Method, device, equipment and computer storage medium for determining target user group

Abstract

The embodiment of the present application provides a method, device, equipment and computer storage medium for determining a target user group, the method comprising: obtaining a first-tier target user group including multiple target users according to target feature data and a classification model of multiple users, the classification model being obtained by learning the mapping relationship between the feature data of sample users and target users; obtaining the T-tier target user group by the following steps, T>1: obtaining the target feature data of the first user, the target feature data comprising multiple sub-target feature data; calculating the m quantiles of each sub-target feature data of the first user; determining the target quantile of each sub-target feature data from the m quantiles based on a threshold search algorithm; obtaining the T-tier target user group according to each sub-target feature data and the target quantile of the first user. The target user group determined by the embodiment of the present application has the advantages of moderate scale and precise positioning.

Description

Technical Field

[0001] The present application belongs to the field of Internet application technology, and in particular, relates to a method, device, equipment and computer storage medium for determining a target user group. Background Art

[0002] With the development of communication technology and the widespread use of communication equipment, the number of users who have not yet applied for mobile phone numbers is becoming fewer and fewer, and the service demand of users who have already applied for mobile phone numbers is also tending to be saturated, making it increasingly difficult for operators to expand new users or new services.

[0003] In order to improve the success rate of business promotion among users, the existing technology usually screens out a target user group with a higher expected promotion success rate from a wide range of users. However, the existing technology has the defects of too large a scale of the selected target user group and inaccurate positioning when screening the target user group. Summary of the invention

[0004] The embodiments of the present application provide a method, apparatus, device and computer storage medium for determining a target user group, which can solve the technical problems in the prior art of the target user group being too large and the positioning being inaccurate.

[0005] On the one hand, an embodiment of the present application provides a method for determining a target user group, the method comprising:

[0006] According to the target feature data of multiple users and the pre-established user group classification model, a first-stage target user group including multiple target users is obtained, and the user group classification model is obtained by learning the mapping relationship between the feature data of multiple sample users and the target users;

[0007] The target user group of the T-th stage is obtained by the following steps, where T is a positive integer greater than 1:

[0008] Acquire target feature data of a first user, the target feature data including a plurality of sub-target feature data, the first user including users other than users in a first-tier target user group to a T-1-tier target user group among the plurality of users;

[0009] Calculate m quantiles of each sub-target feature data of the first user, where m is a positive integer;

[0010] Determine the target quantile of each sub-target feature data from the m quantiles based on the threshold search algorithm;

[0011] According to each sub-target feature data and target quantile of the first user, the T-th level target user group is obtained.

[0012] As an implementation manner, before obtaining a first-tier target user group including a plurality of target users according to target feature data of a plurality of users and a pre-established user group classification model, the method further includes:

[0013] Obtain characteristic data of multiple users;

[0014] Preprocessing the acquired feature data of the multiple users to remove duplicate data and abnormal data beyond a preset range in the feature data of the multiple users, and filling in missing data in the feature data of the multiple users;

[0015] The preprocessed feature data of multiple users are screened according to the target feature to obtain target feature data of multiple users corresponding to the target feature.

[0016] As an implementation mode, preprocessing the acquired characteristic data of multiple users specifically includes:

[0017] Detecting whether there is duplicate data in the characteristic data of multiple users, where the duplicate data includes multiple first data, and when duplicate data exists, retaining only one first data in the duplicate data;

[0018] When abnormal data exists in the feature data of multiple users, the value of the abnormal data is replaced with the target value.

[0019] As an implementation mode, when each user corresponds to a plurality of sub-feature data, and each sub-feature data corresponds to a plurality of users, preprocessing is performed on the acquired feature data of the plurality of users, including:

[0020] When there are missing data in the i-th sub-feature data and it is determined that the type of the i-th sub-feature data is discrete, the missing data of the i-th sub-feature data is filled with the mode of the i-th sub-feature data; when there are missing data in the i-th sub-feature data and it is determined that the type of the i-th sub-feature data is non-discrete, the missing data of the i-th sub-feature data is filled with the mean of the i-th sub-feature data, where i is a positive integer;

[0021] Determine the total number of users corresponding to the i-th sub-feature data and the first number of users corresponding to the missing data of the i-th sub-feature data;

[0022] When the ratio of the first quantity to the total quantity is greater than a preset threshold, the i-th sub-feature data is deleted.

[0023] As an implementation mode, when abnormal data exists in the feature data of multiple users, replacing the value of the abnormal data with the target value includes:

[0024] When the value of the abnormal data is less than the minimum value of the preset range, the value of the abnormal data is replaced by the minimum value;

[0025] When the value of the abnormal data is greater than the maximum value of the preset range, the value of the abnormal data is replaced by the maximum value.

[0026] As an implementation mode, the pre-processed feature data of multiple users are screened according to the target feature to obtain target feature data of multiple users corresponding to the target feature, including:

[0027] Acquiring feature data of a plurality of second users, where the second users include users for whom actual classification results are known;

[0028] Determining the importance of each sub-feature data in the feature data of the plurality of second users;

[0029] Acquire sub-feature data of which importance is greater than a first threshold value from the feature data of the plurality of second users;

[0030] When the correlation coefficient of the sub-feature data whose importance is greater than the first threshold is greater than the second threshold, the sub-feature data except the sub-feature data with the greatest importance in the feature data of the plurality of second users are eliminated;

[0031] Determine a feature corresponding to the remaining sub-feature data among the plurality of sub-feature data whose importance is greater than the first threshold, and obtain a target feature;

[0032] Target feature data corresponding to the target feature is screened out from the feature data of multiple users.

[0033] As an implementation manner, determining the importance of each sub-feature data in the feature data of the plurality of second users includes:

[0034] For the j-th sub-feature data in the feature data of the second user, the following steps are performed:

[0035] Using multiple classifiers to classify the second user according to the j-th sub-feature data;

[0036] Determine the number of classifiers whose classification results are the same as or different from the actual classification results;

[0037] Determine a first error rate of the j-th sub-feature data according to the number of classifiers whose determined classification results are the same as or different from the actual classification results;

[0038] Adding noise interference to the j-th sub-feature data, and determining a second error rate after the j-th sub-feature data is added with the noise interference;

[0039] The importance of the j-th sub-feature data is determined according to the first error rate and the second error rate.

[0040] As an implementation method, determining a target quantile of each sub-target feature data from m quantiles based on a threshold search algorithm includes:

[0041] Determine the number of users who handle the target product when each sub-target characteristic data is greater than each quantile of each sub-target characteristic data, and calculate the ratio of the number of users who handle the target product determined each time to the total number of first users;

[0042] determining a maximum ratio from among the calculated multiple ratios;

[0043] Determine the quantile corresponding to the largest ratio and obtain the target quantile of each sub-target feature data.

[0044] As an implementation mode, according to each sub-target feature data and target quantile of the first user, a T-th level target user group is obtained, including:

[0045] Determine from the first users the users whose each sub-target feature data is greater than the target quantile of each sub-target feature data, and obtain the Tth target user group.

[0046] On the other hand, an embodiment of the present application provides a device for determining a target user group, the device comprising:

[0047] A first determination module is used to obtain a first-tier target user group including multiple target users according to target feature data of multiple users and a pre-established user group classification model, wherein the user group classification model is obtained by learning a mapping relationship between feature data of multiple sample users and target users;

[0048] The second determination module is used to obtain the T-th level target user group through the following steps, where T is a positive integer greater than 1:

[0049] Acquire target feature data of a first user, the target feature data including a plurality of sub-target feature data, the first user including users other than users in a first-tier target user group to a T-1-tier target user group among the plurality of users;

[0050] Calculate m quantiles of each sub-target feature data of the first user;

[0051] Determine the target quantile of each sub-target feature data from the m quantiles based on the threshold search algorithm;

[0052] According to each sub-target feature data and target quantile of the first user, the T-th level target user group is obtained.

[0053] In another aspect, an embodiment of the present application provides an electronic device, the device comprising: a processor and a memory storing computer program instructions;

[0054] When the processor executes the computer program instructions, the method for determining the target user group as described above is implemented.

[0055] On the other hand, an embodiment of the present application provides a computer storage medium having computer program instructions stored thereon, and the computer program instructions, when executed by a processor, implement the method for determining a target user group as described above.

[0056] The method, device, equipment and computer storage medium for determining the target user group of the embodiment of the present application can classify users based on the pre-established user group classification model and the input target feature data of multiple users, and obtain the first-tier target user group with the highest predicted promotion success rate; for the remaining users, the threshold search algorithm is used to traverse the quantiles of each sub-target feature data to obtain the optimal solution of the quantiles of each sub-target feature data - the target quantile, and according to each sub-target feature data and the target quantile, the T-tier target user group is obtained. Compared with the prior art, the target user group determined by the embodiment of the present application has the advantages of moderate scale and precise positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0058] Figure 1 is a flow chart of a method for determining a target user group provided in an embodiment of the present application;

[0059] Figure 2 A schematic diagram of a process for obtaining target feature data of multiple users according to an embodiment of the present application;

[0060] Figure 3 A processing flow of a method for determining a target user group according to an embodiment of the present application is schematically shown;

[0061] Figure 4 A schematic diagram of the hardware structure of a device for determining a target user group provided in an embodiment of the present application is shown;

[0062] Figure 5 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only configured to explain the present application and are not configured to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.

[0064] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0065] With the potential user base of operators approaching saturation and number portability being fully liberalized, competition among operators for market share is becoming increasingly fierce. Therefore, with limited resources, it is particularly important to accurately target target users with a high promotion success rate when expanding new services.

[0066] When screening or selecting target users, the existing technology usually sets the selection conditions based on the business experience of the personnel. When setting the selection conditions, usually only one or a limited number of features are used as the selection conditions. The conditions are wide, which makes the scale of the selected target users large. At the same time, it fails to combine other business behavior information of users for multi-dimensional mining, which is costly and inaccurate. When selecting target users, other existing methods based on machine learning algorithms only output information such as the target user's mobile phone number list, and cannot output the user's feature data, which is not conducive to the promotion personnel to understand the user's situation, resulting in a low promotion success rate.

[0067] In order to solve the problems in the prior art, embodiments of the present application provide a method, apparatus, device and computer storage medium for determining a target user group.

[0068] The technical concept of the embodiment of the present application is: based on a pre-established user group classification model and the input target feature data of multiple users, users are classified to obtain the first-tier target user group with the highest predicted promotion success rate; for the remaining users, the threshold search algorithm is used to traverse the quantiles of each sub-target feature data to obtain the optimal solution of the quantiles of each sub-target feature data - the target quantile, and according to each sub-target feature data and the target quantile, the T-th tier target user group is obtained.

[0069] First, it should be noted that in the method for determining the target user group provided in the embodiment of the present application, it is necessary to use a pre-established user group classification model to classify users to obtain a first-tier target user group. Therefore, before obtaining the first-tier target user group using the user group classification model, it is necessary to establish and train the user group classification model. Therefore, in order to facilitate the understanding of the embodiment of the present application, the establishment and training process of the user group classification model provided in the embodiment of the present application is first described below.

[0070] In order to make the first-tier target user group more accurate, as an example, the embodiment of the present application automatically optimizes model parameters and compares multiple machine learning algorithms, and autonomously selects the most effective machine learning algorithm and model parameters from multiple machine learning algorithms and numerous model parameters, and constructs a user group classification model based on the selected machine learning algorithm and model parameters, so that the constructed user group classification model is more consistent with the actual classification results, and the first-tier target user group is accurately obtained. In the embodiment of the present application, the multiple machine learning algorithms may include a decision tree algorithm, a logistic regression algorithm, a random forest algorithm, an adaptive boosting Adaboost algorithm, and an extreme gradient boosting Xgboost algorithm.

[0071] Specifically, first, for example, for each machine learning algorithm, a grid search is performed by setting the step size of the parameter optimization interval to obtain the optimal model parameters corresponding to each machine learning algorithm. Next, the current model parameters of each machine learning algorithm are set as the optimal model parameters corresponding to each machine learning algorithm. Next, the target feature data of multiple sample users are obtained, and each machine learning algorithm is trained with the target feature data of multiple sample users to learn the mapping relationship between the target feature data of multiple sample users and the target users. Next, based on the training results of each machine learning algorithm, the machine learning algorithm with the best training effect is determined. Finally, based on the determined machine learning algorithm and the optimal model parameters of the machine learning algorithm, a user group classification model is obtained.

[0072] In the embodiment of the present application, the target user refers to, for example, a user in the first-tier target user group.

[0073] After introducing the establishment and training process of the user group classification model, the method for determining the target user group provided in the embodiment of the present application is introduced below.

[0074] Figure 1 FIG. 2 is a flow chart showing a method for determining a target user group provided by an embodiment. Figure 1 As shown, the method may include the following steps:

[0075] S101: obtaining a first-tier target user group including multiple target users according to target feature data of multiple users and a pre-established user group classification model;

[0076] S102: Obtain the T-th level target user group through the following steps, where T is a positive integer greater than 1:

[0077] S102 may specifically include the following steps:

[0078] S102.1: Obtain target feature data of a first user, the target feature data including a plurality of sub-target feature data, the first user including users other than users in the first-tier target user group to the T-1-tier target user group among the plurality of users;

[0079] S102.2: Calculate m quantiles of each sub-target feature data of the first user;

[0080] S102.3: Determine the target quantile of each sub-target feature data from the m quantiles based on a threshold search algorithm;

[0081] S102.4: Obtain the T-th level target user group based on each sub-target feature data and target quantile of the first user.

[0082] The specific implementation of the above steps will be described in detail below.

[0083] The method for determining the target user group in the embodiment of the present application can classify users based on the pre-established user group classification model and the input target feature data of multiple users, and obtain the first-tier target user group with the highest predicted promotion success rate; for the remaining users, the threshold search algorithm is used to traverse the quantiles of each sub-target feature data to obtain the optimal solution of the quantiles of each sub-target feature data - the target quantile, and according to each sub-target feature data and the target quantile, the T-tier target user group is obtained. Compared with the prior art, the target user group determined in the embodiment of the present application has the advantages of moderate scale and precise positioning.

[0084] The specific implementation methods of the above steps are introduced below.

[0085] First, S101 is introduced: according to target feature data of multiple users and a pre-established user group classification model, a first-tier target user group including multiple target users is obtained.

[0086] Specifically, after obtaining the trained user group classification model, the target feature data of multiple users to be classified are input into the user group classification model. Since the user group classification model has trained the mapping relationship between the target feature data and the target users, after internal calculations of the user group classification model, the first-level target user group containing multiple target users can be output.

[0087] In an embodiment of the present application, the first-tier target user group may be the target user group with the highest expected promotion success rate, that is, compared with the subsequent second-tier target user group, third-tier target user group and other-tier target user groups, the users of the first-tier target user group have the highest expected promotion success rate.

[0088] The target characteristic data may be one or more of the characteristic data. To facilitate understanding of the characteristic data and target characteristic data of the embodiment of the present application, the characteristic data and target characteristic data of the embodiment of the present application are described below in conjunction with Table 1.

[0089] Table 1 schematically shows characteristic data and target characteristic data of multiple users in an embodiment of the present application.

[0090]

[0091] As shown in Table 1, the features may include, for example, age, gender, call minutes and text message sending volume, the feature data may be, for example, data corresponding to all features (such as age, gender, call minutes and text message sending volume), and the target feature data may be, for example, data corresponding to the target features (such as call minutes and text message sending volume), but the present invention is not limited to this.

[0092] As an example, the target feature data is obtained by:

[0093] In order to avoid the dimensionality disaster and save computing resources, as an example, the embodiment of the present application screens out the target features according to the OOB algorithm and variable correlation analysis, and then obtains the target feature data corresponding to the target features. Compared with the prior art, the target features objectively screened out by the embodiment of the present application are more in line with the user's preferences, and the target user group obtained based on the target feature data is more accurate.

[0094] Specifically, before S101, target feature data of multiple users are obtained through the following steps:

[0095] Step 1: Obtain feature data of multiple users;

[0096] The second step: screening the characteristic data of multiple users according to the target characteristics to obtain the target characteristic data of multiple users corresponding to the target characteristics.

[0097] Combine the following Figure 2 The process of acquiring target feature data in the embodiment of the present application is described in detail. Figure 2 FIG. 1 is a flow chart of obtaining target feature data of multiple users according to an embodiment of the present application. Figure 2 As shown, obtaining target feature data of multiple users may include the following steps:

[0098] S201: Acquire feature data of multiple second users, where the second users include users for whom actual classification results are known.

[0099] Specifically, in the integrated machine learning algorithm, there will always be a portion of data that does not participate in the model training in terms of probability. These data that do not participate in the model training are called out-of-bag data (OOB for short), also known as out-of-band data. In the embodiment of the present application, for example, feature screening is performed using feature data of multiple second users that do not participate in the user group classification model training, so as to obtain target features through subsequent steps.

[0100] S202: Determine the importance of each sub-feature data in the feature data of the plurality of second users.

[0101] Here, let's take the jth sub-feature data in the feature data of the second user as an example to illustrate S202, where j is a positive integer. It is easy to understand that the importance of each sub-feature data in the feature data of multiple second users can be calculated in the same way as the jth sub-feature data. In S202, for the jth sub-feature data, the following steps are performed:

[0102] First, multiple classifiers are used to classify the second user according to the jth sub-feature data. For example, the jth sub-feature data may be data corresponding to the number of call minutes, and the number of classifiers is flexibly set according to actual conditions, for example, 100. Here, for example, the second users are divided into two categories, A and B, and the actual classification result of each second user is known. For example, 100 classifiers are used to classify each second user according to the data corresponding to the number of call minutes of each second user.

[0103] Next, the number of classifiers whose classification results are the same as and different from the actual classification results is determined.

[0104] The classification results of each classifier are recorded, and the number of classifiers whose classification results are the same as the actual classification results and the number of classifiers whose classification results are different from the actual classification results are counted respectively.

[0105] Next, according to the number of classifiers whose determined classification results are the same as or different from the actual classification results, the first error rate of the j-th sub-feature data is determined;

[0106] Specifically, the expression of the first error rate is as follows:

[0107] error_rate1=n p / (n t +n p )

[0108] Where error_rate1 represents the first error rate of the jth sub-feature data; n p The number of classifiers whose classification results are different from the actual classification results; n t Indicates the number of classifiers whose classification results are the same as the actual classification results.

[0109] Next, noise interference is added to the j-th sub-feature data to determine a second error rate after the noise interference is added to the j-th sub-feature data.

[0110] Specifically, adding noise interference to the j-th sub-feature data can be understood as randomly changing the value of the j-th sub-feature data. After the value of the j-th sub-feature data is randomly changed, the value of the j-th sub-feature data that has been randomly changed is put into the classifier again for classification, and the second error rate error_rate2 after the j-th sub-feature data is added with noise interference is calculated in the same way as the first error rate is calculated.

[0111] Next, the importance of the j-th sub-feature data is determined according to the first error rate and the second error rate.

[0112] Specifically, the ratio of the difference between the first error rate and the second error rate to the total number of classifiers is calculated to obtain the importance of the j-th sub-feature data.

[0113] The expression of the importance of the j-th sub-feature data is as follows:

[0114] Imp=(error_rate1-error_rate2) / n

[0115] Among them, Imp represents the importance of the j-th sub-feature data, also known as the accuracy of the reduction of the j-th sub-feature data; n represents the total number of classifiers.

[0116] Continue to see Figure 2 After determining the importance of each sub-feature data in the feature data of the plurality of second users, executing S203: acquiring the sub-feature data whose importance is greater than a first threshold in the feature data of the plurality of second users.

[0117] The first threshold value can be flexibly set according to actual conditions, and the present invention is not limited thereto. For example, the importance of 10 sub-feature data is calculated in S202, and in S203, the importance of each sub-feature data is compared with the first threshold value, and the sub-feature data with an importance greater than or equal to the first threshold value is retained, and the sub-feature data with an importance less than the first threshold value is eliminated.

[0118] After the sub-feature data with importance greater than or equal to the first threshold is screened out by the above OOB algorithm, correlation screening is then performed to eliminate the sub-feature data with correlation coefficient greater than the second threshold.

[0119] S204: When the correlation coefficient of the sub-feature data with an importance greater than the first threshold is greater than the second threshold, the sub-feature data except the sub-feature data with the greatest importance among the feature data of the plurality of second users are eliminated.

[0120] Specifically, for sub-feature data with importance greater than the first threshold, first calculate the correlation coefficients between the sub-feature data, for example, calculate the correlation coefficient matrix between the sub-feature data. Then, set a feature empty set S f , to the characteristic empty set S f Each sub-feature data is added in turn until all sub-feature data with importance greater than the first threshold are added. In the process of adding each sub-feature data, when the newly added sub-feature data is equal to the feature empty set S f When the correlation coefficient of the sub-feature data added in is less than or equal to the second threshold, the newly added sub-feature data is retained in the feature empty set S f In; when the newly added sub-feature data and the feature empty set S f When the correlation coefficient of any sub-feature data (for example, sub-feature data a) added to the feature set is greater than the second threshold, the importance of the newly added sub-feature data is compared with that of the sub-feature data a. When the importance of the newly added sub-feature data is higher than that of the sub-feature data a, the newly added sub-feature data is retained in the feature empty set S. f Otherwise, the newly added sub-feature data is not retained in the feature empty set S f middle.

[0121] Continue to see Figure 2 After the sub-feature data are screened, S205 is executed: determining features corresponding to the remaining sub-feature data among the plurality of sub-feature data whose importance is greater than the first threshold, and obtaining target features.

[0122] For example, after filtering in S204, only two sub-feature data (such as data corresponding to the number of call minutes and the amount of text messages sent) remain, then S205 can determine the target feature based on the filtered sub-feature data, such as the number of call minutes and the amount of text messages sent.

[0123] S206: Filtering out target feature data corresponding to the target feature from the feature data of multiple users.

[0124] Here, the target features are taken as the number of call minutes and the amount of text messages sent. After obtaining the feature data of multiple users to be classified, for example, the data corresponding to the number of call minutes and the amount of text messages sent are filtered out.

[0125] The above is a specific implementation of S101. The following describes a specific implementation of S102.

[0126] Continue to see Figure 1 After obtaining the first-tier target user group through S101, in the next S102, for example, the second-tier target user group, the third-tier target user group to the N-tier target user group with expected promotion success rates decreasing step by step can be obtained in sequence. The specific number of N is flexibly set according to actual conditions, and the present invention is not limited to this.

[0127] Specifically, for each of the second to N-th target user groups, S102 may be described by taking the T-th target user group as an example, where T is a positive integer greater than 1 and less than or equal to N. It is easy to understand that each of the second to N-th target user groups may be obtained in the same manner as the T-th target user group.

[0128] Specifically, the target user group of the T-th level is obtained through the following steps:

[0129] S102.1: Obtain target feature data of a first user, where the first user includes users other than users in the first-tier target user group to the T-1th-tier target user group among a plurality of users.

[0130] In order to avoid the same user being assigned to multiple target user groups, when obtaining the T-th target user group, users of the previously assigned target user group are excluded. In S102.1, users other than the users of the first to T-1th target user groups are screened out from the initially obtained multiple users.

[0131] S102.2: Calculate m quantiles of each sub-target feature data of the first user. Taking the quartile as an example, the quantile refers to the values ​​at the three split point positions when all the values ​​of the sub-target feature data are arranged from small to large and divided into four equal parts. In the embodiment of the present application, the value of m can be, for example, 5 or other values, but the present invention is not limited thereto.

[0132] In S102.2, each sub-target feature data of the first user is represented by X1, X2, ..., XN For example, when m=5, the quantiles from the first sub-target feature data to the Nth sub-target feature data are, for example, (X 1Q1 ,X 1Q2 ,X 1Q3 ,X 1Q4 ,X 1Q5 ),(X 2Q1 ,X 2Q2 ,X 2Q3 ,X 2Q4 ,X 2Q5 ),……,(X NQ1 ,X NQ2 ,X NQ3 ,X NQ4 ,X NQ5 ).

[0133] The vector matrix of quantiles is:

[0134] S102.3: Determine the target quantile of each sub-target feature data from the m quantiles based on a threshold search algorithm.

[0135] Specifically, first determine the number of users who handle the target product when each sub-target characteristic data is greater than each quantile of each sub-target characteristic data, and calculate the ratio of the number of users who handle the target product determined each time to the total number of first users; then determine the largest ratio from the multiple calculated ratios; finally determine the quantile corresponding to the largest ratio to obtain the target quantile of each sub-target characteristic data.

[0136] In the embodiment of the present application, the ratio of the number of users who handle the target product to the total number of first users is called the handling rate. In S102.3, it is first determined that X1>X 1Qi , X2>X 2Qi , ..., X N >X NQi The processing rate at that time, where X 1Qi represents the i-th quantile of the first sub-feature data, X 2Qi represents the i-th quantile of the second sub-feature data, X NQi Represents the i-th quantile of the N-th sub-feature data.

[0137] For ease of understanding, S102.3 is described below in conjunction with Table 2. Table 2 schematically shows the quantiles of the characteristic data of the three sub-targets.

[0138] age Talk minutes / month SMS sent / month The 1st quantile 20 100 50 The second quantile 30 150 60 The third quantile 40 200 70 The 4th quantile 50 250 80 5th quantile 60 300 90

[0139] As shown in Table 2, assume that there are three sub-target characteristic data in S102.3, for example, data X1, X2 and X3 corresponding to age, number of call minutes and number of text messages sent, and each sub-target characteristic data has 5 quantiles.

[0140] In S102.3, first determine in sequence the processing rate when X1>20, X2>100, X3>50, the processing rate when X1>20, X2>100, X3>60, the processing rate when X1>20, X2>100, X3>70, ..., the processing rate when X1>60, X2>300, X3>80, and the processing rate when X1>60, X2>300, X3>90.

[0141] After calculating the handling rate of each quantile, determine the maximum handling rate from the calculated multiple handling rates. Finally, determine the quantile corresponding to the maximum handling rate, and obtain the target quantile of each sub-target feature data, denoted as (X 1Q ',X 2Q ',…,X NQ '), for example (20,100,70).

[0142] S102.4: Obtain the T-th level target user group based on each sub-target feature data and target quantile of the first user.

[0143] Specifically, determine from the first users the users whose feature data of each sub-target is greater than the target quantile of each sub-target feature data, that is, satisfying X1>X 1Q ', X2>X 2Q ', ..., X N >X NQ 'Condition users, get the Tth target user group.

[0144] As an example, the iteration number of step S102 may be set to output each level target user group in sequence. If the current iteration number is less than or equal to the set iteration number, S102 is executed, and if the current iteration number is greater than the set iteration number, the process ends.

[0145] In order to ensure the accuracy of the characteristic data of multiple users or the target characteristic data and prevent abnormal data in the characteristic data of multiple users from affecting the screening of the target user group, as an example, before executing step S101, the embodiment of the present application may also pre-process the characteristic data of multiple users, specifically including the following steps:

[0146] The acquired feature data of the multiple users are preprocessed to remove duplicate data and abnormal data beyond a preset range in the feature data of the multiple users, and to fill in missing data in the feature data of the multiple users.

[0147] Specifically, removing duplicate data from the feature data of multiple users includes: detecting whether there is duplicate data in the feature data of multiple users, the duplicate data includes multiple first data, and when duplicate data exists, retaining only one first data in the duplicate data. For example, based on the generation time of the multiple first data, retaining the first data among the multiple first data whose generation time is closest to the current time, that is, the latest first data.

[0148] Filling missing data in feature data of multiple users includes the following steps:

[0149] When there are missing data in the i-th sub-feature data and it is determined that the type of the i-th sub-feature data is discrete, the missing data of the i-th sub-feature data is filled with the mode of the i-th sub-feature data;

[0150] When there are missing data in the i-th sub-feature data and it is determined that the type of the i-th sub-feature data is non-discrete, the missing data of the i-th sub-feature data is filled with the mean value of the i-th sub-feature data.

[0151] For ease of understanding, the following describes how to fill in missing data in conjunction with Table 3. Table 3 shows a plurality of sub-feature data.

[0152]

[0153] As shown in Table 3, the second sub-feature data (data corresponding to gender) and the third sub-feature data (data corresponding to call minutes) have missing data, where the data in the second sub-feature data is male or female, and the data value is discontinuous, that is, the type of the second sub-feature data is discrete, and the missing data is filled with the mode of the second sub-feature data, that is, male. The type of the third sub-feature data is non-discrete, and the missing data is filled with the mean of the third sub-feature data, that is, (100+120+200+180) / 4=150.

[0154] In addition, preprocessing also includes deleting sub-feature data with serious missing data, including:

[0155] Determine the total number of users corresponding to the i-th sub-feature data and the first number of users corresponding to the missing data of the i-th sub-feature data; when the ratio of the first number to the total number is greater than a preset threshold, delete the i-th sub-feature data. The preset threshold can be flexibly set according to actual conditions, and the present invention is not limited thereto.

[0156] Processing of abnormal data beyond the preset range, including:

[0157] When the value of the abnormal data is less than the minimum value of the preset range, the value of the abnormal data is replaced by the minimum value;

[0158] When the value of the abnormal data is greater than the maximum value of the preset range, the value of the abnormal data is replaced by the maximum value.

[0159] As an example, the preset range may be [Q1-1.5(IQR), Q3+1.5(IQR)], where Q1 represents the 1 / 4 quantile of the i-th sub-feature data, Q3 represents the 3 / 4 quantile of the i-th sub-feature data, and IQR=Q3-Q1, representing a four-point difference.

[0160] Therefore, by removing duplicate data and abnormal data beyond a preset range in the feature data of multiple users and filling in missing data in the feature data of multiple users, the preprocessing process of the feature data of multiple users is completed.

[0161] Figure 3 A processing flow of a method for determining a target user group according to an embodiment of the present application is schematically shown. Figure 3 As shown, the embodiment of the present application combines a machine learning algorithm with a threshold search algorithm, and uses a machine learning algorithm or autonomous machine learning to obtain a first-tier target user group with the highest expected promotion success rate, and uses a threshold search algorithm to successively obtain a second-tier target user group, a third-tier target user group, to an N-tier target user group with gradually decreasing expected promotion success rates.

[0162] More specifically, after data cleaning (data preprocessing) of the feature data of multiple users, feature extraction is performed to extract the target feature data of multiple users. Then, a user group classification model is constructed using the optimal learning algorithm selected from the decision tree algorithm, the logistic regression algorithm, the random forest algorithm, the adaptive boosting Adaboost algorithm, and the extreme gradient boosting Xgboost algorithm and the optimal model parameters determined from multiple model parameters to obtain the first-tier target user group. For the remaining users, the target feature data (TopN variables) of the remaining users are obtained, and the m quantiles of each sub-target feature data of the remaining users are calculated in turn. According to the handling rate and based on the threshold search algorithm, the target quantile of each sub-target feature data is determined from the m quantiles, and the optimal threshold vector of the target quantile is obtained, for example, the above-mentioned (X 1Q ',X 2Q ',…,X NQ '); Finally, according to each sub-target feature data and target quantile of the first user, the T-th level target user group is obtained.

[0163] The embodiment of the present application combines a machine learning algorithm with a threshold search algorithm to sequentially obtain multiple step target user groups with decreasing expected promotion success rates, which can provide a basic reference for the reasonable configuration of subsequent business promotion.

[0164] For example, users in the first-tier target user group with a higher expected promotion success rate can be selected as the main promotion targets, and the preferential strength of the service can be appropriately reduced, and the promotion strategy can be formulated according to the target characteristic data of the users; for users in the first-tier target user group with a lower expected promotion success rate, the preferential strength of the service can be appropriately increased.

[0165] In order to verify the effect of the method for determining the target user group according to the embodiment of the present application, the inventor applied the method for determining the target user group to a business promotion activity.

[0166] The goal of this business promotion activity is to promote the Spring Festival traffic package to users. The Spring Festival traffic package was promoted to 3.82 million users by using the method of determining the target user group, including 500,000 users in the first-tier target user group, 1 million users in the second-tier target user group, and 2.32 million users in the third-tier target user group. The statistical results show that during the activity, a total of 3.81 million users were contacted and successfully promoted to 44,000 users, with a promotion success rate of 1.16%. The promotion success rates of the first to third-tier target user groups were 3.21%, 1.60%, and 0.53%, respectively. The promotion success rate of the first-tier target user group is 5.35 times the promotion success rate of 0.6% of the existing promotion method, and the average promotion success rate of the three-tier target user groups is twice that of the existing promotion method, indicating that the target user group determined by the embodiment of the present application is more accurate.

[0167] Based on the method for determining the target user group provided in the above embodiment, the present application also provides a specific implementation of the device for determining the target user group. Please refer to the following embodiment.

[0168] Figure 4 FIG. 1 is a schematic diagram showing the hardware structure of the device for determining the target user group provided in an embodiment of the present application. Figure 4 As shown, the device 400 for determining a target user group provided in an embodiment of the present application may include:

[0169] A first determination module 401 is used to obtain a first-tier target user group including multiple target users according to target feature data of multiple users and a pre-established user group classification model, wherein the user group classification model is obtained by learning a mapping relationship between feature data of multiple sample users and target users;

[0170] The second determination module 402 is used to obtain the T-th level target user group through the following steps, where T is a positive integer greater than 1:

[0171] Acquire target feature data of a first user, the target feature data including a plurality of sub-target feature data, the first user including users other than users in a first-tier target user group to a T-1-tier target user group among the plurality of users;

[0172] Calculate m quantiles of each sub-target feature data of the first user;

[0173] Determine the target quantile of each sub-target feature data from the m quantiles based on the threshold search algorithm;

[0174] According to each sub-target feature data and target quantile of the first user, the T-th level target user group is obtained.

[0175] As an implementation manner, the device 400 for determining a target user group may further include a preprocessing module, which is used to perform the following steps before obtaining a first-stage target user group including a plurality of target users according to the target feature data of a plurality of users and a pre-established user group classification model:

[0176] Obtain characteristic data of multiple users;

[0177] Preprocessing the acquired feature data of the multiple users to remove duplicate data and abnormal data beyond a preset range in the feature data of the multiple users, and filling in missing data in the feature data of the multiple users;

[0178] The preprocessed feature data of multiple users are screened according to the target feature to obtain target feature data of multiple users corresponding to the target feature.

[0179] As an implementation mode, the preprocessing module is specifically used to: detect whether there is duplicate data in the feature data of multiple users, the duplicate data includes multiple first data, and when duplicate data exists, only retain one first data in the duplicate data; when there is abnormal data in the feature data of multiple users, replace the value of the abnormal data with the target value.

[0180] As an implementation manner, the preprocessing module is further specifically used for: when each user corresponds to a plurality of sub-feature data, and each sub-feature data corresponds to a plurality of users, performing the following steps:

[0181] When there are missing data in the i-th sub-feature data and it is determined that the type of the i-th sub-feature data is discrete, the missing data of the i-th sub-feature data is filled with the mode of the i-th sub-feature data; when there are missing data in the i-th sub-feature data and it is determined that the type of the i-th sub-feature data is non-discrete, the missing data of the i-th sub-feature data is filled with the mean of the i-th sub-feature data, where i is a positive integer;

[0182] Determine the total number of users corresponding to the i-th sub-feature data and the first number of users corresponding to the missing data of the i-th sub-feature data;

[0183] When the ratio of the first quantity to the total quantity is greater than a preset threshold, the i-th sub-feature data is deleted.

[0184] As an implementation method, the preprocessing module is specifically used for:

[0185] When the value of the abnormal data is less than the minimum value of the preset range, the value of the abnormal data is replaced by the minimum value;

[0186] When the value of the abnormal data is greater than the maximum value of the preset range, the value of the abnormal data is replaced by the maximum value.

[0187] As an implementation mode, the preprocessing module is specifically used to: obtain feature data of multiple second users, where the second users include users whose actual classification results are known; determine the importance of each sub-feature data in the feature data of multiple second users; obtain sub-feature data with an importance greater than a first threshold in the feature data of multiple second users; when the correlation coefficient of the sub-feature data with an importance greater than the first threshold is greater than a second threshold, eliminate the sub-feature data except the sub-feature data with the greatest importance in the feature data of multiple second users; determine the feature corresponding to the remaining sub-feature data in the multiple sub-feature data with an importance greater than the first threshold, and obtain the target feature; and filter out the target feature data corresponding to the target feature from the feature data of multiple users.

[0188] As an implementation manner, determining the importance of each sub-feature data in the feature data of the plurality of second users includes:

[0189] For the j-th sub-feature data in the feature data of the second user, the following steps are performed:

[0190] Using multiple classifiers to classify the second user according to the j-th sub-feature data;

[0191] Determine the number of classifiers whose classification results are the same as or different from the actual classification results;

[0192] Determine a first error rate of the j-th sub-feature data according to the number of classifiers whose determined classification results are the same as or different from the actual classification results;

[0193] Adding noise interference to the j-th sub-feature data, and determining a second error rate after the j-th sub-feature data is added with the noise interference;

[0194] The importance of the j-th sub-feature data is determined according to the first error rate and the second error rate.

[0195] As an implementation manner, the second determining module 402 is specifically configured to:

[0196] Determine the number of users who handle the target product when each sub-target characteristic data is greater than each quantile of each sub-target characteristic data, and calculate the ratio of the number of users who handle the target product determined each time to the total number of first users;

[0197] determining a maximum ratio from among the calculated multiple ratios;

[0198] Determine the quantile corresponding to the largest ratio and obtain the target quantile of each sub-target feature data.

[0199] As an implementation manner, the second determining module 402 is specifically configured to:

[0200] Determine from the first users the users whose each sub-target feature data is greater than the target quantile of each sub-target feature data, and obtain the Tth target user group.

[0201] Figure 4 Each module / unit in the device shown has the function of realizing Figure 1 The functions of each step in the process can achieve the corresponding technical effects, which will not be described in detail here for the sake of brevity.

[0202] Based on the method for determining the target user group provided in the above embodiment, the present application also provides a specific implementation of the electronic device accordingly. Please refer to the following embodiment.

[0203] Figure 5 A schematic diagram of the electronic hardware structure provided in an embodiment of the present application is shown.

[0204] like Figure 5 As shown, the electronic device may include a processor 501 and a memory 502 storing computer program instructions.

[0205] Specifically, the processor 501 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0206] The memory 502 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 502 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, the memory 502 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 502 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 502 is a non-volatile solid-state memory. In a specific embodiment, the memory 502 includes a read-only memory (ROM). In appropriate cases, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM) or a flash memory or a combination of two or more of these.

[0207] The processor 501 implements any one of the methods for determining a target user group in the above embodiments by reading and executing computer program instructions stored in the memory 502 .

[0208] In one example, the electronic device may further include a communication interface 503 and a bus 510. Figure 5 As shown, the processor 501, the memory 502, and the communication interface 503 are connected via a bus 510 and communicate with each other.

[0209] The communication interface 503 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0210] Bus 510 includes hardware, software or both, and the parts of online data flow billing equipment are coupled to each other. For example, but not limitation, bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front-end bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 510 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the present application considers any suitable bus or interconnection.

[0211] In addition, in combination with the method for determining the target user group in the above embodiment, the embodiment of the present application can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any one of the methods for determining the target user group in the above embodiment is implemented.

[0212] In summary, the method, device, equipment and computer storage medium for determining the target user group in the embodiment of the present application can classify users based on the pre-established user group classification model and the input target feature data of multiple users, and obtain the first-tier target user group with the highest predicted promotion success rate; for the remaining users, the threshold search algorithm is used to traverse the quantiles of each sub-target feature data to obtain the optimal solution of the quantiles of each sub-target feature data - the target quantile, and according to each sub-target feature data and the target quantile, the T-tier target user group is obtained. Compared with the prior art, the target user group determined by the embodiment of the present application has the advantages of moderate scale and precise positioning.

[0213] In addition, the embodiments of the present application can output visualized target feature data and screening conditions for screening target user groups, making it easier for marketing personnel to understand user behavior habits and solve the problems of vague user features and poor business explanation.

[0214] The embodiment of the present application introduces recursive thinking to obtain target user groups at each level with expected promotion success rates decreasing step by step in sequence, which can solve the problem of flat preferences of existing user groups and can flexibly adjust the promotion scale by setting the number of recursive iterations.

[0215] The embodiment of the present application divides the target user groups into different levels with decreasing expected promotion success rates. According to the characteristics of decreasing business needs of users at different levels, the main promotion objects can be set to avoid disturbing users with lower expected promotion success rates, thereby improving user experience.

[0216] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0217] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0218] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.

[0219] Aspects of the present disclosure are described above with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0220] The above is only a specific implementation of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.

Claims

1. A method for determining a target user group, characterized in that: include: Screening the characteristic data of multiple users according to the target characteristic to obtain the target characteristic data of multiple users corresponding to the target characteristic; According to target feature data of multiple users and a pre-established user group classification model, a first-tier target user group including multiple target users is obtained, wherein the user group classification model is obtained by learning a mapping relationship between target feature data of multiple sample users and the target users; The target user group of the T-th stage is obtained by the following steps, where T is a positive integer greater than 1: Acquire target feature data of a first user, the target feature data including a plurality of sub-target feature data, the first user including users other than users of the first-tier target user group to the T-1-tier target user group among the plurality of users; Calculate m quantiles of each sub-target feature data of the first user, where m is a positive integer; Determine a target quantile of each sub-target feature data from the m quantiles based on a threshold search algorithm, wherein the target quantile is determined based on a quantile corresponding to a maximum value among a plurality of ratios, wherein the plurality of ratios include a ratio of the number of users determined each time to the total number of the first users, and the number of users determined each time includes the number of users who handle the target product when each sub-target feature data is greater than each quantile of each sub-target feature data; Obtaining the T-th level target user group according to each sub-target feature data of the first user and the target quantile; The step of screening the feature data of multiple users according to the target feature to obtain the target feature data of multiple users corresponding to the target feature includes: obtaining feature data of multiple second users, wherein the second users include users with known actual classification results; determining the importance of each sub-feature data in the feature data of the multiple second users; determining the target feature according to the importance of each sub-feature data; and screening the target feature data corresponding to the target feature from the feature data of the multiple second users; The determining the importance of each sub-feature data in the feature data of the plurality of second users comprises: for the j-th sub-feature data in the feature data of the second user, where j is a positive integer, performing the following steps: classifying the second user according to the j-th sub-feature data using a plurality of classifiers; determining the number of classifiers whose classification results are the same as or different from the actual classification results; determining a first error rate of the j-th sub-feature data according to the number of classifiers whose determined classification results are the same as or different from the actual classification results; adding noise interference to the j-th sub-feature data, and determining a second error rate of the j-th sub-feature data after the noise interference is added; and determining the importance of the j-th sub-feature data according to the first error rate and the second error rate.

2. The method according to claim 1, characterized in that Before obtaining a first-tier target user group including a plurality of target users according to the target feature data of the plurality of users and a pre-established user group classification model, the method further comprises: Acquiring characteristic data of the multiple users; Preprocessing the acquired feature data of the multiple users to remove duplicate data and abnormal data beyond a preset range in the feature data of the multiple users, and filling in missing data in the feature data of the multiple users; The step of screening the characteristic data of multiple users according to the target characteristic to obtain the target characteristic data of multiple users corresponding to the target characteristic includes: The preprocessed feature data of the multiple users are screened according to the target feature to obtain the target feature data of the multiple users corresponding to the target feature.

3. The method according to claim 2, characterized in that The preprocessing of the acquired characteristic data of the multiple users specifically includes: Detecting whether the duplicate data exists in the characteristic data of the multiple users, the duplicate data including multiple first data, and when the duplicate data exists, retaining only one of the first data in the duplicate data; When the abnormal data exists in the feature data of the plurality of users, the value of the abnormal data is replaced with a target value.

4. The method according to claim 3, characterized in that When each user corresponds to a plurality of sub-feature data, and each sub-feature data corresponds to a plurality of users, preprocessing the acquired feature data of the plurality of users includes: When there are missing data in the i-th sub-feature data and it is determined that the type of the i-th sub-feature data is discrete, the missing data of the i-th sub-feature data is filled with the mode of the i-th sub-feature data; when there are missing data in the i-th sub-feature data and it is determined that the type of the i-th sub-feature data is non-discrete, the missing data of the i-th sub-feature data is filled with the mean of the i-th sub-feature data, where i is a positive integer; Determine the total number of users corresponding to the i-th sub-feature data and the first number of users corresponding to the missing data of the i-th sub-feature data; When the ratio of the first number to the total number is greater than a preset threshold, the i-th sub-feature data is deleted.

5. The method according to claim 3, characterized in that: When the abnormal data exists in the characteristic data of the plurality of users, replacing the value of the abnormal data with the target value comprises: When the value of the abnormal data is less than the minimum value of the preset range, replacing the value of the abnormal data with the minimum value; When the value of the abnormal data is greater than the maximum value of the preset range, the value of the abnormal data is replaced by the maximum value.

6. The method according to claim 1, characterized in that Determining the target feature according to the importance of each sub-feature data includes: Acquire sub-feature data of which importance is greater than a first threshold value from the feature data of the plurality of second users; When the correlation coefficient of the sub-feature data whose importance is greater than the first threshold is greater than the second threshold, the sub-feature data except the sub-feature data with the greatest importance in the feature data of the plurality of second users are eliminated; Determine a feature corresponding to the remaining sub-feature data among the plurality of sub-feature data whose importance is greater than the first threshold, and obtain the target feature.

7. The method according to claim 1, characterized in that The determining the target quantile of each sub-target feature data from the m quantiles based on a threshold search algorithm comprises: Determine the number of users who handle the target product when each sub-target characteristic data is greater than each quantile of each sub-target characteristic data, and calculate the ratio of the number of users who handle the target product determined each time to the total number of the first users; determining a maximum ratio from among the calculated multiple ratios; The quantile corresponding to the maximum ratio is determined to obtain the target quantile of each sub-target feature data.

8. The method according to claim 1, characterized in that The step of obtaining the T-th level target user group according to each sub-target feature data of the first user and the target quantile includes: Determine from the first users the users whose each sub-target feature data is greater than the target quantile of each sub-target feature data to obtain the T-th step target user group.

9. A device for determining a target user group, characterized in that: include: A first determination module is used to obtain a first-tier target user group including multiple target users according to target feature data of multiple users and a pre-established user group classification model, wherein the user group classification model is obtained by learning a mapping relationship between feature data of multiple sample users and the target users; The second determination module is used to obtain the T-th level target user group through the following steps, where T is a positive integer greater than 1: Acquire target feature data of a first user, the target feature data including a plurality of sub-target feature data, the first user including users other than users of the first-tier target user group to the T-1-tier target user group among the plurality of users; Calculate m quantiles of each sub-target feature data of the first user; Determine a target quantile of each sub-target feature data from the m quantiles based on a threshold search algorithm, wherein the target quantile is determined based on a quantile corresponding to a maximum value among a plurality of ratios, wherein the plurality of ratios include a ratio of the number of users determined each time to the total number of the first users, and the number of users determined each time includes the number of users who handle the target product when each sub-target feature data is greater than each quantile of each sub-target feature data; Obtaining the T-th level target user group according to each sub-target feature data of the first user and the target quantile; The first determination module is further used to screen the feature data of multiple users according to the target feature to obtain target feature data of multiple users corresponding to the target feature; The first determination module is specifically used to obtain feature data of multiple second users, where the second users include users whose actual classification results are known; determine the importance of each sub-feature data in the feature data of the multiple second users; and determine the target feature according to the importance of each sub-feature data; Filtering the target feature data corresponding to the target feature from the feature data of the plurality of second users; The first determination module is further specifically used to perform the following steps for the j-th sub-feature data in the feature data of the second user, where j is a positive integer: classify the second user according to the j-th sub-feature data using multiple classifiers; determine the number of classifiers whose classification results are the same as or different from the actual classification results; determine a first error rate of the j-th sub-feature data according to the number of classifiers whose determined classification results are the same as or different from the actual classification results; add noise interference to the j-th sub-feature data to determine a second error rate of the j-th sub-feature data after the noise interference is added; and determine the importance of the j-th sub-feature data according to the first error rate and the second error rate.

10. An electronic device, characterized in that: The device comprises: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the method for determining a target user group according to any one of claims 1 to 8 is implemented.

11. A computer storage medium, characterized in that: The computer storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the method for determining a target user group according to any one of claims 1 to 8 is implemented.

Images