User group division method and apparatus, device, and storage medium

Abstract

The present application relates to the field of data processing, and discloses a user group division method and apparatus, a device, and a storage medium. The method comprises: acquiring user characteristics and value characteristics of a plurality of users, wherein the value characteristics are used for representing the value of the users to a service; clustering the plurality of users on the basis of the user characteristics and N groups of clustering hyper-parameters to obtain N user group division results, wherein one user group division result corresponds to one group of clustering hyper-parameters, and the user group division results comprise user groups obtained by clustering division; on the basis of the user characteristics and the value characteristics of the users in the user groups in the user group division results, obtaining by calculation a first discrimination index and a second discrimination index corresponding to each user group division result; and on the basis of the first discrimination index and the second discrimination index, selecting a user group division result having the highest comprehensive discrimination from among the N user group division results as a target user group division result.

Description

User group division method, device, equipment and storage medium

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to Chinese patent application 202311022391.2, filed on August 14, 2023, entitled “User group segmentation method, apparatus, device and storage medium,” and the entire contents of that application are incorporated herein by reference. Technical Field

[0003] The present application relates to the field of data processing, and in particular to a method, apparatus, device, and storage medium for dividing user groups. Background Art

[0004] With the advancement of business digitalization, various types of businesses are gradually shifting towards refinement and intelligence. In order to facilitate the development of different types of refined services for users, it is necessary to divide users into different user groups and provide adaptive refined services for each user group, thereby improving the user experience of the business.

[0005] For users, both their user characteristics and their user value to the business are important factors in segmenting user groups. Currently, value and user characteristic labels can be generated, and then combined to achieve user group segmentation using a pre-set, strong rule-based approach. However, this strong rule-based approach struggles to balance both the value and user characteristic dimensions, resulting in segmented user groups with better differentiation in the value dimension but poorer differentiation in the user characteristic dimension. Alternatively, segmented user groups may have better differentiation in the user characteristic dimension but poorer differentiation in the value dimension, leading to poor differentiation in the segmented user groups.

[0006] Summary of the Invention

[0007] The embodiments of the present application provide a method, apparatus, device, and storage medium for dividing user groups, which can improve the differentiation of divided user groups.

[0008] In a first aspect, an embodiment of the present application provides a method for user group segmentation, comprising: obtaining user characteristics and value characteristics of multiple users, where the value characteristics are used to characterize the value of the user to the business; clustering the multiple users based on the user characteristics and N preset groups of clustering hyperparameters to obtain N types of user group segmentation results, where one user group segmentation result corresponds to one group of clustering hyperparameters, and the user group segmentation result includes the user group obtained by clustering segmentation, where N is an integer greater than or equal to 2; calculating a first discrimination index and a second discrimination index corresponding to each user group segmentation result based on the user characteristics and value characteristics of the users in the user group in the user group segmentation result; selecting the user group segmentation result with the highest comprehensive discrimination from the N types of user group segmentation results as the target user group segmentation result based on the first discrimination index and the second discrimination index.

[0009] In a second aspect, an embodiment of the present application provides a user group segmentation device, comprising: a feature acquisition module for acquiring user features and value features of multiple users, where the value features are used to characterize the value of users to the business; a clustering module for clustering multiple users based on user features and N preset groups of clustering hyperparameters to obtain N types of user group segmentation results, where one user group segmentation result corresponds to one group of clustering hyperparameters, and the user group segmentation result includes the user group obtained by clustering segmentation, where N is an integer greater than or equal to 2; an indicator determination module for calculating a first discrimination index and a second discrimination index corresponding to each user group segmentation result based on the user features and value features of the users in the user group in the user group segmentation result; a user group determination module for selecting the user group segmentation result with the highest comprehensive discrimination as the target user group segmentation result from the N types of user group segmentation results based on the first discrimination index and the second discrimination index.

[0010] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the user group division method of the first aspect is implemented.

[0011] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the user group segmentation method of the first aspect is implemented.

[0012] The embodiments of the present application provide a method, apparatus, device and storage medium for user group segmentation, which can cluster users based on their user characteristics and multiple sets of clustering hyperparameters, and obtain multiple user group segmentation results corresponding to the multiple sets of clustering hyperparameters. According to the user characteristics and value characteristics of the users in the user group of the user group segmentation result, a first discrimination index that can characterize the discrimination of the user characteristic dimension and a second discrimination index that can characterize the discrimination of the value characteristic dimension are obtained. According to the first discrimination index and the second discrimination index, the comprehensive discrimination of the user group in each user group segmentation result can be comprehensively evaluated from the two aspects of the user characteristic dimension and the value characteristic dimension. The user group in the user group segmentation result with the highest comprehensive discrimination is a user group that takes into account both the discrimination of the user characteristic dimension and the discrimination of the value characteristic dimension, and the discrimination of the user characteristic dimension and the discrimination of the value characteristic dimension are both better, thereby improving the discrimination of the user group obtained by segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions 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.

[0014] FIG1 is a flow chart of a method for dividing user groups provided in one embodiment of the present application;

[0015] FIG2 is a schematic diagram of an example of the distribution of user groups in terms of user feature dimensions provided by an embodiment of the present application;

[0016] FIG3 is a schematic diagram of an example of the distribution of user groups in the value feature dimension provided by an embodiment of the present application;

[0017] FIG4 is a flow chart of a method for dividing user groups provided by another embodiment of the present application;

[0018] FIG5 is a flowchart of a method for dividing user groups provided in another embodiment of the present application;

[0019] FIG6 is a schematic structural diagram of a device for dividing user groups according to an embodiment of the present application;

[0020] FIG7 is a schematic structural diagram of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0021] 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 intended to explain the present application, rather than 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 merely to provide a better understanding of the present application by illustrating examples of the present application. It should be noted that the acquisition, storage, use, processing, etc. of information and data in the embodiments of the present application are authorized by the user or relevant agencies and comply with the relevant provisions of national laws and regulations.

[0022] With the advancement of business digitalization, various types of businesses are gradually shifting towards refinement and intelligence. In order to facilitate the development of different types of refined services for users, it is necessary to divide users into different user groups and provide adapted refined services for each user group, thereby improving the user experience of the service. For users, both the user characteristics and the user value of the user to the service are important factors in dividing user groups. Currently, it is possible to form user value feature labels and user feature labels, and divide user groups layer by layer according to the strong rules of value feature labels and user feature labels. For example, users can be divided into three user groups based on value feature labels. Then, each user group can be divided into three labels based on a user feature label. And so on, the user group division dimensions are continuously increased to obtain user groups. However, when there are too many dimensions of user characteristics and value characteristics, an excessive number of user groups will be formed. This excessive number of user groups will defeat the purpose of user group segmentation and will not balance the discrimination of both the user characteristic dimension and the value characteristic dimension. When there are fewer dimensions of user characteristics and value characteristics, user group segmentation will only have good discrimination performance in terms of discrimination of some user characteristic dimensions or some value characteristic dimensions, failing to balance the discrimination of both the user characteristic dimension and the value characteristic dimension, resulting in poor discrimination of the divided user groups. In some cases, value characteristics and user characteristics can be combined and clustered according to the combined value characteristics and user characteristics to divide user groups. However, since value characteristics generally have fewer dimensions, such as a single dimension, the impact of the value characteristic dimension on the distance calculation structure in the clustering algorithm is very small, making it impossible to ensure that the divided user groups have good discrimination in terms of the value characteristic dimension. It is difficult to balance the discrimination of both the user characteristic dimension and the value characteristic dimension, resulting in poor discrimination of the divided user groups.

[0023] The present application provides a method, apparatus, device and storage medium for user group segmentation, which utilizes multiple sets of clustering hyperparameters to cluster users based on user features with relatively more dimensions, and each set of clustering hyperparameters can correspond to a user group segmentation result. The user groups in each user group segmentation result are comprehensively evaluated from two aspects: the discrimination index of the user feature dimension and the discrimination index of the user value feature dimension, and the comprehensive discrimination of each user group segmentation structure in the user feature dimension and the value feature dimension is obtained. The user group segmentation result with the highest comprehensive discrimination is used as the target user group segmentation result for use. The comprehensive discrimination comprehensively considers the discrimination of the user feature dimension and the discrimination of the value feature dimension. Therefore, the target user group segmentation structure can take into account the discrimination of the user feature dimension and the discrimination of the value feature dimension, thereby improving the discrimination of the user group obtained by segmentation, and further improving the adaptability of the user group obtained by segmentation to the business.

[0024] In a first aspect, the present application provides a method for segmenting user groups, which can be applied to user group segmentation scenarios that require consideration of both user characteristics and user value. The method can be performed by a user group segmentation device, apparatus, or the like, and is not limited thereto. FIG1 is a flowchart of the method for segmenting user groups provided in one embodiment of the present application. As shown in FIG1 , the method for segmenting user groups can include steps S101 to S104.

[0025] In step S101 , user characteristics and value characteristics of multiple users are obtained.

[0026] The multiple users here are the users to be classified. User features can be used to characterize the attributes of the users themselves. In some examples, user features may include, but are not limited to, one or more of the following: user demographic attributes, user resource attributes, user risk attributes, user transaction habit attributes, and user resource channel attributes. User demographic attributes characterize the user's demographic attributes. For example, user demographic attributes may include, but are not limited to, one or more of the user's age, gender, and location. User resource attributes characterize the user's resources. For example, user resource attributes may include, but are not limited to, one or more of the user's account balance, real estate owned by the user, and vehicles owned by the user. User risk attributes characterize the user's transaction risk. For example, user risk attributes may include, but are not limited to, one or more of the user's loan amount and risky transactions. User transaction habit attributes characterize the user's transaction habits. For example, user transaction habit attributes may include, but are not limited to, one or more of the transaction time, transaction amount, and transaction scenario. User resource channel attributes characterize information related to the user's resource channels. For example, user resource channel attributes may include, but are not limited to, one or more of the user's resource card holdings and resource card level.

[0027] Value features are used to represent the value of a user to a business and can be defined and selected based on business requirements. A feature may be a value feature for one business, but not for another. For example, for some transaction businesses, the total transaction amount may be used as the user's value feature; for other transaction businesses, the total installment amount may be used as the user's value feature.

[0028] In some examples, the dimensionality of user features can be higher than that of value features. The dimensionality of user features can be reflected by the number of items included in the user features, while the dimensionality of value features can be reflected by the number of items included in the value features. For example, if user features include user age, gender, location, account balance, loan amount, transaction time, transaction amount, transaction scenario, and the number of resource cards held by the user, while the value features include the user's total transaction amount, the dimensionality of user features is higher than that of value features.

[0029] In step S102, multiple users are clustered based on user characteristics and N preset groups of clustering hyperparameters to obtain N types of user group division results.

[0030] N is an integer greater than or equal to 2. Multiple sets of clustering hyperparameters can be selected, and each set of clustering hyperparameters can be used to cluster multiple users based on user characteristics. The user group division results obtained by clustering with different clustering hyperparameters are different, and one user group division result corresponds to a set of clustering hyperparameters. The clustering algorithm used to cluster multiple users is not limited here. For example, the clustering algorithm used may include but is not limited to the distance-based K-Means clustering algorithm, the density-based DBSCAN (i.e., Density-Based Spatial Clustering of Applications with Noise) algorithm, etc. The type of clustering hyperparameter is related to the clustering algorithm. The clustering hyperparameters may include but are not limited to the number of clusters, the density of clusters, the distance to distinguish clusters, etc., where the clusters are user groups.

[0031] The user grouping results obtained by clustering include user groups obtained by clustering. The N user grouping results corresponding to N sets of clustering hyperparameters may result in different user groups. For example, if the clustering hyperparameters include the number of clusters, the number of user groups obtained by clustering may vary among the N user grouping results. The relatively optimal user grouping result is selected from the N user grouping results obtained by clustering.

[0032] In step S103, a first discrimination index and a second discrimination index corresponding to each user group division result are calculated based on the user characteristics and value characteristics of the users in the user groups in the user group division result.

[0033] A first discrimination index corresponding to each user group segmentation result can be calculated based on the user characteristics of the users in the user groups in the user group segmentation results. The first discrimination index can be used to characterize the discrimination of the user characteristic dimension and can be considered an indicator for evaluating the discrimination of the user characteristics. Specifically, based on the user characteristics of the users in the user groups in the user group segmentation results, the degree of similarity in user characteristics between each two user groups in the user group segmentation results can be calculated. Based on the degree of similarity in user characteristics between each two user groups in the user group segmentation results, the first discrimination index corresponding to the user group segmentation results can be evaluated.

[0034] The second discrimination index corresponding to each user group segmentation result can be calculated based on the value characteristics of the users in the user groups in the user group segmentation results. The second discrimination index can be used to characterize the discrimination of the value characteristic dimension and can be regarded as an evaluation index for the discrimination of the value characteristics. Specifically, based on the value characteristics of the users in the user groups in the user group segmentation results, the similarity in value characteristics between each two user groups in the user group segmentation results can be obtained. Based on the similarity in value characteristics between each two user groups in the user group segmentation results, the second discrimination index corresponding to the user group segmentation results can be evaluated.

[0035] In step S104, based on the first discrimination index and the second discrimination index, the user group classification result with the highest comprehensive discrimination is selected from the N user group classification results as the target user group classification result.

[0036] The first discrimination index can characterize the discrimination of the user groups in the user group segmentation results in terms of the user feature dimension, and the second discrimination index can characterize the discrimination of the user groups in the user group segmentation results in terms of the value feature dimension. The fusion of the first discrimination index and the second discrimination index can yield a comprehensive discrimination, which is obtained by fusion of the discrimination in terms of the user feature dimension and the discrimination in terms of the value feature dimension. The user group in the user group segmentation result with the highest comprehensive discrimination among the N types of user group segmentation results is the user group that can take into account both the discrimination in terms of the user feature dimension and the discrimination in terms of the value feature dimension, and has better discrimination in terms of both the user feature dimension and the value feature dimension.

[0037] In an embodiment of the present application, users can be clustered based on their user characteristics and multiple sets of clustering hyperparameters to obtain a variety of user group division results corresponding to the multiple sets of clustering hyperparameters. According to the user characteristics and value characteristics of the users in the user group of the user group division result, a first discrimination index that can characterize the discrimination of the user characteristic dimension and a second discrimination index that can characterize the discrimination of the value characteristic dimension are obtained. According to the first discrimination index and the second discrimination index, the comprehensive discrimination of the user group in each user group division result can be comprehensively evaluated from the two aspects of the user characteristic dimension and the value characteristic dimension. The user group in the user group division result with the highest comprehensive discrimination is a user group that takes into account both the discrimination of the user characteristic dimension and the discrimination of the value characteristic dimension, and the discrimination of the user characteristic dimension and the discrimination of the value characteristic dimension are both better, thereby improving the discrimination of the user group obtained by division and the adaptability of the user group to the business.

[0038] The discrimination degree of the user feature dimension and the discrimination degree of the value feature dimension of the user group obtained by the user group segmentation method in the embodiment of the present application are both good. For example, Figure 2 is a schematic diagram of an example of the distribution of the user group provided in the embodiment of the present application in the user feature dimension. Figure 2 uses the t-SNE model to reduce the dimensionality of the multi-dimensional user features of the user group so that it can show the visualization effect after dimensionality reduction. As shown in Figure 2, the user features of the 10 user groups (numbered 0 to 9) obtained by the user group segmentation method in the embodiment of the present application are clearly defined from each other, indicating that the discrimination degree of these 10 user groups in the user feature dimension is good. Figure 3 is a schematic diagram of an example of the distribution of the user group provided in the embodiment of the present application in the value feature dimension. The user group segmentation in Figure 3 is consistent with the user group segmentation shown in Figure 2. As shown in Figure 3, the value feature distribution of the 10 user groups (numbered 0 to 9) obtained by the user group segmentation method in the embodiment of the present application is relatively dispersed, indicating that the discrimination degree of these 10 user groups in the value feature dimension is good. From Figures 2 and 3 , it can be seen that the user groups in the target user group segmentation results obtained by the user group segmentation method in the embodiment of the present application can take into account both the differentiation in the user feature dimension and the differentiation in the value feature dimension, thereby providing support for the refined development of user services.

[0039] In some embodiments, the first discrimination index and the second discrimination index can be obtained by aggregating the user characteristics of the users in the user groups of the user group segmentation results, and the aggregation of the value characteristics of the users in the user groups of the user group segmentation results. Thus, based on the first discrimination index and the second discrimination index, a weighted algorithm is used to obtain a comprehensive discrimination evaluation index that can characterize the comprehensive discrimination. Figure 4 is a flowchart of the user group segmentation method provided by another embodiment of the present application. The difference between Figure 4 and Figure 1 is that step S103 in Figure 1 can be specifically refined into steps S1031 to S1034 in Figure 4, and step S104 in Figure 1 can be specifically refined into steps S1041 and S1042 in Figure 4.

[0040] In step S1031 , based on the user features of the users in the user group in the user group division result, a first distance between the user features corresponding to the user group and the first centroid of the user features corresponding to the user group is determined.

[0041] In the case where the user features have many dimensions, the user features of each user may include multiple fields, and each item in the user features may include one or more fields. The fields of the items may be spliced ​​together to form the user features. The user features of each user can be regarded as a multidimensional point, and the coordinates of the multidimensional point are the user features. The user features of the users in each user group can form a cluster, and the first centroid is the centroid of the cluster formed by the user features of the users in the user group. The first distance includes the distance between the user features of the users in the user group and the first centroid. The first distance can be implemented as Euclidean distance, Manhattan distance, etc., which is not limited here. Other distance algorithms that can characterize distance are also within the protection scope of the embodiments of this application.

[0042] In step S1032, a first discrimination index corresponding to the user group division result is calculated based on the first distance and the first centroid corresponding to every two user groups in the user group division result and the number of user groups in the user group division result.

[0043] Based on the first distance corresponding to each user group in the user group segmentation results, the average first distance from the user characteristics of each user in the user group to the first centroid can be obtained. Based on the average first distance between each two user groups in the user group segmentation results, the degree of convergence, i.e., the degree of similarity, in terms of user characteristics of each two user groups in the user group segmentation results can be obtained. Based on the first centroid corresponding to each two user groups in the user group segmentation results, the degree of separation, i.e., the degree of differentiation, in terms of user characteristics of each two user groups in the user group segmentation results can be obtained. Based on the degree of convergence and separation of user characteristics of each two user groups in the user group segmentation results, as well as the number of user groups in the user group segmentation results, the first differentiation index corresponding to the user group segmentation results can be determined.

[0044] Specifically, the first distance and the first center of mass corresponding to each two user groups in the user group division result can be used to calculate the first similarity parameter of the user characteristics of each two user groups in the user group division result; based on the maximum value of the first similarity parameters of each two user groups in the user group division result and the number of user groups in the user group division result, the first discrimination index corresponding to the user group division result is calculated.

[0045] Based on the first distances corresponding to each two user groups in the user group segmentation results, the mean of the first distances corresponding to each two user groups can be obtained. The mean of the first distances corresponding to each two user groups can be used to calculate the degree of similarity in user characteristics between the two user groups. The distance between the first centroids corresponding to each two user groups in the user group segmentation results can represent the degree of differentiation in user characteristics between the two user groups. The distance algorithm used to calculate the distance between the first centroids is consistent with the distance algorithm used to calculate the first distance. For example, if the first distance is Euclidean distance, the distance between the first centroids is also Euclidean distance; if the first distance is Manhattan distance, the distance between the first centroids is also Manhattan distance. Based on the mean of the first distances corresponding to each two user groups in the user group segmentation results and the distance between the first centroids corresponding to each two user groups, a first similarity parameter can be calculated. The first similarity parameter corresponding to each two user groups can represent the similarity in user characteristics between the two user groups. In some examples, the first similarity parameter is positively correlated with the similarity; that is, the larger the first similarity parameter, the greater the similarity. The first discrimination index corresponding to the user group segmentation result can represent the average similarity in user characteristics between each user group in the user group segmentation result and its most similar user group. The higher the average similarity, the lower the discrimination. For example, the first discrimination index can be implemented as the Davies-Bouldin Index (DBI for short). The first discrimination index corresponding to the user group segmentation result can be calculated according to the following formulas (1) to (3):

[0046] Among them, DBI f is the first discrimination index; K is the number of user groups in the user group division result; max is the maximum value operation; L1 is the number of users in the i-th user group; u l is the user feature of the lth user in the user group; is the first centroid corresponding to the i-th user group; ‖‖2 is the distance algorithm for the first distance, which is the Euclidean distance operation here, and other distance operation algorithms can also be used here; is the first distance between the user feature of the lth user in the i-th user group and the first centroid corresponding to the i-th user group; is the average of the first distances between the user features of the users in the i-th user group and the first centroid corresponding to the i-th user group; L2 is the number of users in the j-th user group; is the first centroid corresponding to the j-th user group; is the first distance between the user feature of the lth user in the jth user group and the first centroid corresponding to the jth user group; is the average of the first distances between the user features of the users in the j-th user group and the first centroid corresponding to the j-th user group; is the distance between the first centroid corresponding to the i-th user group and the first centroid corresponding to the j-th user group, which can represent the degree of differentiation between the i-th user group and the j-th user group in terms of user characteristics; It can represent the similarity between the i-th user group and the j-th user group in terms of user characteristics; is the first similarity parameter.

[0047] In step S1033, based on the value features of the users in the user group in the user group division result, a second distance between the value feature corresponding to the user group and the second centroid of the value feature corresponding to the user group is determined.

[0048] The value feature of each user may include one or more fields, and each item in the value feature may include one or more fields. The fields of the items may be concatenated to form the value feature. The value features of the users in each user group may form a cluster, and the second centroid is the centroid of the cluster formed by the value features of the users in the user group. The second distance includes the distance between the value features of the users in the user group and the second centroid. The second distance can be implemented as Euclidean distance, Manhattan distance, etc., which is not limited here. Other distance algorithms that can represent distance are also within the scope of protection of the embodiments of this application.

[0049] In step S1034, a second discrimination index corresponding to the user group division result is calculated based on the second distance and the second centroid corresponding to every two user groups in the user group division result and the number of user groups in the user group division result.

[0050] Based on the second distance corresponding to each user group in the user group segmentation results, the average second distance from the value feature of each user in the user group to the second centroid can be obtained. Based on the average second distance between each two user groups in the user group segmentation results, the degree of aggregation, i.e., the degree of similarity, of the value features of each two user groups in the user group segmentation results can be obtained. Based on the second centroid corresponding to each two user groups in the user group segmentation results, the degree of separation, i.e., the degree of differentiation, of the value features of each two user groups in the user group segmentation results can be obtained. Based on the degree of aggregation and separation of the value features of each two user groups in the user group segmentation results and the number of user groups in the user group segmentation results, the second differentiation index corresponding to the user group segmentation results can be determined.

[0051] Specifically, the second distance and second center of mass corresponding to each two user groups in the user group division result can be used to calculate the second similarity parameter of each two user groups in the user group division result in terms of value characteristics; based on the maximum value of the second similarity parameters of each two user groups in the user group division result and the number of user groups in the user group division result, the second discrimination index corresponding to the user group division result is calculated.

[0052] Based on the second distance corresponding to each two user groups in the user group segmentation results, the mean of the second distances corresponding to each two user groups can be obtained. Using this mean of the second distances corresponding to each two user groups, the degree of similarity between the two user groups in terms of value characteristics can be calculated. The distance between the second centroids corresponding to each two user groups in the user group segmentation results can represent the degree of differentiation between the two user groups in terms of value characteristics. The distance algorithm used to calculate the distance between the second centroids is consistent with the distance algorithm used to calculate the second distance. For example, if the second distance is Euclidean distance, the distance between the second centroids is also Euclidean distance; if the second distance is Manhattan distance, the distance between the second centroids is also Manhattan distance. Based on the mean of the second distances corresponding to each two user groups in the user group segmentation results and the distance between the second centroids corresponding to each two user groups, a second similarity parameter can be calculated. The second similarity parameter corresponding to each two user groups can represent the similarity between the two user groups in terms of value characteristics. In some examples, the second similarity parameter is positively correlated with the similarity; that is, the larger the second similarity parameter, the greater the similarity. The second discrimination index corresponding to the user group segmentation result can represent the average similarity in value characteristics between each user group in the user group segmentation result and its most similar user group. The higher the average similarity, the lower the discrimination. For example, the second discrimination index can be implemented as the Davies-Bouldin index. The second discrimination index corresponding to the user group segmentation result can be calculated according to the following formulas (4) to (6):

[0053] Among them, DBIv is the second discrimination index; K is the number of user groups in the user group division result; max is the maximum value operation; L1 is the number of users in the i-th user group; v l is the value feature of the lth user in the user group; is the second centroid corresponding to the i-th user group; ‖‖2 is the distance algorithm for the second distance, which is the Euclidean distance operation here, and other distance operation algorithms can also be used here; is the second distance from the value feature of the lth user in the i-th user group to the second centroid corresponding to the i-th user group; is the average value of the second distance from the value feature of the users in the i-th user group to the second centroid corresponding to the i-th user group; L2 is the number of users in the j-th user group; is the second centroid corresponding to the j-th user group; is the second distance from the value feature of the lth user in the jth user group to the second centroid corresponding to the jth user group; is the average value of the second distance from the value feature of the users in the j-th user group to the second centroid corresponding to the j-th user group; is the distance between the second centroid corresponding to the i-th user group and the second centroid corresponding to the j-th user group, which can represent the degree of differentiation between the i-th user group and the j-th user group in terms of value characteristics; It can represent the similarity between the value characteristics of the i-th user group and the j-th user group; is the second similarity parameter.

[0054] The execution order between step S1031 to step S1032 and step S1033 to step S1034 is not limited here. Step S1031 to step S1032 can be executed before step S1033 to step S1034, or after step S1033 to step S1034. Step S1031 to step S1032 and step S1033 to step S1034 can also be executed synchronously.

[0055] In step S1041, a weighted algorithm is used to obtain a comprehensive discrimination evaluation index corresponding to the user group segmentation result based on the first discrimination index, the first weight corresponding to the first discrimination index, the second discrimination index, and the second weight corresponding to the second discrimination index.

[0056] The first weight corresponds to the first discrimination index, the second weight corresponds to the second discrimination index, and the sum of the first weight and the second weight is 1. The values ​​of the first weight and the second weight can be set according to the scenario, needs, experience, etc. For example, the first weight is 0.5 and the second weight is 0.5. The comprehensive discrimination evaluation index can be used to characterize the comprehensive discrimination in the user feature dimension and the value feature dimension. Similarly, the comprehensive discrimination evaluation index can also be used to characterize the comprehensive similarity in the user feature dimension and the value feature dimension. The comprehensive discrimination evaluation index is a discrimination evaluation index that integrates the discrimination in the user feature dimension and the discrimination in the value feature dimension. The comprehensive discrimination evaluation index can be positively correlated with the comprehensive discrimination, that is, negatively correlated with the comprehensive similarity, or negatively correlated with the comprehensive discrimination, that is, positively correlated with the comprehensive similarity. For example, the comprehensive discrimination evaluation index is negatively correlated with the comprehensive discrimination, and the comprehensive discrimination evaluation index corresponding to the nth user group division result can be calculated according to the following formula (7):

[0057] in, is the comprehensive discrimination evaluation index corresponding to the nth user group classification result. The larger the value of the comprehensive discrimination evaluation index, the smaller the corresponding comprehensive discrimination. The smaller the value of the comprehensive discrimination evaluation index, the larger the corresponding comprehensive discrimination. (1-α) is the first weight. DBI f is the first discrimination index corresponding to the nth user group classification result; α is the second weight; DBI f The second discrimination index corresponding to the n-th user group classification result.

[0058] In step S1042 , the user group division result with the highest comprehensive discrimination represented by the comprehensive discrimination evaluation index is selected from the N user group division results as the target user group division result.

[0059] Each user group segmentation result corresponds to a comprehensive discrimination evaluation index. When the comprehensive discrimination evaluation index is negatively correlated with the comprehensive discrimination, the user group segmentation result with the smallest comprehensive discrimination evaluation index is selected from the N user group segmentation results as the target user group segmentation result; when the comprehensive discrimination evaluation index is positively correlated with the comprehensive discrimination, the user group segmentation result with the largest comprehensive discrimination evaluation index is selected from the N user group segmentation results as the target user group segmentation result. The user group in the user group segmentation result with the highest comprehensive discrimination among the N user group segmentation results has the best comprehensive performance in terms of discrimination in the user feature dimension and the value feature dimension. This user group segmentation result is determined as the target user group segmentation result, and the target user group segmentation result is the user group segmentation result that is finally selected. Subsequently, the business can be refined based on the user groups in the target user group segmentation results.

[0060] In some embodiments, before clustering multiple users based on user features and N sets of preset clustering hyperparameters, the user features may be cleaned and normalized. This allows subsequent clustering and other related calculations to be performed using the cleaned and normalized user features, thereby improving data processing efficiency and accuracy. Figure 5 is a flowchart of a user group segmentation method provided in another embodiment of the present application. The difference between Figure 5 and Figure 1 is that the user group segmentation method shown in Figure 5 may also include steps S105 to S107.

[0061] In step S105 , the user characteristics are processed using the M-quantile method to determine the normal value interval and abnormal value in the user characteristics.

[0062] M is an integer greater than 2. The value of M can be set based on the scenario, needs, experience, etc. For example, M can be 4. The M-quantile method can be used to process user features to obtain the M-quantile of the user features. The upper and lower limits can be obtained based on the upper and lower M-quantiles, and the difference between the upper and lower M-quantiles. These upper and lower limits can be used to define the normal value interval. The normal value interval is the interval within which the normal value of the user feature falls. If the user feature falls within the normal value interval, it is considered a normal value; if the user feature falls outside the normal value interval, it is considered an abnormal value. For example, M is 4, that is, the user characteristics are processed using the quartile method to obtain the upper quartile QU and the lower quartile QL, and the difference IQR between the upper quartile QU and the lower quartile QL is obtained. The lower limit of the normal value interval is QL-k×IQR, and the upper limit of the normal value interval is QU+k×IQR. k is the adjustment coefficient, and the value of k can be set according to the scenario, needs, experience, etc. Correspondingly, the normal value interval is [QL-k×IQR, QU+k×IQR], and the value outside the normal value interval is an abnormal value.

[0063] In step S106 , the upper limit or lower limit of the normal value interval is used to replace the abnormal value to obtain the cleaned user features.

[0064] To prevent outliers from adversely affecting subsequent clustering and related calculations, outliers above the upper limit of the normal value interval can be replaced with the upper limit, and outliers below the lower limit can be replaced with the lower limit. The cleaned user features include those within the normal value interval, those replaced with the upper limit of the normal value interval, and those replaced with the lower limit of the normal value interval.

[0065] In step S107 , the cleaned user features are standardized to obtain standardized user features.

[0066] The value of each field in the normalized user feature is within a preset standard value range. The preset standard value range can be set based on the scenario, requirements, experience, etc. For example, the preset standard value range can be [0, 1].

[0067] The standardization process performed on the cleaned user features is related to the data type of the user features. In some examples, when the user features belong to ordered discrete numerical data, the standardization process may include a numerical bin setting process and a one-hot encoding process. The numerical bin setting process may refer to pre-dividing the value range of the user feature into multiple numerical bins, and determining the numerical bin to which the cleaned user features belong. The one-hot encoding process may refer to performing one-hot encoding on each numerical bin to obtain a one-hot encoding corresponding to each numerical bin. The one-hot encoding may include one or more fields, and the encoding corresponding to the data bin to which the cleaned user features belong is the user feature after standardization. For example, if user features such as user age and the number of resource cards held by the user belong to ordered discrete numerical data, standardization can be achieved through numerical bin setting and one-hot encoding. The ordered discrete numerical data after standardization includes multiple fields, and each field can be 0 or 1.

[0068] In other examples, when the user features are continuous numerical data, the standardization process includes a numerical range narrowing process and a maximum and minimum normalization process. There may be a very large gap between the values ​​of continuous numerical data. In order to facilitate processing, the numerical range narrowing process can be used to narrow the value of the user features to a certain range. For example, the numerical range narrowing process can be implemented as a logarithmic operation process, but is not limited to this. The numerical range narrowing process cannot directly obtain user features whose values ​​of each field are within the preset standard value range. After the numerical range narrowing process is performed, the user features after the numerical range narrowing process can be subjected to a maximum and minimum normalization (i.e., Min-Max Normalization) process to obtain user features whose values ​​are within the preset standard value range. For example, user features such as transaction amount and user account deposit amount are continuous numerical data, and can be standardized through numerical range narrowing process and maximum and minimum normalization process. The continuous numerical data after standardization can include a field whose value is greater than or equal to 0 and less than or equal to 1.

[0069] In other examples, when user features are coded data, the standardization process includes one-hot encoding. One-hot encoding can be directly performed on the coded data to obtain the one-hot encoding corresponding to the coded data. The user features belonging to the coded data after standardization are the one-hot encoded user features. For example, if user features such as user gender and user location belong to coded data, standardization can be achieved through one-hot encoding. The user features belonging to the coded data after standardization can include one field or two or more fields, each of which has a value of 0 or 1.

[0070] It also removes the dimension of user features, retaining the cleaned and standardized values, making them comparable. Subsequent user features used in clustering and other related calculations include the cleaned and standardized de-dimensionalized user features.

[0071] Through data cleaning and standardization, the accuracy and processability of user features can be improved, thereby improving the efficiency and accuracy of subsequent clustering and other related calculations based on user features.

[0072] FIG6 is a schematic diagram of the structure of the user group segmentation device provided in one embodiment of the present application. As shown in FIG6 , the user group segmentation device 200 may include a feature acquisition module 201 , a clustering module 202 , an indicator determination module 203 , and a user group determination module 204 .

[0073] The feature acquisition module 201 may be used to acquire user features and value features of multiple users.

[0074] Value features are used to characterize the value of users to the business.

[0075] In some examples, user characteristics include, but are not limited to, one or more of the following categories: user demographic characteristics, user resource characteristics, user risk characteristics, user transaction habit characteristics, and user resource channel characteristics.

[0076] The clustering module 202 may be used to cluster multiple users based on user characteristics and N preset groups of clustering hyperparameters to obtain N types of user group division results.

[0077] A user group segmentation result corresponds to a set of clustering hyperparameters. The user group segmentation result includes the user groups obtained by clustering. N is an integer greater than or equal to 2.

[0078] The index determination module 203 may be configured to calculate a first discrimination index and a second discrimination index corresponding to each user group segmentation result based on the user characteristics and value characteristics of the users in the user groups in the user group segmentation result.

[0079] The user group determination module 204 may be configured to select a user group division result with the highest comprehensive discrimination degree from the N user group division results according to the first discrimination degree index and the second discrimination degree index as the target user group division result.

[0080] In an embodiment of the present application, users can be clustered based on their user characteristics and multiple sets of clustering hyperparameters to obtain a variety of user group division results corresponding to the multiple sets of clustering hyperparameters. According to the user characteristics and value characteristics of the users in the user group of the user group division result, a first discrimination index that can characterize the discrimination of the user characteristic dimension and a second discrimination index that can characterize the discrimination of the value characteristic dimension are obtained. According to the first discrimination index and the second discrimination index, the comprehensive discrimination of the user group in each user group division result can be comprehensively evaluated from the two aspects of the user characteristic dimension and the value characteristic dimension. The user group in the user group division result with the highest comprehensive discrimination is a user group that takes into account both the discrimination of the user characteristic dimension and the discrimination of the value characteristic dimension, and the discrimination of the user characteristic dimension and the discrimination of the value characteristic dimension are both better, thereby improving the discrimination of the user group obtained by division and the adaptability of the user group to the business.

[0081] In some embodiments, the indicator determination module 203 can be specifically used to: determine the first distance between the user characteristics corresponding to the user group and the first centroid of the user characteristics corresponding to the user group based on the user characteristics of the users in the user group in the user group division result; calculate the first discrimination indicator corresponding to the user group division result according to the first distance, the first centroid and the number of user groups in the user group division result corresponding to every two user groups in the user group division result; determine the second distance between the value characteristics corresponding to the user group and the second centroid of the value characteristics corresponding to the user group based on the value characteristics of the users in the user group in the user group division result; calculate the second discrimination indicator corresponding to the user group division result according to the second distance, the second centroid and the number of user groups in the user group division result corresponding to every two user groups in the user group division result.

[0082] In some examples, the indicator determination module 203 can be specifically used to: use the first distance and first center of mass corresponding to each two user groups in the user group division result to calculate the first similarity parameter of the user characteristics of each two user groups in the user group division result; calculate the first discrimination indicator corresponding to the user group division result based on the maximum value of the first similarity parameters of each two user groups in the user group division result and the number of user groups in the user group division result.

[0083] In some examples, the indicator determination module 203 can be specifically used to: use the second distance and second center of mass corresponding to each two user groups in the user group division result to calculate the second similarity parameter of each two user groups in the user group division result in terms of value characteristics; calculate the second discrimination indicator corresponding to the user group division result based on the maximum value of the second similarity parameters of each two user groups in the user group division result and the number of user groups in the user group division result.

[0084] In some embodiments, the user group determination module 204 can be specifically used to: obtain a comprehensive discrimination evaluation index corresponding to the user group segmentation result based on the first discrimination index, the first weight corresponding to the first discrimination index, the second discrimination index, and the second weight corresponding to the second discrimination index using a weighted algorithm; and select the user group segmentation result with the highest comprehensive discrimination represented by the comprehensive discrimination evaluation index from N types of user group segmentation results as the target user group segmentation result.

[0085] In some embodiments, the user group segmentation device 200 may further include a data preprocessing module. The data preprocessing module may be configured to: process user features using the M-quantile method to determine normal value intervals and abnormal values ​​in the user features, where M is an integer greater than 2; replace abnormal values ​​with the upper or lower limit of the normal value interval to obtain cleaned user features; and perform standardization on the cleaned user features to obtain standardized user features, wherein the value of each field in the standardized user features is within a preset standard value range.

[0086] In some examples, when user features are ordered discrete numerical data, normalization includes value binning and one-hot encoding. When user features are continuous numerical data, normalization includes value range reduction and maximum and minimum normalization. When user features are coded data, normalization includes one-hot encoding.

[0087] In a third aspect, the present application further provides an electronic device. FIG7 is a schematic diagram of the structure of an electronic device provided in one embodiment of the present application. As shown in FIG7 , the electronic device 300 includes a memory 301, a processor 302, and a computer program stored in the memory 301 and executable on the processor 302.

[0088] In some examples, the processor 302 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.

[0089] The memory 301 may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Therefore, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the user group segmentation method according to the embodiment of the present application.

[0090] The processor 302 reads the executable program code stored in the memory 301 to run a computer program corresponding to the executable program code, so as to implement the user group division method in the above embodiment.

[0091] In some examples, the electronic device 300 may further include a communication interface 303 and a bus 304. As shown in FIG7, the memory 301, the processor 302, and the communication interface 303 are connected via the bus 304 and communicate with each other.

[0092] The communication interface 303 is mainly used to implement communication between the modules, devices, units and / or equipment in the embodiment of the present application. Input devices and / or output devices can also be connected through the communication interface 303.

[0093] Bus 304 includes hardware, software, or both that couples components of electronic device 300 to each other. By way of example, and not limitation, bus 304 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-E) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of the above. Bus 304 may include one or more buses, where appropriate. Although embodiments herein describe and illustrate a particular bus, this application contemplates any suitable bus or interconnect.

[0094] In a fourth aspect, the present application further provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the user group segmentation method in the above-mentioned embodiment can be implemented, and the same technical effect can be achieved. To avoid repetition, the above-mentioned computer-readable storage medium may include a non-transitory computer-readable storage medium, such as a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., which is not limited here.

[0095] The present application provides a computer program product. When the instructions in the computer program product are executed by the processor of an electronic device, the electronic device executes the user group division method in the above embodiment and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0096] It should be understood that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. For device embodiments, equipment embodiments, computer-readable storage medium embodiments, and computer program product embodiments, the relevant parts can be referred to the description section of the method embodiment. This application is not limited to the specific steps and structures described above and shown in the figures. Those skilled in the art can make various changes, modifications and additions, or change the order of the steps after understanding the spirit of this application. In addition, for the sake of brevity, a detailed description of known method technologies is omitted here.

[0097] Aspects of the present application have been described above with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. 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 the 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 via 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. This 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 is also understood that each box in the block diagram and / or the flowchart and the combination of the boxes in the block diagram and / or the flowchart can also be implemented by the dedicated hardware that performs the specified function or action, or can be implemented by the combination of dedicated hardware and computer instructions.

[0098] Those skilled in the art should understand that the above embodiments are illustrative rather than restrictive. Different technical features appearing in different embodiments can be combined to achieve beneficial effects. Based on a study of the drawings, the specification and the claims, those skilled in the art should be able to understand and implement other variations of the disclosed embodiments. In the claims, the term "comprising" does not exclude other devices or steps; the quantifier "one" does not exclude a plurality; the terms "first" and "second" are used to identify names rather than to indicate any specific order. Any figure marks in the claims should not be understood as limiting the scope of protection. The functions of multiple parts appearing in the claims can be implemented by a separate hardware or software module. The fact that certain technical features appear in different dependent claims does not mean that these technical features cannot be combined to achieve beneficial effects.

Claims

1. A method for dividing user groups, comprising: Acquire user characteristics and value characteristics of multiple users, where the value characteristics are used to characterize the value of the users to the service; Based on the user characteristics and N preset groups of clustering hyperparameters, multiple users are clustered to obtain N user group division results, where one user group division result corresponds to one group of clustering hyperparameters, the user group division result includes user groups obtained by clustering division, and N is an integer greater than or equal to 2; According to the user characteristics and the value characteristics of the users in the user group in the user group division results, a first discrimination index and a second discrimination index corresponding to each of the user group division results are calculated; According to the first discrimination index and the second discrimination index, the user group division result with the highest comprehensive discrimination is selected from the N user group division results as the target user group division result.

2. The method according to claim 1, wherein: The calculating, according to the user characteristics and the value characteristics of the users in the user groups in the user group division results, the first discrimination index and the second discrimination index corresponding to each user group division result comprises: Based on the user features of the users in the user group in the user group division result, determining a first distance between the user features corresponding to the user group and a first centroid of the user features corresponding to the user group; Calculate the first discrimination index corresponding to the user group division result according to the first distance corresponding to every two user groups in the user group division result, the first centroid, and the number of user groups in the user group division result; Based on the value features of the users in the user group in the user group division result, determining a second distance between the value feature corresponding to the user group and a second centroid of the value feature corresponding to the user group; The user group division result is calculated based on the second distance corresponding to every two user groups in the user group division result, the second centroid and the number of user groups in the user group division result. The second discrimination index corresponding to the score result.

3. The method according to claim 2, wherein: The calculating the first discrimination index corresponding to the user group division result according to the first distance corresponding to every two user groups in the user group division result, the second centroid, and the number of user groups in the user group division result includes: By using the first distance and the first centroid corresponding to every two user groups in the user group division result, a first similarity parameter of the user features between every two user groups in the user group division result is calculated; The first discrimination index corresponding to the user group division result is calculated according to the maximum value of the first similarity parameters between every two user groups in the user group division result and the number of user groups in the user group division result.

4. The method according to claim 2, wherein: The calculating, according to the second distance corresponding to every two user groups in the user group division result and the number of user groups in the user group division result, the second discrimination index corresponding to the user group division result includes: By using the second distance and the second centroid corresponding to every two user groups in the user group division result, a second similarity parameter of the value feature between every two user groups in the user group division result is calculated; The second discrimination index corresponding to the user group division result is calculated according to the maximum value of the second similarity parameters between every two user groups in the user group division result and the number of user groups in the user group division result.

5. The method according to claim 1, wherein: The selecting, according to the first discrimination index and the second discrimination index, the user group division result with the highest comprehensive discrimination degree from the N user group division results as the target user group division result includes: Obtaining a comprehensive discrimination evaluation index corresponding to the user group segmentation result by using a weighted algorithm according to the first discrimination index, the first weight corresponding to the first discrimination index, the second discrimination index, and the second weight corresponding to the second discrimination index; The user group division result with the highest comprehensive discrimination represented by the comprehensive discrimination evaluation index is selected from the N user group division results as the target user group division result.

6. The method according to claim 1, wherein: Before clustering multiple users based on the user features and the preset N groups of clustering hyperparameters, the method further includes: Processing the user characteristics by using the M-quantile method to determine the normal value interval and abnormal value in the user characteristics, where M is an integer greater than 2; Replacing the abnormal value with the upper limit or lower limit of the normal value interval to obtain the cleaned user feature; The cleaned user features are standardized to obtain the standardized user features, wherein the value of each field in the standardized user features is within a preset standard value range.

7. The method according to claim 6, wherein: In the case where the user feature is an ordered discrete numerical data, the standardization process includes a numerical bin setting process and a one-hot encoding process; In the case where the user feature is continuous numerical data, the standardization process includes a numerical range reduction process and a maximum and minimum normalization process; In the case where the user feature belongs to encoding type data, the standardization process includes one-hot encoding process.

8. The method according to any one of claims 1 to 7, wherein: The user characteristics include one or more of the following categories: User demographic characteristics, user resource characteristics, user risk characteristics, user transaction habit characteristics, and user resource channel characteristics.

9. A user group division device, comprising: A feature acquisition module, used to acquire user features and value features of multiple users, wherein the value features are used to characterize the value of the user to the service; A clustering module, configured to cluster multiple users based on the user characteristics and N preset groups of clustering hyperparameters to obtain N user group division results, wherein one user group division result corresponds to one group of clustering hyperparameters, and the user group division result includes user groups obtained by clustering division, and N is an integer greater than or equal to 2; An indicator determination module, configured to calculate a first discrimination index and a second discrimination index corresponding to each user group segmentation result according to the user characteristics and the value characteristics of the users in the user group in the user group segmentation result; The user group determination module is used to select the user group division result with the highest comprehensive discrimination degree from the N user group division results as the target user group division result according to the first discrimination degree index and the second discrimination degree index.

10. An electronic device comprising: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the method for dividing user groups according to any one of claims 1 to 8 is implemented.

11. A computer-readable storage medium, wherein computer program instructions are stored on the computer-readable storage medium, and when the computer program instructions are executed by a processor, the method for dividing user groups according to any one of claims 1 to 8 is implemented.

Images