Information processing method and device
By constructing and optimizing the weight matrix, the problem of inaccurate causal relationship in causal analysis is solved, and the weight matrix error is corrected, and more accurate causal relationship inference and video recommendation stability is achieved.
Patent Information
- Application Number
- CN202111510017.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-10
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2041-12-10
AI Technical Summary
In the prior art, weight matrix estimation errors caused by missing variables in causal analysis, resulting in inaccurate inference of causal relationships, especially when some variable data cannot be observed, such as the idle time of the user account, which affects the accuracy of video recommendations. sex.
By obtaining the experimental data set and observed feature data set, constructing the input matrix and tool variable matrix, optimizing the predefined objective function to obtain the weight matrix, limiting the value of the weight matrix elements, and using the external randomness of the experimental grouping to correct the weight matrix matrix to ensure the stability of causal relationship.
The problem of inaccurate inference of causal relationships was solved, the estimation error of the weight matrix was corrected, and more accurate causal relationships were obtained, which improved the stability of the causal graph and the satisfaction of user needs.
Smart Images

Figure CN114186096B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of data analysis, and in particular to an information processing method and device. Background Art
[0002] For causal analysis with many independent variables, it is necessary to understand whether there is a relationship between the independent variables, and whether a certain variable has an indirect or direct impact on the dependent variable. This often involves the problem of discrete search. In the related art, the problem of discrete search is converted into the problem of continuous search, so that the generated causal graph satisfies the condition of directed non-closed loop. However, this method may cause a large estimation bias of the weight matrix due to the problem of omitted variables, making the inference of the causal relationship between variables inaccurate. For example, for the network platform, it is impossible to observe the data of some variables such as the daily idle time length of the user account. In this case, when the user wants to watch the video of the network platform in his spare time, the video recommended by the network platform may be too long or too short due to the inability to observe the data of the daily idle time length of the user account, which cannot meet the needs of the user. In order to better meet the needs of users, it is necessary to solve the problem of inaccurate inference of the causal relationship between the operation information of the user account due to omitted data when conducting causal analysis on the data of the network platform. Summary of the invention
[0003] The present disclosure provides an information processing method and device to at least solve the above-mentioned problems in the related technologies, or may not solve any of the above-mentioned problems. The technical solution of the present disclosure is as follows:
[0004] According to a first aspect of an embodiment of the present disclosure, there is provided an information processing method, comprising: acquiring an experimental data set and an observed feature data set, wherein the feature data set includes feature data generated by each user account among multiple user accounts for each variable among multiple variables, the experimental data set is feature data obtained by conducting experiments on the multiple user accounts and experimental variables, the experimental variables include at least the multiple variables, and the feature data is operation information of the user account on a predetermined platform; based on the experimental data set and the observed feature data set, obtaining a causal relationship between the multiple variables, wherein the causal relationship between the multiple variables represents the correlation between different operation information.
[0005] Optionally, obtaining the causal relationship between the multiple variables based on the experimental data set and the observed feature data set includes: obtaining an input matrix based on the observed feature data set, and obtaining an instrumental variable matrix based on the experimental data set, wherein the elements of each row of the input matrix are feature data generated by the same user account for each of the multiple variables, the elements of each column of the input matrix are feature data generated by each of the multiple user accounts for the same variable, the elements of each row of the instrumental variable matrix are feature data generated by the same user account for each of the multiple experiments, the elements of each column of the instrumental variable matrix are feature data generated by each of the multiple user accounts for the same experiment, and each of the multiple experiments is conducted on at least one variable among the experimental variables; obtaining a weight matrix based on the input matrix and the instrumental variable matrix, wherein the weight matrix characterizes the causal relationship between the multiple variables.
[0006] Optionally, obtaining a weight matrix according to the input matrix and the instrumental variable matrix includes: optimizing a predefined objective function according to the input matrix and the instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function.
[0007] Optionally, before optimizing the predefined objective function according to the input matrix and the instrumental variable matrix to obtain the weight matrix that minimizes the value of the predefined objective function, it also includes: restricting the values of the elements in the weight matrix according to the predefined relationship between the multiple variables.
[0008] Optionally, the predefined objective function includes an optimization term, which is the sum of an optimization function of each endogenous variable and an optimization function of each exogenous variable, wherein the endogenous variables and the exogenous variables are variables among the multiple variables, the endogenous variables are related to unobserved variables, the exogenous variables are independent of the unobserved variables, and the unobserved variables affect the causal relationship between the multiple variables and are not variables among the multiple variables.
[0009] Optionally, for the variables corresponding to the elements in the i-th column of the input matrix, when the variables corresponding to the elements in the i-th column are endogenous variables, the optimization function of the variables corresponding to the elements in the i-th column is obtained by the following steps: obtaining the residuals of regressing the i-th column of the input matrix on the columns of the input matrix except the i-th column according to the i-th column of the input matrix, the columns of the input matrix except the i-th column, and the weight matrix; obtaining the weighting matrix according to the residuals, the instrumental variable matrix, and the number of user accounts of the multiple user accounts; obtaining the optimization function of the variables corresponding to the elements in the i-th column according to the residuals, the instrumental variable matrix, and the weighting matrix.
[0010] Optionally, for the variables corresponding to the elements in the i-th column of the input matrix, when the variables corresponding to the elements in the i-th column are exogenous variables, the optimization function of the variables corresponding to the elements in the i-th column is obtained by the following steps: according to the i-th column of the input matrix, the columns of the input matrix other than the i-th column, and the weight matrix, the residual of regressing the i-th column of the input matrix on the columns of the input matrix other than the i-th column is obtained; according to the residual, the optimization function of the variables corresponding to the elements in the i-th column is obtained.
[0011] According to a second aspect of an embodiment of the present disclosure, there is provided an information processing device, comprising: an acquisition unit, configured to: acquire an experimental data set and an observed feature data set, wherein the feature data set includes feature data generated by each user account in a plurality of user accounts for each variable in a plurality of variables, the experimental data set is feature data obtained by conducting experiments on the plurality of user accounts and experimental variables, the experimental variables at least include the plurality of variables, and the feature data is operation information of the user account on a predetermined platform; a determination unit, configured to: obtain a causal relationship between the plurality of variables based on the experimental data set and the observed feature data set, wherein the causal relationship between the plurality of variables represents the correlation relationship between different operation information.
[0012] Optionally, the determination unit is configured to: obtain an input matrix based on the observed feature data set, and obtain an instrumental variable matrix based on the experimental data set, wherein the elements of each row of the input matrix are feature data generated by the same user account for each of the multiple variables, the elements of each column of the input matrix are feature data generated by each of the multiple user accounts for the same variable, the elements of each row of the instrumental variable matrix are feature data generated by the same user account for each of the multiple user accounts for the same variable, the elements of each column of the instrumental variable matrix are feature data generated by the same user account for each of the multiple experiments, and each of the multiple experiments is conducted on at least one variable among the experimental variables; obtain a weight matrix based on the input matrix and the instrumental variable matrix, wherein the weight matrix characterizes the causal relationship between the multiple variables.
[0013] Optionally, the determination unit is configured to: optimize a predefined objective function according to the input matrix and the instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function.
[0014] Optionally, it also includes an adjustment unit, which is configured to: before optimizing a predefined objective function based on the input matrix and the instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function, restrict the values of elements in the weight matrix based on the predefined relationship between the multiple variables.
[0015] Optionally, the predefined objective function includes an optimization term, which is the sum of an optimization function of each endogenous variable and an optimization function of each exogenous variable, wherein the endogenous variables and the exogenous variables are variables among the multiple variables, the endogenous variables are related to unobserved variables, the exogenous variables are independent of the unobserved variables, and the unobserved variables affect the causal relationship between the multiple variables and are not variables among the multiple variables.
[0016] Optionally, for the variables corresponding to the elements of the i-th column of the input matrix, when the variables corresponding to the elements of the i-th column are endogenous variables, the optimization function of the variables corresponding to the elements of the i-th column is obtained by a first function unit; the first function unit is configured to: obtain the residuals of regressing the i-th column of the input matrix on the columns of the input matrix except the i-th column according to the i-th column of the input matrix, the columns of the input matrix except the i-th column, and the weight matrix; obtain the weighting matrix according to the residuals, the instrumental variable matrix, and the number of user accounts of the multiple user accounts; obtain the optimization function of the variables corresponding to the elements of the i-th column according to the residuals, the instrumental variable matrix, and the weighting matrix.
[0017] Optionally, for the variables corresponding to the elements of the i-th column of the input matrix, when the variables corresponding to the elements of the i-th column are exogenous variables, the optimization function of the variables corresponding to the elements of the i-th column is obtained through a second function unit; the second function unit is configured to: obtain the residual of regressing the i-th column of the input matrix on the columns of the input matrix except the i-th column according to the i-th column of the input matrix, the columns of the input matrix except the i-th column, and the weight matrix; and obtain the optimization function of the variables corresponding to the elements of the i-th column according to the residual.
[0018] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising: at least one processor; and at least one memory storing computer-executable instructions, wherein the computer-executable instructions, when executed by the at least one processor, cause the at least one processor to execute an information processing method according to the present disclosure.
[0019] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions stored in the computer-readable storage medium are executed by at least one processor, the at least one processor is prompted to execute the information processing method according to the present disclosure.
[0020] According to a fifth aspect of an embodiment of the present disclosure, a computer program product is provided, comprising computer instructions, which implement the information processing method according to the present disclosure when executed by at least one processor.
[0021] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects:
[0022] According to the information processing method and device disclosed in the present invention, the causal relationship between multiple variables is obtained based on the experimental data set and the observed feature data set, and the experimental data can be introduced into the causal analysis, so that the problem of inaccurate inference of the causal correlation relationship between the operation information of user accounts due to missing data in the related technology is solved, and a more accurate causal relationship between user operation information is obtained.
[0023] In addition, according to the information processing method and device disclosed in the present invention, the estimation bias of the larger weight matrix caused by the problem of omitted variables in the related technology can be corrected, and a stable causal relationship between variables can be obtained.
[0024] In addition, according to the information processing method and device disclosed in the present invention, the instrumental variable matrix can be obtained based on the experimental data set and applied to the predefined objective function. By utilizing the external randomness of the experimental grouping and the non-correlation condition of the moment between the experiment and the residual, the estimation bias of the weight matrix can be corrected.
[0025] 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
[0026] 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.
[0027] Figure 1 The invention is a flowchart of an information processing method according to an exemplary embodiment.
[0028] Figure 2 It is an overall flow chart of an information processing method according to an exemplary embodiment.
[0029] Figure 3 The invention is a block diagram of an information processing device according to an exemplary embodiment.
[0030] Figure 4 is a block diagram of an electronic device 400 according to an exemplary embodiment. DETAILED DESCRIPTION
[0031] 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.
[0032] 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 implementation methods described in the following examples do not represent all implementation methods 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 attached claims.
[0033] It should be noted that the phrase "at least one of the items" in the present disclosure includes three types of parallel situations: "any one of the items", "a combination of any number of the items", and "all of the items". For example, "including at least one of A and B" includes the following three parallel situations: (1) including A; (2) including B; (3) including A and B. Another example is "executing at least one of step 1 and step 2" which means the following three parallel situations: (1) executing step 1; (2) executing step 2; (3) executing step 1 and step 2.
[0034] For estimating the causal relationship between variables, methods such as regression are usually used, but these methods can only obtain the direct impact of the independent variable on the dependent variable. For causal analysis with more independent variables, it is necessary to understand whether there is a relationship between the independent variables, and whether a certain variable has an indirect or direct impact on the dependent variable. To address this problem, the traditional method is based on searching all possible causal graphs based on all possible relationships between all nodes and nodes, and generating the final causal graph according to a certain standard. Here, the nodes can be variables. However, this involves a discrete search problem, which is a typical NP-hard problem. According to the traditional method, the generation of the causal graph will take a very long time, and the existing computing resources are difficult to meet the computing requirements.
[0035] In the related art, the non-combinatorial optimization method via trace exponential and augmented Lagrangian for structure learning (NOTEARS) is an algorithm for learning causal network relationships between multiple variables. NOTEARS can transform the search interval of a discrete causal graph into a continuous search interval, that is, transform the discrete search problem into a continuous search problem, so that the generated causal graph satisfies the condition of being directed and non-closed.
[0036] However, the NOTEARS method is highly hypothetical, assuming that all data of variables related to the causal relationship in the causal graph must be observable and controllable. However, in some cases, it is impossible to observe the data of all variables, and these data may affect the causal relationship. In this case, the use of NOTEARS to generate a causal graph may lead to a large estimation bias of the weight matrix due to the problem of omitted variables. The weight matrix represents the causal relationship between variables, which will make the causal relationship inference in the causal graph unstable, resulting in an unstable structure of the generated causal graph. For example, for the network platform, it is impossible to observe the data of some variables such as the daily idle time length of the user account. In this case, when the user wants to watch the video of the network platform in his spare time, the video duration recommended by the network platform may be too long or too short due to the inability to observe the data of the daily idle time length of the user account, which cannot meet the user's needs. In order to better meet the needs of users, it is necessary to solve the problem of inaccurate inference of causal relationships between the operation information of user accounts caused by missing data when conducting causal analysis on data related to the network platform.
[0037] In order to solve the problems existing in the above-mentioned related technologies, the present disclosure proposes an information processing method and device, which obtains the causal relationship between multiple variables based on an experimental data set and an observed feature data set, and can introduce the experimental data into causal analysis, so that the problem of inaccurate inference of the causal correlation relationship between the operation information of user accounts due to missing data in the related technology is solved, and a more accurate causal relationship between user operation information is obtained.
[0038] Next, we will refer to Figures 1 to 4 The information processing method and device according to the present disclosure are described in detail.
[0039] Figure 1 is a flowchart of an information processing method according to an exemplary embodiment. Figure 1 In step 101, an experimental data set and an observed feature data set may be obtained, wherein the feature data set includes feature data generated by each user account in a plurality of user accounts for each variable in a plurality of variables, the experimental data set is feature data obtained by conducting experiments on a plurality of user accounts and experimental variables, the experimental variables include at least the plurality of variables mentioned above, and the feature data is operation information of the user account on a predetermined platform.
[0040] According to an exemplary embodiment of the present disclosure, the experiment may be, but is not limited to, an AB experiment.
[0041] According to an exemplary embodiment of the present disclosure, a variable may represent a category of feature data. For example, a variable may be the duration of watching a video, and feature data corresponding to the variable may include 30 minutes, 35 minutes, and the like.
[0042] In step 102, a causal relationship between multiple variables may be obtained based on the experimental data set and the observed feature data set, wherein the causal relationship between the multiple variables represents the correlation relationship between different operation information.
[0043] According to an exemplary embodiment of the present disclosure, different operation information may be feature data generated by different variables. Different variables may be each variable among the above-mentioned multiple variables, and different operation information may be different operation information of the same user account.
[0044] According to an exemplary embodiment of the present disclosure, first, an input matrix can be obtained according to an observed feature data set, and an instrumental variable matrix can be obtained according to an experimental data set. Then, a weight matrix can be obtained according to the input matrix and the instrumental variable matrix, wherein the weight matrix represents the causal relationship between multiple variables. Here, the elements of each row of the input matrix are feature data generated by the same user account for each of the multiple variables, the elements of each column of the input matrix are feature data generated by each of the multiple user accounts for the same variable, the elements of each row of the instrumental variable matrix are feature data generated by the same user account for each of the multiple experiments, the elements of each column of the instrumental variable matrix are feature data generated by each of the multiple user accounts for the same experiment, and each of the multiple experiments is conducted on at least one variable among the experimental variables.
[0045] Here, each row of the input matrix can represent the characteristic data of a user account, and the user accounts between rows are not repeated. Each column of the input matrix can represent the characteristic data of a variable, and the variables between columns are not repeated. Each row of the instrumental variable matrix can represent the characteristic data of a user account, and the user accounts between rows are not repeated. Each column of the instrumental variable matrix can represent the characteristic data of an experiment, and the experiments between columns are not repeated.
[0046] According to an exemplary embodiment of the present disclosure, a predefined objective function may be optimized according to an input matrix and an instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function.
[0047] According to an exemplary embodiment of the present disclosure, before optimizing a predefined objective function according to an input matrix and an instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function, the values of the elements in the weight matrix may be restricted according to the relationship between a plurality of predefined variables. For example, a plurality of variables may include the number of times a certain video is displayed and the number of times the video is viewed, and the relationship between a plurality of predefined variables may include that the number of times a certain video is displayed determines the number of times the video is viewed, and based on this relationship, the values of the elements in the weight matrix related to the number of times a certain video is displayed and the number of times the video is viewed may be restricted. Here, restricting the value of an element may be, but is not limited to, restricting the value range of the element and setting the element to a preset value.
[0048] According to an exemplary embodiment of the present disclosure, after restricting the values of elements in the weight matrix according to the relationship between multiple predefined variables, the predefined objective function can be optimized to obtain a weight matrix that minimizes the value of the predefined objective function, and the values of the elements in the obtained weight matrix still meet the restricted range. For example, according to the relationship between multiple predefined variables, the first element in the weight matrix is set to a preset value, and the first element in the weight matrix that minimizes the value of the predefined objective function is still the preset value.
[0049] Here, the predefined objective function may include an optimization term and a penalty term.
[0050] For example, the predefined objective function can be expressed as formula (1):
[0051] F(W)=l(W;X) 优化项 +λ||W||1 (1)
[0052] Among them, F(W) is the predefined objective function, l(W;X) 优化项 is the optimization term, λ||W||1 is the penalty term, W is a weight matrix of k×k dimensions, X is an input matrix of N×k dimensions, N and k are both positive integers, N is the number of user accounts of multiple user accounts, k is the number of variables of multiple variables, λ is an adjustable parameter, and ||W||1 is the L-1 penalty term of W. It should be noted that the L-1 penalty term can be used to penalize sparse matrices, and λ can be determined by cross-validation.
[0053] It should be noted that before optimizing the predefined objective function, each element in the weight matrix exists in the form of an initial value, and the initial value of each element in the weight matrix may be a preset custom value.
[0054] For the optimization items in the predefined objective function, NOTEARS in the related art can be expressed by the following formula (2):
[0055]
[0056] Among them, l(W;X) 优化项 is the optimization term, W is a weight matrix of k×k dimensions, X is an input matrix of N×k dimensions, N and k are both positive integers, N is the number of user accounts of the multiple user accounts, k is the number of variables of the multiple variables, ||X-XW|| F Represents the Frobenius norm of X-XW, where XW is the product of X and W in N×k dimensions.
[0057] It can be seen that the optimization item represented in formula (2) does not classify the variables corresponding to the elements of each column of the input matrix. Based on this, in an exemplary embodiment of the present disclosure, the optimization item can be the sum of the optimization function of each endogenous variable and the optimization function of each exogenous variable, wherein the endogenous variable and the exogenous variable can be variables among multiple variables, the endogenous variable is related to the unobserved variable, the exogenous variable is unrelated to the unobserved variable, and the unobserved variable affects the causal relationship between multiple variables and is not a variable among multiple variables. For example, the unobserved variable can be a variable that affects the causal relationship between multiple variables, but the corresponding feature data cannot be observed.
[0058] According to an exemplary embodiment of the present disclosure, for the variable corresponding to the element in the i-th column of the input matrix, when the variable corresponding to the element in the i-th column is an endogenous variable, the optimization function of the variable corresponding to the element in the i-th column is obtained by the following steps: First, the residual of the i-th column of the input matrix regressing the column in the input matrix except the i-th column can be obtained according to the i-th column of the input matrix, the column in the input matrix except the i-th column, and the weight matrix. Then, the weight matrix can be obtained according to the residual, the instrumental variable matrix, and the number of user accounts of multiple user accounts. Finally, the optimization function of the variable corresponding to the element in the i-th column can be obtained according to the residual, the instrumental variable matrix, and the weight matrix.
[0059] For example, for the variable corresponding to the element in the i-th column of the input matrix, when the variable corresponding to the element in the i-th column is an endogenous variable, the optimization function of the variable corresponding to the element in the i-th column is expressed as the following formula (3):
[0060]
[0061] in, u i =X i -∑- i x -i Wi,-i , 1≤i≤k, l(W;X) is the optimization function of the variable corresponding to the element in the i-th column, W is a weight matrix of k×k dimensions, X is an input matrix of N×k dimensions, Z is an instrumental variable matrix of N×L dimensions, N, k, and L are all positive integers, N is the number of user accounts for multiple user accounts, k is the number of variables for multiple variables, L is the number of experiments for multiple experiments, L≥k, Z′ is the transposed matrix of Z, and X i is the i-th column of X, X -i is the column of X except the i-th column, W i,-i is the i-th row of W and the columns of W except the i-th column, M k is the weighting matrix, u i is the residual of the regression of the i-th column of X on the columns of X other than the i-th column, u i ′ is the transposed matrix of ui.
[0062] According to an exemplary embodiment of the present disclosure, for the variables corresponding to the elements in the i-th column of the input matrix, when the variables corresponding to the elements in the i-th column are exogenous variables, the optimization function of the variables corresponding to the elements in the i-th column is obtained by the following steps: according to the i-th column of the input matrix, the columns of the input matrix except the i-th column, and the weight matrix, the residual of regressing the i-th column of the input matrix on the columns of the input matrix except the i-th column is obtained; according to the residual, the optimization function of the variables corresponding to the elements in the i-th column is obtained.
[0063] For example, for the variable corresponding to the element in the i-th column of the input matrix, when the variable corresponding to the element in the i-th column is an exogenous variable, the optimization function of the variable corresponding to the element in the i-th column is expressed as the following formula (4):
[0064]
[0065] Wherein, 1≤i≤k, l(W;X) is the optimization function of the variable corresponding to the element in the i-th column, W is a weight matrix of k×k dimensions, X is an input matrix of N×k dimensions, N and k are both positive integers, N is the number of user accounts of the multiple user accounts, k is the number of variables of the multiple variables, and X i is the i-th column of X, X -i is the column of X except the i-th column, W i,-i is the i-th row of W and the columns of W except the i-th column. It should be noted that (X i -∑ -i X -i W i,-i ) is the residual of regressing the i-th column of X on the columns of X except the i-th column.
[0066] According to an exemplary embodiment of the present disclosure, a causal graph regarding a plurality of variables may be generated according to the plurality of variables and causal relationships between the plurality of variables.
[0067] Figure 2 is an overall flow chart of an information processing method according to an exemplary embodiment. Figure 2 , an overall description is given of the information processing method in the exemplary embodiment of the present disclosure.
[0068] First, an experimental data set and an observed feature data set may be obtained. The feature data set includes feature data generated by each user account in a plurality of user accounts for each variable in a plurality of variables, the experimental data set is feature data obtained by conducting experiments on a plurality of user accounts and experimental variables, the experimental variables at least include the plurality of variables, and the feature data may be operation information of the user account on a predetermined platform.
[0069] Here, the input matrix can be obtained based on the observed characteristic data set, and the instrumental variable matrix can be obtained based on the experimental data set.
[0070] Secondly, the values of the elements in the weight matrix included in the predefined objective function can be restricted according to the relationship between the predefined multiple variables.
[0071] Then, the predefined objective function can be optimized according to the input matrix and the instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function, wherein the weight matrix represents the causal relationship between multiple variables. The causal relationship between multiple variables reflects the correlation between different operation information.
[0072] Here, the predefined objective function is optimized, and reference may be made to equations (1), (3) and (4) above.
[0073] Finally, a causal graph about the multiple variables can be generated according to the multiple variables and the causal relationship between the multiple variables.
[0074] The information processing method according to the present disclosure is described below with an illustrative example.
[0075] First, an experimental data set and an observed feature data set can be obtained. Among them, the feature data set includes feature data of each user account in multiple user accounts for multiple variables such as the average daily video viewing time on the predetermined platform and the average daily number of clicks on the video displayed on the homepage of the predetermined platform. The experimental data set is feature data obtained by conducting experiments on multiple user accounts and experimental variables. The experimental variables can be the average daily video viewing time on the predetermined platform, the average daily live broadcast viewing time on the predetermined platform, the average daily number of clicks on the video displayed on the homepage on the predetermined platform, and the average daily number of clicks on the live broadcast displayed on the homepage on the predetermined platform.
[0076] It should be noted that the characteristic data may be the operation information of the user account on the predetermined platform. For example, the characteristic data may be that the average daily video viewing time of the first user account on the predetermined platform is 30 minutes, and the average daily video viewing time of the second user account on the predetermined platform is 50 minutes, the average number of times the first user account clicks on the video displayed on the homepage of the predetermined platform is 10 times per day, and the average number of times the second user account clicks on the video displayed on the homepage of the predetermined platform is 20 times per day, etc.
[0077] Here, the input matrix can be obtained based on the observed characteristic data set, and the instrumental variable matrix can be obtained based on the experimental data set.
[0078] Secondly, the values of the elements in the weight matrix included in the predefined objective function can be restricted according to the relationship between the predefined multiple variables. For example, the relationship between the predefined multiple variables can be that the average number of times a video displayed on the homepage is clicked on the predetermined platform per day and the average daily viewing time of the video on the predetermined platform are positively correlated. Then, the values of the elements in the weight matrix related to the relationship described above can be restricted. For example, if the element in the weight matrix related to the relationship described above is the element in the ath row and the bth column, the element in the ath row and the bth column of the weight matrix can be set to a preset value.
[0079] Finally, the predefined objective function can be optimized according to the input matrix and the instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function, wherein the weight matrix represents the causal relationship between multiple variables.
[0080] It should be noted that the causal relationship between multiple variables reflects the correlation between the average daily video viewing time of each user account on the predetermined platform and the average daily number of clicks on the video displayed on the homepage of the predetermined platform.
[0081] For example, the causal relationship between multiple variables can be that the average number of times each user account clicks on a video displayed on the homepage of a predetermined platform per day is the cause, and the average daily video viewing time on the predetermined platform is the result.
[0082] Here, the predefined objective function is optimized, and reference may be made to equations (1), (3) and (4) above.
[0083] It should also be noted that the exemplary embodiment of the present disclosure only takes the causal relationship between determining the average daily video viewing time of each user account on the predetermined platform and the average daily number of clicks on the video displayed on the homepage of the predetermined platform as an example. The present disclosure also protects the causal relationship between other operation information about the user account on the predetermined platform, such as the number of times the video is watched on the predetermined platform and the number of videos collected on the predetermined platform, which will not be repeated here.
[0084] Figure 3 is a block diagram of an information processing device according to an exemplary embodiment. Figure 3 , the information processing device 300 includes an acquisition unit 301 and a determination unit 302.
[0085] The acquisition unit 301 can acquire an experimental data set and an observed feature data set, wherein the feature data set includes feature data generated by each user account in multiple user accounts for each variable in multiple variables, and the experimental data set is feature data obtained by conducting experiments on multiple user accounts and experimental variables, and the experimental variables include at least the above-mentioned multiple variables, and the feature data is the operation information of the user account on a predetermined platform.
[0086] According to an exemplary embodiment of the present disclosure, the experiment may be, but is not limited to, an AB experiment.
[0087] According to an exemplary embodiment of the present disclosure, a variable may represent a category of feature data. For example, a variable may be the duration of watching a video, and feature data corresponding to the variable may include 30 minutes, 35 minutes, and the like.
[0088] The determination unit 302 may obtain a causal relationship between multiple variables according to the experimental data set and the observed feature data set, wherein the causal relationship between the multiple variables represents the correlation relationship between different operation information.
[0089] According to an exemplary embodiment of the present disclosure, different operation information may be feature data generated by different variables. Different variables may be each variable among the above-mentioned multiple variables, and different operation information may be different operation information of the same user account.
[0090] According to an exemplary embodiment of the present disclosure, first, the determination unit 302 may obtain an input matrix based on the observed feature data set, and obtain an instrumental variable matrix based on the experimental data set. Then, the determination unit 302 may obtain a weight matrix based on the input matrix and the instrumental variable matrix, wherein the weight matrix characterizes the causal relationship between multiple variables. Here, the elements of each row of the input matrix are feature data generated by the same user account for each of the multiple variables, the elements of each column of the input matrix are feature data generated by each of the multiple user accounts for the same variable, the elements of each row of the instrumental variable matrix are feature data generated by the same user account for each of the multiple experiments, the elements of each column of the instrumental variable matrix are feature data generated by each of the multiple user accounts for the same experiment, and each of the multiple experiments is conducted on at least one variable among the experimental variables.
[0091] Here, each row of the input matrix can represent the characteristic data of a user account, and the user accounts between rows are not repeated. Each column of the input matrix can represent the characteristic data of a variable, and the variables between columns are not repeated. Each row of the instrumental variable matrix can represent the characteristic data of a user account, and the user accounts between rows are not repeated. Each column of the instrumental variable matrix can represent the characteristic data of an experiment, and the experiments between columns are not repeated.
[0092] According to an exemplary embodiment of the present disclosure, the determination unit 302 may optimize a predefined objective function according to the input matrix and the instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function.
[0093] According to an exemplary embodiment of the present disclosure, an adjustment unit is also included. The adjustment unit can optimize the predefined objective function according to the input matrix and the instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function, and then restrict the values of the elements in the weight matrix according to the relationship between the predefined multiple variables. For example, the multiple variables may include the number of times a certain video is displayed and the number of times the video is viewed. The relationship between the predefined multiple variables may include that the number of times a certain video is displayed determines the number of times the video is viewed, and then the adjustment unit can restrict the values of the elements in the weight matrix related to the number of times a certain video is displayed and the number of times the video is viewed based on the relationship. Here, restricting the value of the element can be, but is not limited to, restricting the value range of the element and setting the element to a preset value.
[0094] According to an exemplary embodiment of the present disclosure, after the adjustment unit restricts the values of the elements in the weight matrix according to the relationship between the predefined multiple variables, the determination unit 302 can optimize the predefined objective function to obtain a weight matrix that minimizes the value of the predefined objective function, and the values of the elements in the obtained weight matrix still meet the restricted range. For example, according to the relationship between the predefined multiple variables, the first element in the weight matrix is set to a preset value, and the first element in the weight matrix that minimizes the value of the predefined objective function is still the preset value.
[0095] Here, the predefined objective function may include an optimization term and a penalty term.
[0096] For example, the predefined objective function can be expressed as the above formula (1).
[0097] It should be noted that before optimizing the predefined objective function, each element in the weight matrix exists in the form of an initial value, and the initial value of each element in the weight matrix may be a preset custom value.
[0098] For the optimization items in the predefined objective function, NOTEARS in the related art can be expressed by the above formula (2).
[0099] The optimization item represented in formula (2) does not classify the variables corresponding to the elements of each column of the input matrix. Based on this, in an exemplary embodiment of the present disclosure, the optimization item may be the sum of the optimization function of each endogenous variable and the optimization function of each exogenous variable, wherein the endogenous variable and the exogenous variable may be variables among multiple variables, the endogenous variable is related to the unobserved variable, the exogenous variable is unrelated to the unobserved variable, the unobserved variable affects the causal relationship between multiple variables and is not a variable among multiple variables. For example, the unobserved variable may be a variable that affects the causal relationship between multiple variables, but the corresponding feature data cannot be observed.
[0100] According to an exemplary embodiment of the present disclosure, for the variables corresponding to the elements in the i-th column of the input matrix, when the variables corresponding to the elements in the i-th column are endogenous variables, the optimization function of the variables corresponding to the elements in the i-th column is obtained by the first function unit; the first function unit is configured to: obtain the residual of the i-th column of the input matrix regressing the columns of the input matrix except the i-th column according to the i-th column of the input matrix, the columns of the input matrix except the i-th column, and the weight matrix; obtain the weighting matrix according to the residuals, the instrumental variable matrix, and the number of user accounts of multiple user accounts; obtain the optimization function of the variables corresponding to the elements in the i-th column according to the residuals, the instrumental variable matrix, and the weighting matrix.
[0101] For example, for the variable corresponding to the element in the i-th column of the input matrix, when the variable corresponding to the element in the i-th column is an endogenous variable, the optimization function of the variable corresponding to the element in the i-th column is expressed as the above formula (3).
[0102] According to an exemplary embodiment of the present disclosure, for the variables corresponding to the elements in the i-th column of the input matrix, when the variables corresponding to the elements in the i-th column are exogenous variables, the optimization function of the variables corresponding to the elements in the i-th column is obtained by the second function unit; the second function unit is configured to: obtain the residual of regressing the i-th column of the input matrix on the columns of the input matrix except the i-th column according to the i-th column of the input matrix, the columns of the input matrix except the i-th column, and the weight matrix; and obtain the optimization function of the variables corresponding to the elements in the i-th column according to the residual.
[0103] For example, for the variable corresponding to the element in the i-th column of the input matrix, when the variable corresponding to the element in the i-th column is an exogenous variable, the optimization function of the variable corresponding to the element in the i-th column is expressed as the above formula (4).
[0104] According to an exemplary embodiment of the present disclosure, the information processing may further include a generating unit, and the generating unit may generate a causal graph regarding the plurality of variables according to the plurality of variables and the causal relationship between the plurality of variables.
[0105] Figure 4 is a block diagram of an electronic device 400 according to an exemplary embodiment.
[0106] Reference Figure 4 The electronic device 400 includes at least one memory 401 and at least one processor 402, wherein the at least one memory 401 stores a set of computer executable instructions, and when the computer executable instruction set is executed by the at least one processor 402, an information processing method according to an exemplary embodiment of the present disclosure is executed.
[0107] As an example, the electronic device 400 may be a PC, a tablet device, a personal digital assistant, a smart phone, or other device capable of executing the above-mentioned instruction set. Here, the electronic device 400 is not necessarily a single electronic device, but may also be any device or circuit collection capable of executing the above-mentioned instructions (or instruction sets) individually or in combination. The electronic device 400 may also be part of an integrated control system or system manager, or may be configured as a portable electronic device interconnected with a local or remote (e.g., via wireless transmission) interface.
[0108] In the electronic device 400, the processor 402 may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller or a microprocessor. As an example and not limitation, the processor may also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, etc.
[0109] The processor 402 may execute instructions or codes stored in the memory 401, which may also store data. Instructions and data may also be sent and received over a network via a network interface device, which may employ any known transmission protocol.
[0110] The memory 401 may be integrated with the processor 402, for example, by placing RAM or flash memory within an integrated circuit microprocessor or the like. In addition, the memory 401 may include a separate device, such as an external disk drive, a storage array, or any other storage device that can be used by a database system. The memory 401 and the processor 402 may be operatively coupled, or may communicate with each other, such as through an I / O port, a network connection, etc., so that the processor 402 can read files stored in the memory.
[0111] In addition, the electronic device 400 may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, a mouse, a touch input device, etc.) All components of the electronic device 400 may be connected to each other via a bus and / or a network.
[0112] According to an exemplary embodiment of the present disclosure, a computer-readable storage medium may also be provided, wherein when the instructions stored in the computer-readable storage medium are executed by at least one processor, the at least one processor is prompted to perform the information processing method according to the exemplary embodiment of the present disclosure. Examples of computer-readable storage media here include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disk storage, hard disk drive (HDD), solid state drive (SSD), card storage (such as, multimedia card, secure digital (SD) card or extreme digital (XD) card), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk and any other device, any other device is configured to store computer programs and any associated data, data files and data structures in a non-transitory manner and provide the computer programs and any associated data, data files and data structures to a processor or computer so that the processor or computer can execute the computer program. The computer program in the above-mentioned computer-readable storage medium can be run in an environment deployed in a computer device such as a client, a host, an agent device, a server, etc. In addition, in one example, the computer program and any associated data, data files and data structures are distributed on a networked computer system, so that the computer program and any associated data, data files and data structures are stored, accessed and executed in a distributed manner by one or more processors or computers.
[0113] According to an exemplary embodiment of the present disclosure, a computer program product may also be provided. Instructions in the computer program product may be executed by a processor of a computer device to implement the information processing method according to the exemplary embodiment of the present disclosure.
[0114] According to the information processing method and device disclosed in the present invention, the causal relationship between multiple variables is obtained based on the experimental data set and the observed feature data set, and the experimental data can be introduced into the causal analysis, so that the problem of inaccurate inference of the causal correlation relationship between the operation information of user accounts due to missing data in the related technology is solved, and a more accurate causal relationship between user operation information is obtained.
[0115] In addition, according to the information processing method and device disclosed in the present invention, the estimation bias of the larger weight matrix caused by the problem of omitted variables in the related technology can be corrected, and a stable causal relationship between variables can be obtained.
[0116] In addition, according to the information processing method and device disclosed in the present invention, the instrumental variable matrix can be obtained based on the experimental data set and applied to the predefined objective function. By utilizing the external randomness of the experimental grouping and the non-correlation condition of the moment between the experiment and the residual, the estimation bias of the weight matrix can be corrected.
[0117] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.
[0118] 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. An information processing method, characterized in that: include: Acquire an experimental data set and an observed feature data set, wherein the feature data set includes feature data generated by each user account in a plurality of user accounts for each variable in a plurality of variables, the experimental data set is feature data obtained by conducting experiments on the plurality of user accounts and experimental variables, the experimental variables at least include the plurality of variables, and the feature data is operation information of the user account on a predetermined platform; Obtaining a causal relationship between the multiple variables according to the experimental data set and the observed feature data set, wherein the causal relationship between the multiple variables represents an association relationship between different operation information; Wherein, obtaining the causal relationship between the multiple variables according to the experimental data set and the observed feature data set includes: obtaining an input matrix according to the observed feature data set, and obtaining an instrumental variable matrix according to the experimental data set; A weight matrix is obtained according to the input matrix and the instrumental variable matrix, wherein the weight matrix represents the causal relationship between the multiple variables; Wherein, the weight matrix is obtained according to the input matrix and the instrumental variable matrix, including: Optimizing a predefined objective function according to the input matrix and the instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function; The predefined objective function includes an optimization term, which is the sum of the optimization function of each endogenous variable and the optimization function of each exogenous variable. Among them, the endogenous variable and the exogenous variable are variables among the multiple variables, the endogenous variable is related to the unobserved variable, the exogenous variable is unrelated to the unobserved variable, the unobserved variable affects the causal relationship between the multiple variables and is not a variable among the multiple variables.
2. The information processing method according to claim 1, characterized in that: in, The elements of each row of the input matrix are feature data generated by the same user account for each of the multiple variables, the elements of each column of the input matrix are feature data generated by each of the multiple user accounts for the same variable, the elements of each row of the instrumental variable matrix are feature data generated by the same user account for each of the multiple experiments, the elements of each column of the instrumental variable matrix are feature data generated by each of the multiple user accounts for the same experiment, and each of the multiple experiments is conducted on at least one variable among the experimental variables.
3. The information processing method according to claim 1, wherein: Before optimizing the predefined objective function according to the input matrix and the instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function, the method further includes: The values of the elements in the weight matrix are restricted according to the predefined relationship between the multiple variables.
4. The information processing method according to claim 1, wherein: For the variables corresponding to the elements in the ith column of the input matrix, when the variables corresponding to the elements in the ith column are endogenous variables, the optimization function of the variables corresponding to the elements in the ith column is obtained by the following steps: According to the i-th column of the input matrix, the columns of the input matrix other than the i-th column, and the weight matrix, obtaining the residual of regressing the i-th column of the input matrix on the columns of the input matrix other than the i-th column; Obtaining a weighting matrix according to the residual, the instrumental variable matrix, and the number of user accounts of the plurality of user accounts; An optimization function of the variables corresponding to the elements in the i-th column is obtained according to the residual, the instrumental variable matrix and the weighting matrix.
5. The information processing method according to claim 1, wherein: For the variable corresponding to the element in the i-th column of the input matrix, when the variable corresponding to the element in the i-th column is an exogenous variable, the optimization function of the variable corresponding to the element in the i-th column is obtained by the following steps: According to the i-th column of the input matrix, the columns of the input matrix other than the i-th column, and the weight matrix, obtaining the residual of regressing the i-th column of the input matrix on the columns of the input matrix other than the i-th column; According to the residual, an optimization function of the variable corresponding to the element in the i-th column is obtained.
6. An information processing device, characterized in that: include: an acquisition unit, configured to: acquire an experimental data set and an observed feature data set, wherein the feature data set includes feature data generated by each user account in a plurality of user accounts for each variable in a plurality of variables, the experimental data set is feature data obtained by conducting experiments on the plurality of user accounts and experimental variables, the experimental variables at least include the plurality of variables, and the feature data is operation information of the user account on a predetermined platform; A determination unit is configured to: obtain a causal relationship between the multiple variables according to the experimental data set and the observed feature data set, wherein the causal relationship between the multiple variables represents an association relationship between different operation information; Wherein, the determining unit is configured as: obtaining an input matrix according to the observed feature data set, and obtaining an instrumental variable matrix according to the experimental data set; A weight matrix is obtained according to the input matrix and the instrumental variable matrix, wherein the weight matrix represents the causal relationship between the multiple variables; Wherein, the determining unit is configured as: Optimizing a predefined objective function according to the input matrix and the instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function; The predefined objective function includes an optimization term, which is the sum of the optimization function of each endogenous variable and the optimization function of each exogenous variable. Among them, the endogenous variable and the exogenous variable are variables among the multiple variables, the endogenous variable is related to the unobserved variable, the exogenous variable is unrelated to the unobserved variable, the unobserved variable affects the causal relationship between the multiple variables and is not a variable among the multiple variables.
7. The information processing device according to claim 6, wherein: in, The elements of each row of the input matrix are feature data generated by the same user account for each of the multiple variables, the elements of each column of the input matrix are feature data generated by each of the multiple user accounts for the same variable, the elements of each row of the instrumental variable matrix are feature data generated by the same user account for each of the multiple experiments, the elements of each column of the instrumental variable matrix are feature data generated by each of the multiple user accounts for the same experiment, and each of the multiple experiments is conducted on at least one variable among the experimental variables.
8. The information processing device according to claim 6, wherein: Also included is an adjustment unit configured to: Before optimizing a predefined objective function based on the input matrix and the instrumental variable matrix to obtain a weight matrix that minimizes the value of the predefined objective function, the values of elements in the weight matrix are restricted based on the predefined relationship between the multiple variables.
9. The information processing device according to claim 6, wherein: For the variable corresponding to the element in the i-th column of the input matrix, when the variable corresponding to the element in the i-th column is an endogenous variable, the optimization function of the variable corresponding to the element in the i-th column is obtained by the first function unit; The first functional unit is configured as: According to the i-th column of the input matrix, the columns of the input matrix other than the i-th column, and the weight matrix, obtaining the residual of regressing the i-th column of the input matrix on the columns of the input matrix other than the i-th column; Obtaining a weighting matrix according to the residual, the instrumental variable matrix, and the number of user accounts of the plurality of user accounts; An optimization function of the variables corresponding to the elements in the i-th column is obtained according to the residual, the instrumental variable matrix and the weighting matrix.
10. The information processing device according to claim 6, wherein: For the variable corresponding to the element in the i-th column of the input matrix, when the variable corresponding to the element in the i-th column is an exogenous variable, the optimization function of the variable corresponding to the element in the i-th column is obtained by the second function unit; The second functional unit is configured as: According to the i-th column of the input matrix, the columns of the input matrix other than the i-th column, and the weight matrix, obtaining the residual of regressing the i-th column of the input matrix on the columns of the input matrix other than the i-th column; According to the residual, an optimization function of the variable corresponding to the element in the i-th column is obtained.
11. An electronic device, characterized in that: include: at least one processor; at least one memory storing computer executable instructions, Wherein, when the computer executable instructions are executed by the at least one processor, the at least one processor is prompted to execute the information processing method according to any one of claims 1 to 5.
12. A computer-readable storage medium, characterized in that: When the instructions stored in the computer-readable storage medium are executed by at least one processor, the at least one processor is prompted to perform the information processing method according to any one of claims 1 to 5.
13. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by at least one processor, the information processing method according to any one of claims 1 to 5 is implemented.
