrinsic attributes corresponding to the at least one user account; a discretization processing module 660, which is used to discretize the user intrinsic attributes corresponding to the at least one user account to obtain the user intrinsic attributes expressed in the form of discrete numerical values.

[0100] In one example, the user intrinsic attributes include attributes of at least one dimension; Figure 7 As shown, the discretization processing module 660 is used to: obtain the quantile information corresponding to the attributes of the at least one dimension; determine the discretization intervals corresponding to the attributes of the at least one dimension and the discretization values ​​corresponding to each discretization interval based on the quantile information corresponding to the attributes of the at least one dimension; discretize the user intrinsic attributes corresponding to the at least one user account according to the discretization intervals corresponding to the attributes of the at least one dimension and the discretization values ​​corresponding to each discretization interval, so as to obtain the user intrinsic attributes expressed in the form of discretization values.

[0101] In one example, the user intrinsic attributes include attributes of at least one of the following dimensions: user age corresponding to the user account, user economic income corresponding to the user account, and user occupation corresponding to the user account.

[0102] In one example, if Figure 7 As shown, the device 600 also includes: an estimation determination module 640, which is also used to determine at least one causal effect estimation, each causal effect estimation is used to indicate the causal influence of a user external feature on user activity; a difference analysis module 670, which is used to analyze the difference between the causal influences of at least one user external feature on user activity based on the at least one causal effect estimation.

[0103] In one example, the at least one user-external feature includes at least one operating system on which the application is mounted; Figure 7As shown, the device 600 also includes: a system determination module 680, which is used to determine the operating system to be referenced and the operating system to be optimized based on the analysis, wherein the operating system to be referenced is the operating system with a larger causal impact on the user activity among the at least one operating system, and the operating system to be optimized is the operating system with a smaller causal impact on the user activity among the at least one operating system; a data packet acquisition module 690, which is used to acquire program data packets to be referenced and program data packets to be optimized, wherein the program data packets to be referenced are the program data packets of the application program in the operating system to be referenced, and the program data packets to be optimized are the program data packets of the application program in the operating system to be optimized; a data packet improvement module 600, which is used to improve the program data packets to be optimized based on the program data packets to be referenced.

[0104] To sum up, the technical solution provided by the embodiment of the present application takes into account the correlation introduced by the user's intrinsic attributes between the user's extrinsic features and the user's activity when estimating the impact of the user's extrinsic features on the user's activity, and marginalizes the user's intrinsic attributes, thereby estimating the causal impact of the user's extrinsic features on the user's activity, avoiding the erroneous judgment of the correlation between the user's extrinsic features and the user's activity as the impact of the user's extrinsic features on the user's activity, improving the accuracy of the user activity correlation analysis, and facilitating the subsequent precise operation of the application and guiding the user's operation based on the analysis results, so as to effectively improve the user's activity.

[0105] It should be noted that the device provided in the above embodiment only uses the division of the above functional modules as an example to implement its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and their specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0106] In an exemplary embodiment, a computer device is further provided. The computer device includes a processor and a memory. A computer program is stored in the memory. The computer program is loaded and executed by the processor to implement the method for determining causal influence as described above.

[0107] In an exemplary embodiment, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, the method for determining causal influence as described above is implemented.

[0108] In an exemplary embodiment, a computer program product is also provided. When the computer program product is run on a computer device, the computer device is enabled to perform the method for determining causal influence as described above.

[0109] The above description is only an exemplary embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for determining causal influence, characterized in that: The method comprises: Acquire at least one set of user information corresponding to the first user's external characteristics from the user information corresponding to at least one user account logged in to the application, wherein each set of user information includes a user's intrinsic attribute, a user's external characteristics, and a user's activity level corresponding to one user account, wherein the user's intrinsic attribute includes at least one of the user's age, the user's occupation, and the user's economic income, and the user's external characteristics include at least one of the terminal type used by the user, the operating system on which the application is installed, and the method of registering the application; Determining the occurrence probability of each type of user intrinsic attribute based on the user intrinsic attributes included in the at least one set of user information; Based on the user intrinsic attributes and user activity included in the at least one set of user information, determining the mathematical expectation of the user activity when each type of user intrinsic attribute is given; The occurrence probability of each type of user intrinsic attribute is used as a weight, and a weighted summation process is performed on the mathematical expectation of the user activity when each type of user intrinsic attribute is given to obtain a first causal effect estimator, where the first causal effect estimator is used to indicate the causal influence of the first user extrinsic feature on the user activity.

2. The method according to claim 1, characterized in that The causal relationship between the user's intrinsic attributes, the user's extrinsic features, and the user's activity is as follows: The user's intrinsic attributes have a causal effect on the user's extrinsic characteristics; The user's intrinsic attributes have a causal impact on the user's activity; The user's external characteristics have a causal impact on the user's activity.

3. The method according to claim 1, characterized in that The user's intrinsic attribute is expressed in a discretized value; the method further includes: Obtaining user intrinsic attributes corresponding to the at least one user account; Discretization processing is performed on the user intrinsic attribute corresponding to the at least one user account to obtain the user intrinsic attribute expressed in the form of discrete numerical values.

4. The method according to claim 3, characterized in that The user intrinsic attribute includes an attribute of at least one dimension; the discretization of the user intrinsic attribute corresponding to the at least one user account to obtain the user intrinsic attribute expressed in the form of a discretized numerical value includes: Obtaining quantile information corresponding to the attributes of the at least one dimension respectively; Based on the quantile information respectively corresponding to the attributes of the at least one dimension, determining the discretization intervals respectively corresponding to the attributes of the at least one dimension, and the discretization values ​​corresponding to each discretization interval; According to the discretization intervals corresponding to the attributes of the at least one dimension and the discretization values ​​corresponding to each discretization interval, the user intrinsic attributes corresponding to the at least one user account are discretized to obtain the user intrinsic attributes expressed in the form of discretization values.

5. The method according to claim 3, characterized in that: The user intrinsic attributes include attributes of at least one of the following dimensions: user age corresponding to the user account, user economic income corresponding to the user account, and user occupation corresponding to the user account.

6. The method according to claim 1, characterized in that The method further comprises: Determine at least one causal effect estimator, each causal effect estimator is used to indicate the causal influence of a user external feature on the user activity; Based on the at least one causal effect estimator, the difference between the causal effects of at least one user external characteristic on the user activity is analyzed.

7. The method according to claim 6, characterized in that The at least one user-external characteristic includes at least one operating system on which the application is mounted; After analyzing the difference between the causal effects of at least one user external feature on the user activity based on the at least one causal effect estimator, the method further includes: Based on the analysis, determining an operating system to be referenced and an operating system to be optimized, wherein the operating system to be referenced is an operating system having a greater causal impact on user activity among the at least one operating system, and the operating system to be optimized is an operating system having a smaller causal impact on user activity among the at least one operating system; Acquire a program data packet to be referenced and a program data packet to be optimized, wherein the program data packet to be referenced is a program data packet of the application program in the operating system to be referenced, and the program data packet to be optimized is a program data packet of the application program in the operating system to be optimized; Based on the program data package to be referenced, the program data package to be optimized is improved.

8. A device for determining causal influence, characterized in that: The device comprises: an information acquisition module, configured to acquire at least one set of user information corresponding to an external feature of a first user from user information corresponding to at least one user account logged in to the application, wherein each set of user information includes a user intrinsic attribute, a user external feature, and a user activity level corresponding to a user account, wherein the user intrinsic attribute includes at least one of the user's age, the user's occupation, and the user's economic income, and the user external feature includes at least one of the type of terminal used by the user, the operating system on which the application is mounted, and the method of registering the application; A probability determination module, configured to determine the occurrence probability of each type of user intrinsic attribute based on the user intrinsic attributes included in the at least one set of user information; an expectation determination module, configured to determine the mathematical expectation of the user activity when each type of user intrinsic attribute is given, based on the user intrinsic attributes and user activity included in the at least one set of user information; The estimation calculation module is used to perform weighted summation processing on the mathematical expectations of user activity when each type of user intrinsic attribute is given, using the occurrence probability of each type of user intrinsic attribute as a weight, to obtain a first causal effect estimation, wherein the first causal effect estimation is used to indicate the causal influence of the first user extrinsic feature on the user activity.

9. A computer device, characterized in that: The computer device includes a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the method for determining causal influence according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for determining causal influence according to any one of claims 1 to 7 is implemented.

Images