User data processing method and apparatus, computer device, and storage medium

Abstract

The application relates to a user data processing method and device, computer equipment and a storage medium. The method comprises the following steps: obtaining a target user group to be classified, the target user group comprising a plurality of target users; dividing the target user group based on a candidate feature dimension in a candidate feature dimension set, to obtain a set of divided user groups corresponding to the candidate feature dimension; obtaining user group features corresponding to each divided user group in the set of divided user groups; obtaining feature differences between the user group features; obtaining feature discrimination degrees corresponding to the candidate feature dimension based on the feature differences, the feature discrimination degrees being in a positive correlation with the feature differences; and selecting a target feature dimension from the candidate feature dimension set based on the feature discrimination degrees, the target feature dimension being used for classifying the target user group. In the application, a clustering algorithm or a neural network model based on artificial intelligence can be used to divide the target user group. The method can improve the accuracy of dividing the user group.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a user data processing method, apparatus, computer equipment, and storage medium. Background Art

[0002] With the development of computer and artificial intelligence technologies, user data processing is often required, such as user grouping. User grouping involves dividing multiple users into groups. Content can also be pushed based on these groups, improving the accuracy of content push. Currently, user grouping can be achieved by manually tagging users or using artificial intelligence algorithms.

[0003] However, existing user clustering methods are unable to accurately divide user groups. For example, manual labeling clustering methods cannot accurately divide user groups due to the influence of subjective consciousness, resulting in low accuracy in dividing user groups. Summary of the Invention

[0004] Based on this, it is necessary to provide a user data processing method, apparatus, computer device and storage medium that can improve the accuracy of dividing user groups in order to address the above technical problems.

[0005] A user data processing method, the method comprising: obtaining a target user group to be classified; the target user group includes multiple target users; dividing the target user group based on candidate feature dimensions in a candidate feature dimension set to obtain a set of divided user groups corresponding to the candidate feature dimensions; obtaining user group features corresponding to each divided user group in the set of divided user groups; obtaining feature differences between the user group features, and obtaining feature discrimination corresponding to the candidate feature dimensions based on the feature differences, wherein the feature discrimination is positively correlated with the feature differences; selecting a target feature dimension from the candidate feature dimension set based on the feature discrimination, and the target feature dimension is used to classify the target user group.

[0006] A user data processing device, the device comprising: a target user group acquisition module, for acquiring a target user group to be classified; the target user group comprises a plurality of target users; a divided user group set acquisition module, for dividing the target user group based on candidate feature dimensions in a candidate feature dimension set, respectively, to obtain a divided user group set corresponding to the candidate feature dimensions; a user group feature acquisition module, for acquiring user group features corresponding to each divided user group in the divided user group set; a feature difference acquisition module, for acquiring feature differences between the user group features, and obtaining feature discrimination corresponding to the candidate feature dimensions based on the feature differences, wherein the feature discrimination is positively correlated with the feature differences; a target feature dimension acquisition module, for selecting a target feature dimension from the candidate feature dimension set based on the feature discrimination, wherein the target feature dimension is used to classify the target user group.

[0007] In some embodiments, the user group feature acquisition module includes: a target statistical value acquisition unit, which is used to perform statistics on multiple user features corresponding to the divided user groups to obtain target statistical values ​​corresponding to each of the user features; a user group feature acquisition unit, which is used to obtain target feature values ​​corresponding to each of the target statistical values, arrange each of the target feature values, and obtain the user group features.

[0008] In some embodiments, the user characteristics include interactive behavior characteristics, and the target statistical value obtaining unit is further used to count users in the divided user groups who have an interactive correlation relationship with the first interactive content to obtain a target statistical value corresponding to the interactive behavior characteristics.

[0009] In some embodiments, the user group feature obtaining unit is also used to obtain a first reference statistical value, which is a statistical value obtained by statistically analyzing the user features corresponding to the reference user group; and obtain the corresponding target feature value based on the numerical relationship between the target statistical value and the first reference statistical value.

[0010] In some embodiments, the numerical relationship includes a ratio relationship, and the user group feature obtaining unit is further used to perform a ratio operation on the target statistical value and the first reference statistical value to obtain a target ratio; obtain the target ratio range in which the target ratio is located, and use the range characteristic value corresponding to the target ratio range as the target characteristic value.

[0011] In some embodiments, the target feature dimension obtaining module is also used to select candidate feature dimensions that meet the discrimination conditions from the candidate feature dimension set based on the feature discrimination, and use the candidate feature dimensions that meet the discrimination conditions as target feature dimensions; the discrimination conditions include at least one of the feature discrimination being greater than the discrimination threshold or the discrimination ranking being before the ranking threshold.

[0012] In some embodiments, the target feature dimension obtaining module is also used to select candidate feature dimensions that meet the discrimination conditions from the candidate feature dimension set based on the feature discrimination, and use each candidate feature dimension that meets the discrimination conditions as a selected feature dimension to form a selected feature dimension set; the discrimination condition includes at least one of the feature discrimination being greater than the discrimination threshold or the discrimination ranking being before the ranking threshold; obtain the feature correlation between the selected feature dimensions in the selected feature dimension set; obtain a selected feature dimension combination whose feature correlation is greater than the correlation threshold, the selected feature dimension combination includes a selected feature dimension whose feature correlation is greater than the correlation threshold, determine the selected feature dimension to be removed from the selected feature dimension combination, remove the selected feature dimension to be removed from the selected feature dimension set, and use at least one of the remaining selected feature dimensions in the selected feature dimension set as the target feature dimension.

[0013] In some embodiments, the target feature dimension obtaining module is also used to obtain the feature coverage of each selected feature dimension in the selected feature dimension combination in the target user group; and the selected feature dimension with smaller feature coverage in the selected feature dimension combination is used as the selected feature dimension to be removed.

[0014] In some embodiments, the feature difference acquisition module includes: a feature distance calculation unit, used to calculate the feature distance between the user group features, and use the calculated feature distance as the feature difference between the user group features; a feature discrimination obtaining unit, used to use the feature distance as the feature discrimination corresponding to the candidate feature dimension.

[0015] In some embodiments, the device further includes: a target sub-user group obtaining module, which is used to classify the target user group based on the target feature dimension to obtain a target sub-user group; a user label obtaining module, which is used to count the user features of users in the target sub-user group, and obtain the user label corresponding to the target sub-user group based on the statistically obtained user statistical features.

[0016] In some embodiments, the user tag obtaining module includes: a user statistical value obtaining unit, which is used to count the users in the target sub-user group who have an interactive association relationship with the second interactive content, and obtain the user statistical value corresponding to the second interactive content; a statistical difference value obtaining unit, which is used to obtain the second reference statistical value corresponding to the second interactive content, and obtain the statistical difference value between the user statistical value and the second reference statistical value; a user tag obtaining unit, which is used to obtain the user tag corresponding to the target sub-user group based on the content attribute corresponding to the second interactive content when the statistical difference value is greater than the difference threshold, so as to push content based on the user tag.

[0017] In some embodiments, the user tag acquisition module also includes: an attribute statistical feature acquisition unit, which is used to count the attribute features of users in the target sub-user group, obtain the attribute statistical features corresponding to the target sub-user group, and use the attribute statistical features as the user tag corresponding to the target sub-user group; an attribute statistical feature acquisition unit, which is used to send the attribute statistical features to the device that sends the attribute feature acquisition request when receiving an attribute feature acquisition request for users in the target sub-user group.

[0018] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above-mentioned user data processing method when executing the computer program.

[0019] A computer-readable storage medium stores a computer program, which implements the steps of the above-mentioned user data processing method when executed by a processor.

[0020] In some embodiments, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of each of the above method embodiments.

[0021] The above-mentioned user data processing method, device, computer equipment and storage medium obtain a target user group to be classified, the target user group includes multiple target users, and divides the target user group based on the candidate feature dimensions in the candidate feature dimension set to obtain a set of divided user groups corresponding to the candidate feature dimensions, obtains user group features corresponding to each divided user group in the divided user group set, obtains feature differences between user group features, obtains feature discrimination corresponding to the candidate feature dimension based on the feature differences, and the feature discrimination is positively correlated with the feature difference. The target feature dimension is selected from the candidate feature dimension set based on the feature discrimination, and the target feature dimension is used to classify the target user group. Since the feature discrimination is positively correlated with the feature difference, and the feature difference is the difference between user group features, the greater the difference between user group features, the greater the feature discrimination. Therefore, based on the feature discrimination, a feature dimension that can distinguish user groups can be selected, thereby improving the accuracy of the obtained feature dimension. Therefore, when the obtained feature dimension is used to classify the user group, the accuracy of dividing the user group can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A diagram of an application environment of a user data processing method in some embodiments;

[0023] Figure 2 is a flowchart of a method for processing user data in some embodiments;

[0024] Figure 3 is a schematic diagram of feature correlation in some embodiments;

[0025] Figure 4 is a schematic diagram of clustering in some embodiments;

[0026] Figure 5 A diagram of an application environment of a user data processing method in some embodiments;

[0027] Figure 6 A diagram of an application environment of a user data processing method in some embodiments;

[0028] Figure 7 A diagram of an application environment of a user data processing method in some embodiments;

[0029] Figure 8 is a schematic diagram of a user data processing method in some embodiments;

[0030] Figure 9 is a structural block diagram of a user data processing device in some embodiments;

[0031] Figure 10 is a diagram of the internal structure of a computer device in some embodiments;

[0032] Figure 11 1 is a diagram of the internal structure of a computer device in some embodiments. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0034] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also involves studying the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0035] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0036] Natural language processing (NLP) is a key area of ​​research in computer science and artificial intelligence. It studies the theories and methods that enable effective communication between humans and computers using natural language. Natural language processing (NLP) integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language we use in everyday life—and is closely linked to the study of linguistics. Natural language processing technologies typically include text processing, semantic understanding, machine translation, robotic question answering, and knowledge graphs.

[0037] Machine learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is at the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0038] With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, autonomous driving, drones, robots, smart medical care, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0039] The solutions provided in the embodiments of this application involve technologies such as natural language processing and machine learning of artificial intelligence, which are specifically illustrated by the following embodiments:

[0040] The user data processing method provided in this application can be applied to Figure 1 The application environment shown in FIG. 1 includes a terminal 102 and a server 104. The terminal 102 and the server 104 communicate via a network. Various applications can run on the terminal 102, such as at least one of a social application, an e-commerce application, a payment application, a voice application, an entertainment application, a video application, or a tool application.

[0041] Specifically, the terminal 102 can display the target content and can send a user analysis request to the server 104 to obtain the analysis results of the user corresponding to the target content from the server 104. For example, when the terminal 102 obtains a user analysis trigger operation, such as obtaining a trigger operation for the control "View User Analysis", it can send a user analysis request for the target content to the server 104. When the server 104 receives the user analysis request sent by the terminal 102, the server 104 can respond to the user analysis request, obtain users corresponding to the target content, and form a target user group to be classified. The target user group includes multiple target users. The server 104 can divide the target user group based on the candidate feature dimensions in the candidate feature dimension set, obtain a set of divided user groups corresponding to the candidate feature dimensions, obtain user group features corresponding to each divided user group in the divided user group set, obtain feature differences between the user group features, and obtain feature discrimination corresponding to the candidate feature dimensions based on the feature differences. The feature discrimination is positively correlated with the feature differences. The target feature dimension is selected from the candidate feature dimension set based on the feature discrimination. The server 104 can classify the target user group based on the target feature dimension, that is, divide the target user group to obtain the user groups to which the target users belong. The target content can be any content, for example, including at least one of video, text, image, or audio. The users corresponding to the target content can be users who interact with the target content, for example, users who share, watch, download, or like the target content.

[0042] A positive correlation means that, with other conditions remaining unchanged, the two variables change in the same direction. When one variable changes from large to small, the other also changes from large to small. It's understandable that a positive correlation here means the direction of change is consistent, but it doesn't require that when one variable changes slightly, the other also changes. For example, you could set variable b to 100 when variable a is between 10 and 20, and 120 when variable a is between 20 and 30. In this way, both a and b change in the same direction: when a increases, b also increases. However, when a is between 10 and 20, b can remain unchanged.

[0043] The terminal 102 may be, but is not limited to, a laptop computer, a smart phone, a tablet computer, a desktop computer, a smart TV, and a portable wearable device, and the server 104 may be implemented as an independent server or a server cluster or cloud server consisting of multiple servers.

[0044] It can be understood that the above application scenario is only an example and does not constitute a limitation on the user data processing method provided in the embodiment of the present application. The method provided in the embodiment of the present application can also be applied in other application scenarios. For example, the user data processing method provided in the present application can be executed by the terminal 102. The terminal 102 can upload the obtained target feature dimension to the server 104. The server 104 can store the target feature dimension and divide the user group based on the target feature dimension. The terminal 102 can also divide the target user group based on the obtained target feature dimension and upload the division result to the server 104.

[0045] In the user data processing method disclosed in this application, the user group segmentation results obtained by segmenting the target user group can be stored in a blockchain. For example, a query code corresponding to the user group segmentation result can be generated, and the query code and the user group segmentation result can be stored on the blockchain. When the user group segmentation result is needed, the corresponding user group segmentation result can be queried from the blockchain based on the query code, thereby improving the security of the user group segmentation result. The user group segmentation result can include the number of sub-user groups obtained by segmentation and at least one of attribute information or behavior information of users corresponding to each sub-user group.

[0046] In some embodiments, as Figure 2 As shown, a user data processing method is provided. The method can be executed by a terminal or a server, or by both the terminal and the server. In the embodiment of the present application, the method is applied to Figure 1 Taking the server 104 in the example as an example, the following steps are included:

[0047] S202, obtaining a target user group to be classified; the target user group includes multiple target users.

[0048] A user group refers to a group of multiple users, where "multiple" means at least two. A user group can be associated with a user group identifier, which uniquely identifies the user group. A user can be associated with a user identifier, which uniquely identifies the user and can include at least one of the user's mobile phone number, ID number, or an account registered on the user's application. A target user group refers to the user group to be categorized, i.e., the user group to be divided. Target users refer to users within the target user group.

[0049] The target users in the target user group can be any users or users with the same characteristics. The sources can be the same or different. For example, they can be users who watch the same video, users who watch a certain drama, or users who watch the same channel. They can also be users on the same application, such as users on video applications.

[0050] The target user group may be a user group composed of users with the same user attribute characteristics. User attribute characteristics may be characteristics obtained based on user attribute data. User attribute data may be used as user attribute characteristics, or user attribute data may be calculated, such as normalized, and the result of the calculation may be used as user attribute characteristics. User attribute data is data related to the user's attributes, used to reflect the user's basic information, and may include at least one of age, gender, education level, birthday, occupation, zodiac sign, blood type, or regional information. User attribute characteristics may include at least one of age, gender, education level, birthday, occupation, zodiac sign, blood type, or regional information. Among them, regional information refers to information about the area where the user is located, such as the city level. The target user group may be, for example, a user group composed of users of the same gender.

[0051] The target user group can be a user group consisting of users with the same user behavior characteristics. User behavior characteristics can be characteristics derived from user behavior data. User behavior data can be used as user behavior characteristics, or the user behavior data can be calculated, such as by performing a normalized calculation, with the calculated result used as the user behavior characteristics. User behavior data is data related to user behavior, and can be, for example, user behavior data within a historical time period. For example, it can include at least one of the user's online or offline behavior data. For example, it can include data on the user's historical online operations, such as at least one of the content, time, or number of times the user searches, downloads, listens, shares, or favorites in a music playback application. User behavior data can reflect the user's habits or preferences. For example, when the user behavior data is a record of the user's historical online operations, the user behavior data can reflect the user's online habits or preferences. The historical time period can be set as needed or pre-set, such as at least one of the last month, the last week, or the last three days. For users in video applications, user behavior characteristics can include at least one of the following: monthly active days, video play duration, short video play days, long video play days, short video play duration, long video play duration, or user level. The user level can include at least one of an ordinary user and a VIP (very important person) user. The target user group can be, for example, a user group consisting of users who download the same music in a music playback application. Of course, the users in the target user group can also be users with the same user attribute characteristics and the same behavioral characteristics. The source of the target users in the target user group is not limited in this application.

[0052] Specifically, the server may store a reference user group, which may include users from the same source as well as users from different sources. The server may retrieve individual users from the same source from the reference user group to form a target user group to be classified. Users on the same application software may be considered as users from the same source, or users corresponding to the same content on the same application software may be considered as users from the same source. For example, users watching the same movie on the same video software may be considered as users from the same source. Users corresponding to the same content on different application software may also be considered as users from the same source. For example, users watching the same movie on different video software may be considered as users from the same source.

[0053] In some embodiments, the server may obtain users from the same source from a reference user group to form a candidate user group, and count the number of users included in the candidate user group. When the number of users counted exceeds a user number threshold, the candidate user group is used as a target user group to be classified. The user number threshold can be set as needed or pre-set.

[0054] S204 , dividing the target user group based on the candidate feature dimensions in the candidate feature dimension set, and obtaining a set of divided user groups corresponding to the candidate feature dimensions.

[0055] Here, a feature dimension refers to a dimension used to describe user characteristics. User characteristics may include at least one of user attribute characteristics or user behavior characteristics. Feature dimensions may include attribute feature dimensions or behavior feature dimensions. Attribute feature dimensions refer to dimensions used to describe user attribute characteristics, while behavior feature dimensions refer to dimensions used to describe user behavior characteristics. Attribute feature dimensions may include at least one of gender, age, occupation, city level, or educational background. Behavior feature dimensions may include at least one of monthly active days, video play duration, short video play duration, long video play duration, short video play duration, long video play duration, or user level. The set of candidate feature dimensions may include multiple candidate feature dimensions, where multiple refers to at least two. The candidate feature dimensions in the set of candidate feature dimensions may be pre-set or set as needed. For example, they may be determined based on the source of the target users in the target user group. For example, if the target users are from video applications, the candidate feature dimensions may include at least one of monthly active days, video play duration, short video play duration, long video play duration, short video play duration, long video play duration, or user level.

[0056] Each candidate feature dimension may correspond to a segmented user group set, which may include multiple segmented user groups. The segmented user groups are obtained by segmenting the target user group based on the candidate feature dimension. Users in the segmented user groups belong to the target user group.

[0057] Specifically, the server can use each candidate feature dimension in the candidate feature dimension set to divide the target user group and obtain a set of divided user groups corresponding to each candidate feature dimension. The server can set a corresponding feature threshold for the candidate feature dimension and divide the target user group based on the feature threshold corresponding to the candidate feature dimension. There can be one or more feature thresholds corresponding to the candidate feature dimension, and multiple refers to at least two. For example, the server can determine the current feature dimension from the candidate feature dimension set, obtain the user features corresponding to the current feature dimension of the target user, compare the user features of the current feature dimension with the feature threshold corresponding to the current feature dimension, and divide the target user group according to the comparison result. For example, when the divided user group set includes a first user group and a second user group, and the current feature dimension corresponds to a feature threshold, when the comparison result is that the user features of the current feature dimension are greater than the feature threshold corresponding to the current feature dimension, the target user can be divided into the first user group, and when the comparison result is that the user features of the current feature dimension are less than the feature threshold corresponding to the current feature dimension, the target user can be divided into the second user group. The first user group and the second user group are different user groups. For example, when the current feature dimension is age, the age threshold corresponding to the age can be obtained. The server can obtain the age of the target user and compare the age of the target user with the age threshold. When the age of the target user is greater than the age threshold, the target user is divided into the first user group. When the age of the target user is less than the age threshold, the target user is divided into the second user group.

[0058] S206: Obtain user group features corresponding to each divided user group in the divided user group set.

[0059] Each user group may have a corresponding user group feature. A user group feature refers to a feature corresponding to a user group. The user group feature may be a feature obtained by statistically analyzing user features, wherein the user feature may include at least one of a user attribute feature or a user behavior feature.

[0060] Specifically, the server can perform statistics on the user features of the users in the divided user groups to obtain the target statistical value corresponding to each user feature. The target statistical value can be at least one of the average value corresponding to each user feature, the number of user features, the median of each user feature, or the mode of each user feature. The user group feature is obtained based on each target statistical value. For example, each target statistical value can be arranged and the sequence obtained by arrangement is used as the user group feature. For example, the target statistical value can correspond to a target feature value. The server can obtain the target feature value corresponding to each target statistical value, and obtain the user group feature based on each target feature value. Among them, the target feature value can be pre-set. For example, the target feature value corresponding to the target statistical value within the preset statistical value range can be set to a first value. The target feature value can also be calculated based on the target statistical value. For example, the target statistical value can be normalized and the result of the normalization process is used as the target feature value corresponding to the target statistical value.

[0061] In some embodiments, the target statistical value corresponding to the user feature may include a target attribute statistical value corresponding to the user attribute feature or a target behavior statistical value corresponding to the user behavior feature. The target attribute statistical value may be obtained by statistical calculation based on the user attribute features corresponding to the divided users in the divided user group. For example, the user attribute features corresponding to the divided users in the divided user group may be averaged, and the average value obtained may be used as the target attribute statistical value corresponding to the user attribute feature. The target behavior statistical value may be obtained by statistical calculation based on the user behavior features corresponding to the divided users in the divided user group. For example, the number of users in the divided user group who interact with the same interactive content may be counted, and the target behavior statistical value may be obtained based on the counted number of users. Interactive content refers to content with which users can interact. Interactive content may be content in application software, for example, it may include at least one of video in a video application or audio in an audio application. Interactive behavior refers to the behavioral operation generated on the interactive content. Interactive behavior may include at least one of browsing, reading, clicking, downloading, or sharing the interactive content.

[0062] In some embodiments, the server may arrange each target attribute statistical value and each target behavior statistical value, and use the arranged sequence as the user group feature. The target feature value may include a target attribute feature value corresponding to the target attribute statistical value or a target behavior feature value corresponding to the target behavior statistical value. The server may obtain the target attribute feature value corresponding to the target attribute statistical value and the target behavior feature value corresponding to the target behavior statistical value, arrange each target attribute feature value and each target behavior feature value, and use the arranged sequence as the user group feature. Since user behavior can reflect the user's interest, the sequence obtained by arranging each target attribute feature value and each target behavior feature value can be called an interest vector.

[0063] S208, obtaining feature differences between user group features, and obtaining feature discrimination corresponding to the candidate feature dimensions based on the feature differences, wherein the feature discrimination is positively correlated with the feature differences.

[0064] Among them, feature difference refers to the difference between user group features. Feature discrimination is used to reflect the degree of discrimination between the divided user groups obtained by the candidate feature dimension division. The greater the degree of difference between the divided user groups obtained by the candidate feature dimension division, the greater the feature discrimination, and the smaller the degree of difference between the divided user groups obtained by the candidate feature dimension division, the smaller the feature discrimination. Feature discrimination is positively correlated with feature difference. For example, feature difference can be used as feature discrimination, or a preset value corresponding to feature difference can be used as feature discrimination. The preset value can be pre-set as needed.

[0065] Specifically, the server can obtain the user group features corresponding to each divided user group in the divided user group set obtained based on the current candidate feature dimension as the user group feature set corresponding to the current candidate feature dimension. The current candidate feature dimension can be any candidate feature dimension in the candidate feature dimension set. The server can calculate the feature distance between each user group feature in the user group feature set, and obtain the feature discrimination corresponding to the candidate feature dimension based on the feature distance. For example, the feature distance can be used as the feature discrimination corresponding to the candidate feature dimension, or the feature distance can be normalized and used as the feature discrimination corresponding to the candidate feature dimension. The feature distance refers to the distance between user group features, which can be calculated by the group feature similarity between user group features. The group feature similarity refers to the degree of similarity between user group features. The feature distance is negatively correlated with the group feature similarity. The server can use a distance calculation method to calculate the feature distance between user group features. The distance calculation method can include at least one of a cosine distance calculation method and a Euclidean distance calculation method. The cosine distance calculation method is used to calculate the cosine distance. The Euclidean distance calculation method is used to calculate the Euclidean distance. A negative correlation means that, with other conditions remaining unchanged, two variables change in opposite directions: when one variable changes from large to small, the other variable changes from small to large. While negative correlation refers to the opposite direction of change, it doesn't necessarily mean that a slight change in one variable necessarily requires a change in the other.

[0066] In some embodiments, the user group feature set may include two or more user group features, where multiple means at least three. For example, the user group feature set may include a first user group feature and a second user group feature. The server may calculate the difference between the first user group feature and the second user group feature to obtain the feature difference corresponding to the first user group feature and the second user group feature.

[0067] In some embodiments, the step of calculating the difference between the first user group feature and the second user group feature to obtain the feature difference between the first user group feature and the second user group feature includes: calculating the similarity between the first user group feature and the second user group feature to obtain group feature similarity, and obtaining the corresponding feature difference based on the group feature similarity. The feature difference is negatively correlated with the group feature similarity.

[0068] In some embodiments, the first user group feature and the second user group feature are vectors, and the server can calculate the cosine similarity between the first user group feature and the second user group feature as the group feature similarity. Among them, cosine similarity can also be called cosine distance. Since user behavior can reflect the user's interest, when the user group feature is a vector obtained based on the target behavior feature value. The user group feature can also be called an interest vector. The cosine similarity between user group features can also be called the interest vector cosine distance, and the feature difference can also be called the interest difference. The interest difference is positively correlated with the feature discrimination. The greater the interest difference, the greater the feature discrimination, and the smaller the interest difference, the smaller the feature discrimination.

[0069] S210 , selecting a target feature dimension from the candidate feature dimension set based on feature discrimination, and the target feature dimension is used to classify the target user group.

[0070] Among them, the target feature dimension is a candidate feature dimension selected from the candidate feature dimension set based on feature discrimination. There can be multiple target feature dimensions. Classifying the target user group refers to dividing the target user group. Dividing the user group can be applied to a variety of scenarios, such as scenarios that require artificial recognition of user behavior, scenarios that need to provide different content to different users, scenarios of group recommendations, or scenarios where there are a small number of seed users and it is hoped that the target user will be expanded to more users. Seed users refer to users with relatively typical behaviors. When clustering user groups, they can be used as the starting position for the search, such as the starting centroid of the k-means clustering algorithm.

[0071] Specifically, the server may select a candidate feature dimension that satisfies the discrimination condition from the candidate feature dimension set, and use the candidate feature dimension that satisfies the discrimination condition as the target feature dimension. The discrimination condition includes at least one of the feature discrimination being greater than the discrimination threshold or the discrimination ranking being before the ranking threshold. The discrimination threshold and the ranking threshold may be set as needed or calculated. The discrimination ranking refers to the ranking of the feature discrimination in the feature discrimination sequence. The feature discrimination sequence is a sequence obtained by sorting the feature discrimination in order from large to small. The larger the feature discrimination in the feature discrimination sequence, the higher the ranking.

[0072] In some embodiments, the server can obtain at least one of the feature correlation, feature coverage or value difference between each candidate feature dimension in the candidate feature dimension set, and select the target feature dimension from the candidate feature dimension set based on at least one of the feature correlation, feature coverage or value difference and the feature discrimination. For example, based on the feature discrimination, the candidate feature dimension that meets the discrimination condition can be selected from the candidate feature dimension set, and each candidate feature dimension that meets the discrimination condition can be used as the selected feature dimension to form a selected feature dimension set. The selected feature dimensions in the selected feature dimension set are removed using at least one of the feature correlation, feature coverage or value difference, and the remaining selected feature dimensions in the selected feature dimension set are used as the target feature dimension, where removal means to remove, that is, to discard.

[0073] Among them, feature relevance refers to the degree of association between feature dimensions. The greater the feature relevance, the greater the degree of association between feature dimensions, and the smaller the feature relevance, the smaller the degree of association between feature dimensions. Feature coverage indicates the degree of coverage of feature dimensions and can be calculated based on the ratio of the number of valid users corresponding to the feature dimension to the total number of users. For example, the ratio of the number of valid users corresponding to the feature dimension to the total number of users can be used as feature coverage. Valid users refer to target users in the target user group whose characteristics on the feature dimension are known. The number of valid users refers to the number of valid users in the target user group. The total number of users refers to the total number of all users in the target user group. The value difference reflects the degree of difference between the maximum and minimum features corresponding to a feature dimension. It can be calculated based on the difference between the maximum and minimum features corresponding to the feature dimension. For example, the difference between the maximum and minimum features corresponding to the feature dimension can be used as the value difference, or the difference between the maximum and minimum features corresponding to the feature dimension can be normalized and the normalized result can be used as the value difference. For example, if the feature dimension is "age", the target user group has an age of 70 and a youngest of 10, then the value difference can be 70-10=60. Since the value difference corresponding to a feature dimension is small, the feature dimension may not be able to segment users well. Therefore, removing feature dimensions with small value differences can improve the accuracy of the selected feature dimensions, thereby improving the clustering effect.

[0074] In some embodiments, the server can obtain the feature correlation between the selected feature dimensions in the selected feature dimension set, obtain a selected feature dimension combination whose feature correlation is greater than a correlation threshold, and determine the selected feature dimension to be removed from the selected feature dimension combination. For example, any selected feature dimension can be selected from the selected feature dimension combination as the selected feature dimension to be removed, or a selected feature dimension with a smaller feature coverage in the selected feature dimension combination can be selected as the selected feature dimension to be removed. The selected feature dimension to be removed is removed from the selected feature dimension set, and at least one of the remaining selected feature dimensions in the selected feature dimension set is used as the target feature dimension. For example, all remaining selected feature dimensions in the selected feature dimension set can be directly used as the target feature dimension, or the remaining selected feature dimensions in the selected feature dimension set can be further screened and the target feature dimension determined based on the screening results. For example, when performing further screening, the remaining selected feature dimensions in the selected feature dimension set can be formed into a first set, and the selected feature dimensions whose value difference is less than the difference threshold are removed from the first set, and at least one of the remaining selected feature dimensions in the first set is used as the target feature dimension. For example, all the remaining selected feature dimensions in the first set can be used as target feature dimensions, or a first number of selected feature dimensions can be selected from the remaining selected feature dimensions in the first set as individual target feature dimensions. The first number can be a pre-set fixed value, such as 3. The first number can also be determined based on the number of remaining selected feature dimensions in the first set. For example, the first number can be one-third of the number of remaining selected feature dimensions in the first set. Among them, the correlation threshold can be pre-set or calculated based on the feature correlation between each selected feature dimension. For example, a correlation coefficient can be obtained. The correlation coefficient can be pre-set, and a statistical operation, such as a mean operation, is performed on the feature correlation between each selected feature dimension. The product of the operation result and the correlation coefficient is used as the correlation threshold. The difference threshold can be set as needed or pre-set.

[0075] For example, assuming that the candidate feature dimension set includes feature dimension a, feature dimension b, feature dimension c, feature dimension d, and feature dimension e, if the feature dimensions a, feature dimension b, feature dimension c, feature dimension d, and feature dimension e that meet the discrimination condition are feature dimension a, feature dimension b, feature dimension c, and feature dimension d, then feature dimension a, feature dimension b, feature dimension c, and feature dimension d form the selected feature dimension set. Feature dimensions a, feature dimension b, feature dimension c, and feature dimension d can be combined in pairs to obtain various selected feature dimension combinations. For example, feature dimension a and feature dimension b can form a selected feature dimension combination. If the feature correlation between feature dimension a and feature dimension b is 0.3, the feature correlation between feature dimension a and feature dimension c is 0.1, the feature correlation between feature dimension a and feature dimension d is 0.2, the feature correlation between feature dimension b and feature dimension c is 0.3, the feature correlation between feature dimension b and feature dimension d is 0.4, the feature correlation between feature dimension c and feature dimension d is 0.6, and the correlation threshold is 0.5, since the feature correlation between feature dimension c and feature dimension d is 0.6 greater than the correlation threshold 0.5, then the feature dimension c and feature dimension d can be combined into a Select the one with smaller feature coverage in the feature dimension combination as the selected feature dimension to be removed. For example, if the feature coverage of feature dimension c is 0.8 and the feature coverage of feature dimension d is 0.7, then feature dimension d is used as the selected feature dimension to be removed, and feature dimension d is removed from the selected feature dimension set. The remaining feature dimensions a, feature dimension b and feature dimension c in the selected feature dimension set are combined into a first set. If the value difference of feature dimension c is less than the difference threshold, then feature dimension c is removed from the first set, and the remaining feature dimensions a and feature dimension b in the first set are used as target feature dimensions.

[0076] In some embodiments, the server may use candidate feature dimensions that meet the discrimination condition as selected feature dimensions, form each selected feature dimension into a selected feature dimension set, obtain feature coverage corresponding to the selected feature dimension, remove the selected feature dimensions whose feature coverage is less than the coverage threshold in the selected feature dimension set, and form the remaining selected feature dimensions in the selected feature dimension set into a second set. At least one selected feature dimension in the second set may be used as a target feature dimension. For example, all selected feature dimensions in the second set may be used as target feature dimensions. The selected feature dimensions in the second set may also be further screened, and the target feature dimension may be determined according to the screening results. For example, the server may obtain feature dimensions from the second set. A combination of selected feature dimensions with a correlation greater than a correlation threshold is selected, and the selected feature dimensions to be removed are determined from the combination of selected feature dimensions. The selected feature dimensions to be removed are removed from the second set, and at least one of the remaining selected feature dimensions in the second set is used as the target feature dimension. For example, all the remaining selected feature dimensions in the second set can be used as the target feature dimension, or the remaining selected feature dimensions in the second set can be formed into a third set, and the selected feature dimensions in the third set can be further screened. The target feature dimension is determined based on the screening results. For example, the selected feature dimensions with a value difference less than a difference threshold can be removed from the third set, and at least one of the remaining selected feature dimensions in the third set is used as the target feature dimension. The coverage threshold can be set as needed or pre-set.

[0077] In some embodiments, the server may use candidate feature dimensions that meet the discrimination condition as selected feature dimensions, form each selected feature dimension into a selected feature dimension set, obtain the value difference corresponding to the selected feature dimension, remove the selected feature dimensions whose value difference is less than the difference threshold in the selected feature dimension set, and form the remaining selected feature dimensions in the selected feature dimension set into a fourth set. At least one selected feature dimension in the fourth set may be used as the target feature dimension. For example, all selected feature dimensions in the fourth set may be used as the target feature dimension. The selected feature dimensions in the fourth set may also be further screened, and the target feature dimension may be determined according to the screening results. For example, the server may obtain the selected feature dimensions whose feature correlation is greater than the correlation threshold from the fourth set. Feature dimension combination, determine the selected feature dimensions to be removed from the selected feature dimension combination, remove the selected feature dimensions to be removed in the fourth set, and use at least one of the remaining selected feature dimensions in the fourth set as the target feature dimension. For example, all the remaining selected feature dimensions in the fourth set can be used as the target feature dimension, or the remaining selected feature dimensions in the fourth set can be formed into a fifth set, and the fifth set can be further screened. The target feature dimension is determined according to the screening result. For example, the server can remove the selected feature dimensions whose feature coverage is less than the coverage threshold in the fifth set, and use at least one of the remaining selected feature dimensions in the fifth set as the target feature dimension. For example, all the remaining selected feature dimensions in the fifth set can be used as the target feature dimension.

[0078] In the above-mentioned user data processing method, a target user group to be classified is obtained, the target user group includes multiple target users, and the target user group is divided respectively based on the candidate feature dimensions in the candidate feature dimension set to obtain a set of divided user groups corresponding to the candidate feature dimensions, and the user group features corresponding to each divided user group in the divided user group set are obtained. The feature differences between the user group features are obtained, and the feature discrimination corresponding to the candidate feature dimension is obtained based on the feature differences. The feature discrimination is positively correlated with the feature difference, and the target feature dimension is selected from the candidate feature dimension set based on the feature discrimination. The target feature dimension is used to classify the target user group. Since the feature discrimination is positively correlated with the feature difference, and the feature difference is the difference between the user group features, the greater the difference between the user group features, the greater the feature discrimination. Therefore, based on the feature discrimination, the feature dimension that can distinguish the user groups can be selected, thereby improving the accuracy of the obtained feature dimension. Therefore, when the obtained feature dimension is used to classify the user group, the accuracy of dividing the user group can be improved.

[0079] In some embodiments, obtaining user group characteristics corresponding to each divided user group in the divided user group set includes: performing statistics on multiple user characteristics corresponding to the divided user groups to obtain target statistical values ​​corresponding to each user characteristic; obtaining target feature values ​​corresponding to each target statistical value, arranging each target feature value, and obtaining user group characteristics.

[0080] Specifically, the target statistical value corresponding to the user feature may include a target attribute statistical value corresponding to the user attribute feature or a target behavior statistical value corresponding to the user behavior feature. The target attribute statistical value may be obtained by statistical calculation based on the user attribute features corresponding to the segmented users in the segmented user group. The target behavior statistical value may be obtained by statistical calculation based on the user behavior features corresponding to the segmented users in the segmented user group.

[0081] In some embodiments, the server may perform statistical calculations, such as mean calculations, on the user attribute features corresponding to the divided users in the divided user group, and use the calculated statistical values ​​as the target attribute statistical values ​​corresponding to the user attribute features. For example, the server may obtain user attribute features corresponding to the same attribute feature dimension corresponding to each divided user in the divided user group, form a user attribute feature set, perform statistical calculations, such as mean calculations, on the user attribute feature set, and obtain the target attribute statistical values ​​corresponding to the user attribute features. For example, the server may obtain age features corresponding to the age dimension of each divided user in the divided user group, form an age set, perform mean calculations on the ages in the age set, and obtain the average age corresponding to the age set. The average age may be used as a target attribute statistical value.

[0082] In some embodiments, the server can count the number of users in the divided user group who interact with the same interactive content, and obtain the target behavior statistical value based on the counted number of users. For example, the number of users in the divided user group who interact with the same interactive content can be counted as the first number of interactive users, and the number of all users included in the divided user group can be counted as the number of divided users. Based on the first number of interactive users and the number of divided users, the target behavior statistical value corresponding to the interactive behavior feature (i.e., user behavior feature) is determined. For example, the ratio of the first number of interactive users to the number of divided users can be calculated, and the obtained ratio can be used as the target behavior statistical value corresponding to the interactive behavior feature (i.e., user behavior feature).

[0083] In some embodiments, the target feature value may include a target attribute feature value corresponding to the target attribute statistical value or a target behavior feature value corresponding to the target behavior statistical value. The server may obtain the target attribute feature value corresponding to the target attribute statistical value and the target behavior feature value corresponding to the target behavior statistical value, arrange each target attribute feature value and each target behavior feature value, and use the arranged sequence as the user group feature.

[0084] In some embodiments, the target users in the target user group are from the reference user group. The server can perform statistics on each user feature in the reference user to obtain a first reference statistical value corresponding to each user feature. The target feature value is determined based on the target statistical value and the first reference statistical value. For example, the ratio of the target statistical value to the first reference statistical value can be calculated, and the target feature value is obtained based on the calculated ratio. For example, the target feature value obtained by the calculated ratio can be used as the target feature value, or the range feature value corresponding to the range to which the calculated ratio belongs can be used as the target feature value. The range feature value can be set as needed.

[0085] In this embodiment, statistics are performed on multiple user features corresponding to the divided user groups to obtain target statistical values ​​corresponding to each user feature, target feature values ​​corresponding to each target statistical value are obtained, and each target feature value is arranged to obtain user group features, so that the user group features can accurately reflect the features of the divided user groups, thereby improving the accuracy of the user group features.

[0086] In some embodiments, user characteristics include interactive behavior characteristics, and statistics are performed on multiple user characteristics corresponding to the divided user groups to obtain target statistical values ​​corresponding to each user characteristic, including: statistics are performed on users in the divided user groups who have an interactive correlation relationship with the first interactive content to obtain target statistical values ​​corresponding to the interactive behavior characteristics.

[0087] The interactive behavior characteristics refer to user behavior characteristics. The target statistical value corresponding to the interactive behavior characteristics refers to the target behavior statistical value corresponding to the user behavior characteristics. The first interactive content can be predetermined interactive content or interactive content determined as needed. There can be multiple first interactive contents, and multiple first interactive contents can be interactive content of the same type, for example, multiple first interactive contents can be movie-type videos. Users who have an interactive association relationship with the interactive content refer to users who interact with the interactive content.

[0088] Specifically, the server can count the number of users in the divided user group who have an interactive relationship with the first interactive content to obtain the first number of interactive users, and determine the target behavior statistical value corresponding to the interactive behavior feature based on the first number of interactive users. For example, the first number of interactive users can be used as the target behavior statistical value corresponding to the interactive behavior feature, or the target behavior statistical value corresponding to the interactive behavior feature can be determined based on the first number of interactive users and the number of divided users. For example, the ratio of the first number of interactive users to the number of divided users can be calculated, and the obtained ratio can be used as the target behavior statistical value corresponding to the interactive behavior feature.

[0089] In some embodiments, the server may obtain the number of users in the divided user group who interacted with the same interactive content or the same type of interactive content as the first number of interactive users. For example, the server may obtain multiple first interactive contents to form a first interactive content set, and count the number of users in the divided user group who have an interactive association with at least one first interactive content in the first interactive content set as the first number of interactive users. The first interactive contents in the first interactive content set may be of the same type, such as a movie video.

[0090] In this embodiment, statistics are collected on the users in the divided user group who have an interactive correlation with the first interactive content to obtain a target statistical value corresponding to the interactive behavior feature. Since the number of users who have an interactive correlation with the interactive content can reflect the data characteristics of the interactive behavior feature, statistics are collected on the users who have an interactive correlation with the first interactive content to improve the accuracy of the statistical value of the obtained interactive behavior feature.

[0091] In some embodiments, obtaining the target feature value corresponding to each target statistical value includes: obtaining a first reference statistical value, the first reference statistical value is a statistical value obtained by statistically analyzing the user features corresponding to the reference user group; and obtaining the corresponding target feature value based on the numerical relationship between the target statistical value and the first reference statistical value.

[0092] Among them, the numerical relationship may include a size relationship, and the size relationship may be, for example, a ratio relationship. The reference user group refers to a user group that includes the target user group, that is, the target user in the target user group belongs to the reference user group. The target feature value may be calculated based on the difference between the target statistical value and the first reference statistical value. For example, the difference between the target statistical value and the first reference statistical value may be used as the target feature value, or the difference between the target statistical value and the first reference statistical value may be normalized to obtain the target feature value. The target feature value may also be calculated based on the ratio between the target statistical value and the first reference statistical value. For example, the ratio between the target statistical value and the first reference statistical value may be used as the target feature value, or the difference between the target statistical value and the first reference statistical value may be normalized to obtain the target feature value.

[0093] Specifically, the server may perform statistics on each user feature corresponding to the reference user group to obtain a statistical value corresponding to each user feature as a first reference statistical value. The first reference statistical value may include a first behavior statistical value obtained by performing statistics on user behavior features or a first attribute statistical value obtained by performing statistics on user attribute features. For example, statistics may be performed on users in the reference user group that have an interactive association relationship with the first interactive content, and the calculated statistical results may be used as the first behavior statistical value. Statistics may also be performed on the user attribute features of the reference users in the reference user group, and the calculated statistical results may be used as the first behavior statistical value.

[0094] In some embodiments, the server may perform a division operation on the target statistical value and the first reference statistical value, and determine the target characteristic value corresponding to the target statistical value based on the result of the division operation. For example, the result of the division operation may be used as the target characteristic value corresponding to the target statistical value, or the target characteristic value corresponding to the target statistical value may be determined based on the value range to which the result of the division operation belongs. The target characteristic value may include at least one of a target behavior characteristic value or a target attribute characteristic value. For example, the result obtained by dividing the target attribute statistical value by the first attribute statistical value may be used as the target attribute characteristic value, and the result obtained by dividing the target behavior statistical value by the first behavior statistical value may be used as the target behavior characteristic value.

[0095] In this embodiment, the corresponding target feature value is obtained based on the numerical relationship between the target statistical value and the first reference statistical value. Since the first reference statistical value is a statistical value obtained by statistically analyzing the user features corresponding to the reference user group, the numerical relationship between the target statistical value and the first reference statistical value can reflect the relative relationship of the target statistical value with respect to the first reference statistical value, thereby enabling the obtained target feature value to reflect the relative relationship of the target statistical value with respect to the first reference statistical value, thereby improving the accuracy of the target feature value.

[0096] In some embodiments, user characteristics include interactive behavior characteristics, and statistics are performed on multiple user characteristics corresponding to the divided user groups to obtain target statistical values ​​corresponding to each user characteristic, obtain target feature values ​​corresponding to each target statistical value, and arrange each target feature value. The steps of obtaining user group characteristics include: performing statistics on users in the divided user groups who have an interactive correlation relationship with the first interactive content to obtain target statistical values ​​corresponding to the interactive behavior characteristics, obtaining a first reference statistical value, obtaining corresponding target feature values ​​based on the numerical relationship between the target statistical value and the first reference statistical value, arranging each target feature value, and obtaining user group characteristics.

[0097] In this embodiment, since the first reference statistical value is a statistical value obtained by counting the user characteristics corresponding to the reference user group, and the target statistical value is a statistical value obtained by counting the users in the divided user group who have an interactive correlation relationship with the first interactive content, therefore, the target feature value calculated based on the numerical relationship between the target statistical value and the first reference statistical value can reflect the special characteristics of the divided user group relative to the reference user group, thereby improving the expressiveness of the characteristics and making the obtained user group characteristics more accurate.

[0098] In some embodiments, the numerical relationship includes a ratio relationship, and obtaining the corresponding target characteristic value based on the numerical relationship between the target statistical value and the first reference statistical value includes: performing a ratio operation on the target statistical value and the first reference statistical value to obtain the target ratio; obtaining the target ratio range where the target ratio is located, and taking the range characteristic value corresponding to the target ratio range as the target characteristic value.

[0099] Among them, the target ratio refers to the ratio of the target statistical value to the first reference statistical value. The ratio range is the range of the ratio, which can be expressed by the maximum ratio and the minimum ratio. For example, the ratio range can be [0%, 200%], where 0% in [0, 200%] is the minimum ratio and 200% is the maximum ratio. The target ratio range refers to the value range to which the target ratio belongs. The range characteristic value is used to represent the ratio range. The ratio range and its corresponding range characteristic value can have a one-to-one correspondence. The range characteristic value can be set as needed or pre-set. For example, the range characteristic value of the ratio range [0%, 200%] can be set to 1, and the range characteristic value of the ratio range [200%, 400%] can be set to 2.

[0100] Specifically, the server may use the result obtained by dividing the target statistical value by the first reference statistical value as the target ratio. For example, when the target statistical value is 80% and the first reference statistical value is 40%, the target ratio is 80% ÷ 40% = 200%. The server may determine the candidate ratio range to which the target ratio belongs from the candidate ratio range set as the target ratio range for the target ratio. The candidate ratio range set may include multiple candidate ratio ranges, each of which may be pre-set or set as needed. For example, if the candidate ratio ranges in the candidate ratio range set are [0%, 200%] and [200%, 400%], and the target ratio is 230%, then since 230% belongs to the range [200%, 400%], [200%, 400%] may be used as the target ratio range corresponding to the target ratio of 230%. Each candidate ratio range may correspond to a range characteristic value. Different candidate ratio ranges may correspond to different range characteristic values. The range characteristic value may be pre-set or set as needed. There may be a one-to-one correspondence between the range characteristic value and the candidate ratio range. The range characteristic value corresponding to the candidate ratio range may be determined based on the candidate ratio range, or the candidate ratio range corresponding to the range characteristic value may be determined based on the range characteristic value. The server may use the range characteristic value corresponding to the target ratio range in which the target ratio is located as the target characteristic value.

[0101] In this embodiment, a ratio operation is performed on the target statistical value and the first reference statistical value to obtain a target ratio, a target ratio range in which the target ratio is located is obtained, and the range characteristic value corresponding to the target ratio range is used as the target characteristic value, thereby simplifying the data in the target characteristic value, saving data processing time, and improving data processing efficiency.

[0102] In some embodiments, selecting a target feature dimension from a set of candidate feature dimensions based on feature discrimination includes: selecting a candidate feature dimension that meets a discrimination condition from the set of candidate feature dimensions based on feature discrimination, and using the candidate feature dimension that meets the discrimination condition as the target feature dimension; the discrimination condition includes at least one of the feature discrimination being greater than a discrimination threshold or the discrimination ranking being before a ranking threshold.

[0103] The target feature dimension refers to a candidate feature dimension in the set of candidate feature dimensions that meets the discrimination condition. There can be one or more target feature dimensions, and "multiple" refers to at least two. Discrimination ranking refers to the ranking of feature discrimination in a feature discrimination sequence. The feature discrimination sequence is a sequence obtained by sorting the feature discriminations corresponding to each candidate feature dimension. For example, it can be a sequence obtained by sorting the feature discriminations corresponding to each candidate feature dimension in descending order. The greater the feature discrimination, the higher its ranking in the feature discrimination sequence, i.e., the higher its discrimination ranking.

[0104] The discrimination threshold can be predetermined or dynamically determined based on the feature discrimination corresponding to each candidate feature dimension. For example, the discrimination threshold can be a fixed value such as 0.9 or 0.8. The discrimination threshold can also be calculated based on the feature discrimination corresponding to each candidate feature dimension. For example, the server can perform a mean operation on the feature discrimination corresponding to each candidate feature dimension to obtain a mean discrimination value, obtain a preset discrimination coefficient, multiply the discrimination coefficient by the mean discrimination value, and use the result of the multiplication operation as the discrimination threshold.

[0105] The ranking threshold can be pre-set or determined based on the number of candidate feature dimensions. For example, the ranking threshold can be a fixed value such as 3 or 5, or it can be calculated based on the number of candidate feature dimensions. For example, the server can obtain a ranking coefficient, which can be pre-set, such as 0.5 or 0.6. The ranking coefficient is multiplied by the number of candidate feature dimensions, and the result of the multiplication is used as the ranking threshold.

[0106] Specifically, the server can sort the feature discriminations corresponding to each candidate feature dimension in descending order to obtain a feature discrimination sequence. The greater the feature discrimination, the higher the discrimination ranking, and the smaller the feature discrimination, the lower the discrimination ranking. The discrimination ranking is compared with the sorting threshold. If the discrimination ranking is before the sorting threshold, the feature discrimination corresponding to the discrimination ranking is used as the target feature discrimination, and the candidate feature dimension corresponding to the target feature discrimination is used as the target feature dimension.

[0107] In some embodiments, the server may compare the feature discrimination with a discrimination threshold. When the feature discrimination exceeds the discrimination threshold, the candidate feature dimension corresponding to the feature discrimination is used as the target feature dimension. For example, when the discrimination threshold is 0.8 and the feature discrimination is 0.82, since 0.82 is greater than 0.8, the candidate feature dimension corresponding to the feature discrimination of 0.82 may be used as the target feature dimension.

[0108] In some embodiments, the server may use a candidate feature dimension whose feature discrimination exceeds a discrimination threshold and whose discrimination ranking is before the ranking threshold as a target feature dimension.

[0109] In this embodiment, based on the feature discrimination, candidate feature dimensions that meet the discrimination conditions are selected from the candidate feature dimension set, and the candidate feature dimensions that meet the discrimination conditions are used as target feature dimensions. Since the discrimination conditions include at least one of the feature discrimination being greater than the discrimination threshold or the discrimination ranking being before the ranking threshold, feature dimensions with greater discrimination can be selected, thereby improving the accuracy and rationality of the obtained feature dimensions.

[0110] In some embodiments, selecting a target feature dimension from a set of candidate feature dimensions based on feature discrimination includes: selecting candidate feature dimensions that meet the discrimination condition from the set of candidate feature dimensions based on feature discrimination, and using each candidate feature dimension that meets the discrimination condition as a selected feature dimension to form a selected feature dimension set; the discrimination condition includes at least one of a feature discrimination greater than a discrimination threshold or a discrimination ranking before a ranking threshold; obtaining feature correlations between selected feature dimensions in the set of selected feature dimensions; obtaining a combination of selected feature dimensions whose feature correlations are greater than a correlation threshold, the selected feature dimension combination including a selected feature dimension whose feature correlations are greater than the correlation threshold, determining a selected feature dimension to be removed from the combination of selected feature dimensions, removing the selected feature dimensions to be removed from the set of selected feature dimensions, and using at least one of the remaining selected feature dimensions in the set of selected feature dimensions as a target feature dimension.

[0111] Among them, the selected feature dimension set may include multiple selected feature dimensions, and the selected feature dimension refers to the candidate feature dimension in the candidate feature dimension set that meets the discrimination condition. Feature correlation refers to the degree of correlation between the selected feature dimensions. The greater the feature correlation, the greater the degree of correlation between the selected feature dimensions, and the smaller the feature correlation, the smaller the degree of correlation between the selected feature dimensions. The correlation threshold can be pre-set or calculated based on the feature correlation between each selected feature dimension. For example, a correlation coefficient can be obtained. The correlation coefficient can be pre-set, and a statistical operation, such as a mean operation, is performed on the feature correlation between each selected feature dimension. The product of the operation result and the correlation coefficient is used as the correlation threshold.

[0112] Specifically, the server may select a first selected feature dimension and a second selected feature dimension from the selected feature dimension set to form a selected feature dimension combination, where the first selected feature dimension and the second selected feature dimension are different selected feature dimensions. The server may calculate feature correlations between each selected feature dimension in the selected feature dimension combination to obtain feature correlations corresponding to each selected feature dimension combination.

[0113] In some embodiments, the step of calculating the feature correlation between each selected feature dimension in the selected feature dimension combination includes: obtaining the features of each target user in the target user group on the first selected feature dimension to form a first feature set, obtaining the features of each target user in the target user group on the second selected feature dimension to form a second feature set, and calculating the set correlation between the first feature set and the second feature set as the feature correlation between each selected feature dimension in the selected feature dimension combination. For example, if the first selected feature dimension is the number of days of short video playback and the second selected feature dimension is the number of days of long video playback, then the number of days of short video playback corresponding to each target user in the target user group can be obtained, and each short video playback day can be grouped into a short video playback day set, and the number of long video playback days corresponding to each target user in the target user group can be obtained, and each long video playback day can be grouped into a long video playback day set, and the correlation between the short video playback day set and the long video playback day set can be calculated to obtain the feature correlation between the short video playback day set and the long video playback day set.

[0114] In some embodiments, the step of calculating the set correlation between the first feature set and the second feature set includes: performing a statistical operation on features in the first feature set to obtain a first feature statistical value, performing a statistical operation on features in the second feature set to obtain a second feature statistical value, and performing a correlation calculation based on the first feature statistical value and the second feature statistical value to obtain the set correlation between the first feature set and the second feature set. The first feature statistical value may include a first feature mean, and the second feature statistical value may include a second feature mean. Specifically, the server can perform a mean operation on the features in the first feature set to obtain a first feature mean, perform a mean operation on the features in the second feature set to obtain a second feature mean, calculate the difference between each first feature in the first feature set and the first feature mean to obtain a first feature difference value corresponding to each first feature, sum up the first feature difference values ​​to obtain a first sum value, calculate the difference between each second feature in the second feature set and the second feature mean to obtain a second feature difference value corresponding to each second feature, sum up the second feature difference values ​​to obtain a second sum value, and obtain a set correlation between the first feature set and the second feature set based on the first sum value and the second sum value. The set correlation is positively correlated with the first sum value, and the set correlation is positively correlated with the second sum value. Among them, the first feature refers to the feature in the first feature set. The second feature refers to the feature in the second feature set.

[0115] In some embodiments, the server may compare the feature correlation between the selected feature dimensions in the selected feature dimension combination (recorded as the feature correlation corresponding to the selected feature dimension combination) with the correlation threshold. When the feature correlation corresponding to the selected feature dimension combination is greater than the correlation threshold, the server determines the selected feature dimension to be removed from the selected feature dimension combination, removes the selected feature dimension to be removed from the selected feature dimension set, thereby achieving the removal processing of the selected feature dimensions in the selected feature dimension set, and at least one of the remaining selected feature dimensions in the selected feature dimension set can be used as the target feature dimension. The step of determining the selected feature dimension to be removed from the selected feature dimension combination may include: randomly obtaining a selected feature dimension from the selected feature dimension combination as the selected feature dimension to be removed.

[0116] In this embodiment, when the correlation of certain features is relatively large, for example, it can be inferred from the principle of k-means algorithm that when the correlation of certain features is relatively large, such as when the correlation between members and active days is relatively large, the influence of features with relatively large correlation on user group division may become greater, thereby causing the results of user group division to deviate from the actual situation, resulting in a decrease in the accuracy of user group division. In the embodiment of the present application, feature dimensions are screened in combination with feature discrimination and feature correlation, so that the obtained feature dimensions not only have good discrimination, but also maintain rationality in the correlation between the obtained feature dimensions, thereby improving the accuracy of the obtained feature dimensions.

[0117] In some embodiments, determining the selected feature dimensions to be removed from the selected feature dimension combination includes: obtaining the feature coverage of each selected feature dimension in the selected feature dimension combination in the target user group; and using the selected feature dimensions with smaller feature coverage in the selected feature dimension combination as the selected feature dimensions to be removed.

[0118] Among them, feature coverage refers to the ratio of the number of valid users corresponding to the selected feature dimension to the total number of users. The number of valid users corresponding to the selected feature dimension refers to the number of valid users corresponding to the selected feature dimension in the target user group. Valid users refer to target users in the target user group whose characteristics on the selected feature dimension are known. For example, when the selected feature dimension is age, when the age of the target user is known, the target user is a valid user, and when the age of the target user is unknown, the target user is not a valid user. The total number of users refers to the total number of all users in the target user group. For example, assuming that the selected feature dimension is age, the target user group includes 100 target users, of which the ages of 90 target users are known, and the ages of the remaining 10 target users are unknown, then the feature coverage corresponding to age can be expressed as 90÷100=90%.

[0119] Specifically, the server can obtain the feature coverage corresponding to the first selected feature dimension in the selected feature dimension combination as the first feature coverage, obtain the feature coverage corresponding to the second selected feature dimension in the selected feature dimension combination as the second feature coverage, compare the first feature coverage with the second feature coverage, and when it is determined that the first feature coverage is greater than the second feature coverage, use the second selected feature dimension as the selected feature dimension to be removed; when it is determined that the first feature coverage is less than the second feature coverage, use the first selected feature dimension as the selected feature dimension to be removed.

[0120] like Figure 3 As shown, the selected feature dimension set includes 13 selected feature dimensions, namely age, gender, occupation, city level, education level, whether campus network user, monthly active days, video playback time, short video playback days, long video playback days, short video playback time, long video playback time and user level. 0.0-0.1 means no correlation, 0.1-0.3 means weak correlation, 0.3-0.5 means medium correlation, 0.5-1.0 means strong correlation. Figure 3 It can be seen that the correlation between age and whether or not a user is a campus network user is 0.003, which means that there is no correlation between age and whether or not a user is a campus network user. The correlation between monthly active days and whether or not a user is a campus network user is 0.175, which means that there is a weak correlation between monthly active days and whether or not a user is a campus network user. The correlation between user level and video playback time is 0.301, which means that there is a medium correlation between user level and video playback time. The correlation between occupation and academic qualifications is 0.811, which means that there is a strong correlation between occupation and academic qualifications. At this time, the occupation coverage can be compared with the academic qualifications coverage. When the occupation coverage is greater than the academic qualifications coverage, the occupation can be used as the target feature dimension. When the occupation coverage is less than the academic qualifications coverage, the academic qualifications can be used as the target feature dimension. It should be noted that Figure 3 The absolute value of the value in is the correlation.

[0121] In this embodiment, the feature coverage of each selected feature dimension in the selected feature dimension combination in the target user group is obtained, and the selected feature dimensions with smaller feature coverage in the selected feature dimension combination are used as selected feature dimensions to be removed. Since the greater the coverage of the feature dimension, the better the expressive ability of the feature, the feature dimension with smaller coverage is removed to obtain the feature dimension with higher coverage, thereby improving the accuracy of the feature dimension.

[0122] In some embodiments, obtaining feature differences between user group features and obtaining feature discrimination corresponding to candidate feature dimensions based on the feature differences includes: calculating feature distances between user group features, and using the calculated feature distances as feature differences between user group features; and using the feature distances as feature discrimination corresponding to candidate feature dimensions.

[0123] Specifically, feature distance refers to the distance between user group features. The server can use a distance calculation method to calculate the feature distance between user group features. The distance calculation method may include at least one of a cosine distance calculation method or a Euclidean distance calculation method. The cosine distance calculation method is used to calculate the cosine distance. The Euclidean distance calculation method is used to calculate the Euclidean distance. For example, the server can calculate the cosine similarity between user group features to obtain feature similarity, and obtain the feature distance between user group features based on the feature similarity. The feature distance is negatively correlated with the feature similarity. The greater the feature similarity, the smaller the feature distance, and the smaller the feature similarity, the larger the feature distance. The server can use the feature distance between user group features as the feature difference between user group features, and can use the feature distance between user group features as the feature discrimination corresponding to the candidate feature dimension.

[0124] In this embodiment, since the feature distance between user group features can reflect the difference between user group features, the larger the feature distance, the greater the difference between user group features. Therefore, the feature distance is used as the feature discrimination corresponding to the feature, which improves the accuracy of the feature discrimination.

[0125] In some embodiments, the method further includes: classifying the target user group based on the target feature dimension to obtain a target sub-user group; performing statistics on the user features of users in the target sub-user group, and obtaining user labels corresponding to the target sub-user group based on the statistically obtained user statistical features.

[0126] Among them, the target sub-user group is a user group obtained by dividing the target user group based on the target feature dimension. The target sub-user group may include multiple, and multiple means at least two. The user statistical feature may include at least one of an attribute statistical feature or a behavior statistical feature. The attribute statistical feature is a statistical value obtained by counting the user attribute features of the users in the target sub-user group, and the behavior statistical feature is a statistical value obtained by counting the user behavior features of the users in the target sub-user group. The user tag corresponding to the target sub-user group is used to indicate the attributes or behaviors of the users in the target sub-user group. The target sub-user group may have multiple user tags corresponding to the target sub-user group, and multiple means at least two. The user tag may include at least one of a tag determined based on the attribute statistical feature or a tag determined based on the behavior statistical feature.

[0127] Specifically, the server can generate target user features corresponding to the target user based on the target feature dimension, and classify the target user group based on the target user features corresponding to each target user. The server can obtain the features of the target user in each target feature dimension as the target features corresponding to the target user, arrange the target features, and use the arranged sequence as the target user features corresponding to the target user. The target feature dimension can include at least one feature dimension of the attribute feature dimension or the behavior feature dimension. For example, when the target feature dimension includes age, gender, education level, short video playback time, and long video playback time, the features corresponding to the target user's age, gender, education level, short video playback time, and long video playback time can be obtained as target features, and these target features can be arranged to obtain the target user features corresponding to the target user. When arranging the target features, they can be randomly arranged or arranged according to preset arrangement rules. The preset arrangement rules can be set as needed. For example, the preset arrangement rules can include arranging the features corresponding to the attribute feature dimension before the features of the behavior feature dimension.

[0128] In some embodiments, the server may perform a mean operation on the user attribute characteristics of the users in the target sub-user group, and use the result of the mean operation as the attribute statistical characteristic; perform a mean operation on the user behavior characteristics of the users in the target sub-user group, and use the result of the mean operation as the behavior statistical characteristic; and use the attribute statistical characteristic and the behavior statistical characteristic as the user tags corresponding to the target sub-user group. For example, the server may perform a mean operation on the age of each user in the target sub-user group to obtain the average age, and the average age can be used as an attribute statistical characteristic; the server may perform a mean operation on the movie viewing time of each user in the target sub-user group to obtain the average movie viewing time, and the average movie viewing time can be used as a behavior statistical characteristic, and the average age and the average movie viewing time can be used as the user tags corresponding to the target sub-user group.

[0129] In some embodiments, the target user set can be clustered based on the target user features corresponding to each target user, resulting in clustered sub-user sets as target sub-user groups. For example, the server can use a clustering algorithm to perform cluster analysis on the target user features corresponding to each target user to obtain target sub-user groups. The clustering algorithm used in the cluster analysis can be any clustering algorithm, such as a machine learning-based clustering algorithm, such as k-means. K-means offers good interpretability and computational performance. The following example illustrates the principle of the k-means algorithm: When a data set consisting of N data needs to be clustered into K clusters, the first step is to randomly select K data as centroids. A centroid represents the mean of a cluster. The centroid can be data in the data set or data outside the data set; the second step is to calculate the distance between each data in the data set and each centroid, and obtain the K centroid distances corresponding to each data in the data set. The cluster represented by the centroid corresponding to the minimum distance among these K centroid distances is used as the cluster to which the data belongs, thereby obtaining K clusters; the third step is to calculate the mean of the data in each cluster as the centroid of the next round; the fourth step is to repeat steps 1 to 3 until the centroid of the next round is equal to the centroid of this round or the difference between the centroid of the next round and the centroid of this round is less than the centroid difference threshold, and the obtained K clusters are used as the clustering results corresponding to the data set. The centroid difference threshold can be set as needed and can be pre-set. Among them, when calculating the distance between the data and the centroid, the Euclidean distance can be used. For example, the formula can be used. Calculate the distance between the data and the centroid, where x represents any data in the data set, μ k represents the kth centroid among K centroids, k is greater than or equal to 1 and less than or equal to K, d represents the difference between x and μ k The distance between them. Figure 4 As shown, the clustering principle of the k-means algorithm is demonstrated. Figure 4 The data in the data set shown in (a) Figure 4 The “×” in (b)-(f) represents the center of mass. It can be seen from the figure that the center of mass is changing and the accuracy of the center of mass is constantly improving.

[0130] In some embodiments, the server can obtain a trained user group identification model, and the user group identification model is used to identify the user group to which the user belongs based on the relevant data of the user. The user group identification model can be a neural network model based on artificial intelligence, for example, it can be a model based on a convolutional network. The server can input the target user features into the user group identification model, and the user group identification model can perform convolution processing on the target user features to obtain convolution features, and use the convolution features to determine the probability that the target user belongs to each preset user group, and use the preset group corresponding to the largest probability among the various probabilities as the target user group corresponding to the target user. The preset user group refers to the user group that can be identified by the user group identification model. There can be one or more preset user groups, and multiple refers to at least two.

[0131] In some embodiments, the server can expand the target sub-user group to obtain an expanded target sub-user group. For example, a user in the target sub-user group can be used as a seed user, and users whose similarity with the seed user is greater than a similarity threshold are obtained and added to the target sub-user group to obtain the expanded target sub-user group. Wherein, the seed user refers to a relatively typical user. The similarity threshold can be pre-set or set as needed, for example, it can be 90%. For example, the server can obtain a candidate user, generate a candidate user feature corresponding to the candidate user based on the target feature dimension, calculate the similarity between the candidate user feature and the user statistical feature, and when the similarity is greater than the similarity threshold, add the candidate user to the target sub-user group. The server can also expand the target sub-user group based on the user tag corresponding to the target sub-user group. For example, the server can generate a candidate user tag corresponding to the candidate user, calculate the similarity between the candidate user tag corresponding to the candidate user and the user tag corresponding to the target sub-user group, and when the calculated similarity is greater than the similarity threshold, add the candidate user to the target sub-user group. The server can also utilize lookalike (similar population expansion) technology to achieve the expansion of the user group. Lookalike technology uses a seed user model to identify similar groups of people with potential connections through an algorithmic evaluation model. Lookalike technology is not a specific algorithm, but a general term for a class of methods.

[0132] In this embodiment, since the statistical values ​​corresponding to the information of users in the target sub-user group can reflect the characteristics of the target sub-user group, the user characteristics in the target sub-user group are counted, and the user labels corresponding to the target sub-user group are obtained based on the statistically obtained user statistical characteristics. This can make the user labels accurately reflect the characteristics of the user group, thereby improving the accuracy of the obtained user labels.

[0133] In some embodiments, statistics are performed on the information of users in the target sub-user group, and user tags corresponding to the target sub-user group are obtained based on the user information obtained by statistics, including: statistics are performed on users in the target sub-user group who have an interactive association relationship with the second interactive content, and user statistical values ​​corresponding to the second interactive content are obtained; a second reference statistical value corresponding to the second interactive content is obtained, and a statistical difference value between the user statistical value and the second reference statistical value is obtained; when the statistical difference value is greater than a difference threshold, the user tag corresponding to the target sub-user group is obtained based on the content attribute corresponding to the second interactive content, so as to push content based on the user tag.

[0134] Among them, the second interactive content can be any interactive content or a specified interactive content. The second reference statistical value is a statistical value obtained by counting the users in the reference user group who have interactive behaviors on the second interactive content. The difference threshold can be set as needed or pre-set. Content attributes refer to the attributes of the interactive content. Content attributes may include at least one of the interactive content type, name, key information or key figures. For example, when the interactive content is the idol TV series "ABC" starring Zhang San, the content type may be a TV series with the name "ABC", the key figure may be Zhang San, and the key information may be an idol drama. The difference statistical value may be the difference between the user statistical value and the second reference statistical value, or it may be the ratio between the user statistical value and the second reference statistical value.

[0135] Specifically, the server may calculate the number of users in the target sub-user group who interacted with the second interactive content to obtain the second interactive user number, calculate the total number of all users in the target sub-user group to obtain the number of sub-users, calculate the ratio of the second interactive user number to the number of sub-users, and use the obtained ratio as the user statistics corresponding to the second interactive content. The server may also count the number of users in the reference user group who interacted with the second interactive content as the second reference statistics.

[0136] In some embodiments, the server can calculate the ratio of the user statistical value to the second reference statistical value, and use the calculated ratio as the statistical difference value. The statistical difference value can be compared with the difference threshold. When it is determined that the statistical difference value is greater than the difference threshold, it indicates that the target sub-user group has a greater degree of interest in the second interactive content. The server can obtain the content attributes of the second interactive content and use the content attributes of the second interactive content as the user tag corresponding to the target sub-user group, so as to recommend appropriate content to users in the target sub-user group based on the user tag of the target sub-user group.

[0137] In this embodiment, the statistical difference value is greater than the difference threshold, which can reflect that the interactive correlation relationship between users in the target sub-user group for the second interactive content is relatively close. The user tag corresponding to the target sub-user group is obtained based on the content attribute corresponding to the second interactive content, which can reflect the target sub-user group's preference for the second interactive content. Therefore, content push is performed based on the user tag, and the second interactive content or content whose content difference with the second interactive content is less than the content difference threshold can be pushed to the user, thereby improving the accuracy of content push.

[0138] In some embodiments, statistics are performed on the user characteristics of users in the target sub-user group, and user tags corresponding to the target sub-user group are obtained based on the statistically obtained user statistical characteristics, including: statistics are performed on the attribute characteristics of the users in the target sub-user group to obtain the attribute statistical characteristics corresponding to the target sub-user group, and the attribute statistical characteristics are used as the user tags corresponding to the target sub-user group; when an attribute characteristic acquisition request for users in the target sub-user group is received, the attribute statistical characteristics are sent to the device that sends the attribute characteristic acquisition request.

[0139] The attribute feature acquisition request is used to request the user's attribute features. The device sending the attribute feature acquisition request can be a terminal or a server. For example, the device sending the attribute feature acquisition request can be a terminal, and the terminal can display the attribute statistical features.

[0140] Specifically, when the terminal receives a user data acquisition operation for a target sub-user group triggered by a user, it can send an attribute feature acquisition request for the target sub-user group to the server. The server can respond to the attribute feature acquisition request, obtain the attribute statistical features of the target sub-user group, and return the attribute statistical features of the target sub-user group to the terminal. The terminal can display the attribute statistical features of the target sub-user group.

[0141] In some embodiments, the server may also generate attribute prompt information based on attribute statistical features, and return the attribute prompt information to the terminal, which may then display the target sub-user group and the corresponding attribute prompt information. Figure 5As shown, when the terminal receives a trigger operation for the control "Viewing User Analysis", it can send a request to the server to obtain the attribute characteristics of the user corresponding to the target video. The server can obtain the users corresponding to the target video and form a target user group to be classified. The server can store the user data processing method provided by this application. The target characteristic dimension can be determined based on the user data processing method. The target user group can be divided based on the target characteristic dimension to obtain a target sub-user group. The attribute statistical characteristics of the target sub-user group can be counted, and corresponding attribute prompt information can be generated based on the attribute statistical characteristics. The attribute prompt information 500 is returned to the terminal. Among them, the attribute prompt information 500 is "Users who watch the target video can be divided into the following three categories: 1. Male students in small towns, 2. Young women in the workplace, 3. Middle-aged uncles who follow dramas."

[0142] In some embodiments, the server can also obtain the attribute statistical characteristics and behavioral statistical characteristics of the target sub-user group in response to the attribute feature acquisition request, and return the attribute statistical characteristics and behavioral statistical characteristics of the target sub-user group to the terminal. The terminal can display the attribute statistical characteristics and behavioral statistical characteristics of the target sub-user group accordingly, so that the user's preferences can be intuitively obtained while protecting the user's privacy.

[0143] In some embodiments, when the server receives a content push request for target content, it can obtain the target content attributes corresponding to the target content, match the target content data with the user tags corresponding to the target sub-user group, and when it is determined that the target content attributes match the user tags corresponding to the target sub-user group, push the target content to the terminals corresponding to the users in the target sub-user group. Figure 6 As shown, the server Figure 5 The user behavior characteristics of the three user groups in the example are analyzed, and the corresponding behavioral statistical characteristics of each user group are obtained. The corresponding behavioral statistical characteristics of the "small town male student" are sports, games, and animation, the corresponding behavioral statistical characteristics of the "young career woman" are idol dramas, variety shows, and entertainment, and the corresponding behavioral statistical characteristics of the "middle-aged drama-chasing uncle" are military affairs, finance, and fitness. When the terminal receives a trigger operation on the "push" control 602, it can send a video push request for video 1, video 2, and video 3 to the server. The server can analyze video 1, video 2, and video 3. Assuming that the analysis determines that video 1 belongs to the animation category, video 2 belongs to the idol drama category, and video 3 belongs to the military category, video 1 can be pushed to the terminal of the "small town male student", video 2 can be pushed to the terminal of the "young career woman", and video 3 can be pushed to the terminal of the "middle-aged drama-chasing uncle".

[0144] In this embodiment, the attribute characteristics of the users in the target sub-user group are counted to obtain the attribute statistical characteristics corresponding to the target sub-user group, and the attribute statistical characteristics are used as the user labels corresponding to the target sub-user group, thereby achieving high-quality reconstruction of the user gender and age. When an attribute feature acquisition request for the users in the target sub-user group is received, the attribute statistical characteristics are sent to the device that sends the attribute feature acquisition request. Since the attribute statistical characteristics are sent to the device that sends the attribute feature acquisition request, rather than the attribute characteristics of the real user, the leakage of user data is avoided, the user's privacy is protected, and the security of the user data is improved.

[0145] In some embodiments, a method for processing user data is provided, comprising the following steps:

[0146] 1. The server receives a user analysis request for the target video.

[0147] Among them, Figure 7 As shown in FIG, when the terminal obtains the user analysis trigger operation, it can send a user analysis request for the target video to the server, and the user analysis request can carry the identifier of the target video. Figure 7 In the example, when the terminal obtains the trigger operation for the control "view user analysis", it can be determined that the user analysis trigger operation is obtained.

[0148] 2. The server responds to the user analysis request, obtains the users corresponding to the target video, and forms a target user group to be classified.

[0149] The user corresponding to the target video may be at least one of a user who views the target video, a user who collects the target video, or a user who shares the target video. The server may store the user data processing method proposed in this application. Based on the user data processing method proposed in this application, a target feature dimension is determined, and the target user group is classified based on the target feature dimension to obtain target sub-user groups.

[0150] 3. The server obtains the original user data corresponding to the target user, preprocesses the original user data, and obtains user attribute characteristics and user behavior characteristics based on the preprocessed data.

[0151] The pre-processing of the user original data may include data filtering, and the data filtering may include at least one filtering method of missing data filtering or abnormal data filtering. The "filtering" means to discard or discard.

[0152] User raw data can be obtained from application software. For example, application software can record the operations performed by users on the application software. The server can obtain the user's historical behavior data and user attribute data from the server where the application software is located, pre-process the user's historical behavior data and user attribute data to obtain pre-processed user behavior data and user behavior data, obtain user behavior features based on the pre-processed user behavior data, and obtain user attribute features based on the pre-processed user attribute data. User attribute features can include, for example Figure 8 Gender, age, education level and city level in the user behavior characteristics may include Figure 8 Monthly active days, user level, and tag interests. Tag interests refer to tags related to interests and can include at least one of video tags or music tags.

[0153] 4. The server determines a set of candidate feature dimensions based on user behavior characteristics and user attribute characteristics.

[0154] Specifically, the server can obtain the dimensions corresponding to the user behavior characteristics and obtain each behavior characteristic dimension, and can obtain the dimensions corresponding to the user attribute characteristics and obtain each attribute characteristic dimension. Based on each behavior characteristic dimension and each attribute characteristic dimension, a candidate feature dimension set is obtained. For example, the server can use the set consisting of each behavior characteristic dimension and each attribute characteristic dimension as a candidate feature set, or it can select a preset number of dimensions from each behavior characteristic dimension and each attribute characteristic dimension to form a candidate feature dimension set. The preset number can be set as needed, for example, it can be 10.

[0155] 5. The server divides the target user group based on the candidate feature dimensions in the candidate feature dimension set, obtains a set of divided user groups obtained by dividing each candidate feature dimension, obtains user group features corresponding to each divided user group in the divided user group set, obtains feature differences between the user group features, and obtains feature discrimination corresponding to the candidate feature dimensions based on the feature differences.

[0156] Among them, feature discrimination is positively correlated with feature difference.

[0157] 6. The server obtains the feature correlation between the candidate feature dimensions in the candidate feature dimension set.

[0158] For example, the Pearson correlation coefficient calculation formula may be used to calculate the feature correlation between the candidate feature dimensions.

[0159] 7. The server obtains the feature coverage of the candidate feature dimensions in the candidate feature dimension set in the target user group;

[0160] 8. The server obtains the value difference of the candidate feature dimensions in the candidate feature dimension set in the target user group;

[0161] 9. The server selects a target feature dimension from the candidate feature dimension set based on at least one of feature discrimination, feature correlation, feature coverage, or value difference.

[0162] Among them, Figure 8 As shown, the server can screen the candidate feature dimensions in the candidate feature dimension set based on feature discrimination, feature correlation, feature coverage, and value difference to obtain the target feature dimension.

[0163] 10. The server determines the target user characteristics corresponding to the target users in the target user group based on the target characteristic dimension, divides the target user group into user groups based on each target user characteristic, and uses each user group obtained after the division as the target sub-user group.

[0164] Among them, Figure 8 As shown, the server may cluster the target user group using a clustering algorithm to achieve division of the target user group. For example, the server may cluster the target user group using a clustering algorithm K-means.

[0165] 11. The server analyzes the users in the target sub-user group, obtains the user analysis results corresponding to the target sub-user group, and returns the user analysis results to the terminal so that the terminal can display the target sub-user group and the user analysis results accordingly.

[0166] The user analysis results may include the average value of the attribute features corresponding to the target sub-user group, and may also include the degree to which the target sub-user group likes the target video. The user analysis results may include, for example, Figure 8 At least one of the demographic attributes, activity status, or special interest preference in the target content. The special interest preference of the target sub-user group for the target content refers to the ratio of the target sub-user group's preference for the target content to the reference user group's preference for the target user. For example, if the target sub-user group's preference for movies is 80 points, and the reference user group's preference for movies is 40 points, then the target sub-user group's special interest preference for movies can be 80 / 40=2. For example, in Figure 7 In the example, the server divides the target user group composed of users corresponding to the target video into three target sub-user groups. By analyzing these three target sub-user groups, the three target sub-user groups are respectively determined to be the "small town male student" user group, the "young working woman" user group, and the "middle-aged uncle who chases dramas" user group. The "small town", "male", and "student" in the "small town male student" are attribute statistical information. Figure 7The stars in the chart represent the user group's preference for the target video. The more stars there are, the more they like the target video. Since "Male students in small towns" has 3 stars, "Young working women" has 2 stars, and "Middle-aged uncles who chase dramas" has 1 star, we can know that the target video's target audience is "male students in small towns". Therefore, when pushing the target video, we can focus on pushing the target video to this type of users, thereby improving the accuracy of video push.

[0167] In this embodiment, feature dimensions with high feature coverage, large feature differentiation and independence can be retained, so that the feature dimensions can be optimized, the accuracy of user group segmentation can be improved, and niche user groups can be identified. Niche user groups can, for example, include at least one of "male students in small towns", "middle-aged uncles who chase dramas" or "young working women".

[0168] This application also provides an application scenario, which applies the above-mentioned user data processing method. Specifically, the application of the user data processing method in this application scenario is as follows:

[0169] The server can obtain a target user group to be classified, use the user data processing method provided in this application to obtain a target feature dimension, use the target feature dimension to divide the target user group into user groups, obtain target sub-user groups, analyze each target sub-user group, and obtain user tags. The user tags are used to indicate the attributes or behaviors of the user. The user tags may include at least one of a gender tag, an age tag, or an educational level tag. Based on the user tags, a user profile is obtained. The user profile may include multiple user tags.

[0170] The user data processing method provided in this application is applied to constructing user portraits, which can improve the accuracy of user portraits.

[0171] This application also provides an application scenario, which applies the above-mentioned user data processing method. Specifically, the application of the user data processing method in this application scenario is as follows:

[0172] The server can obtain each user to whom the video is to be pushed, form a video user group, divide the video user group into user groups using the user data processing method provided in this application, obtain each divided video sub-user group, analyze the video tags corresponding to each video sub-user group, determine the video content that matches the video sub-user group based on the video tags corresponding to the video sub-user group, and push the matching video content to the terminals corresponding to each user in the video sub-user group.

[0173] The user data processing method provided in this application is applied to assist recommendation systems in content push, such as video push, and can improve the accuracy of content push.

[0174] This application also provides an application scenario, which applies the above-mentioned user data processing method. Specifically, the application of the user data processing method in this application scenario is as follows:

[0175] The server can obtain each user to whom advertisements are to be delivered, form an advertising user group, divide the advertising user group into user groups using the user data processing method provided in this application, obtain each divided advertising sub-user group, analyze the advertising tags corresponding to each advertising sub-user group, determine the advertising content that matches the advertising sub-user group based on the advertising tags corresponding to the advertising sub-user group, and push the matching advertising content to the terminals corresponding to each user in the advertising sub-user group.

[0176] The user data processing method provided in this application is applied to assist in advertising delivery and can improve the accuracy of advertising delivery.

[0177] It should be understood that although Figure 2-8 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 2-8 At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The order of execution of these steps or stages is not necessarily one by one, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0178] In some embodiments, as Figure 9As shown, a user data processing method and apparatus is provided. The apparatus can be implemented as a software module or a hardware module, or a combination of both, as part of a computer device. The apparatus specifically includes: a target user group acquisition module 902, a module for obtaining a divided user group set 904, a user group feature acquisition module 906, a feature difference acquisition module 908, and a target feature dimension acquisition module 910. The target user group acquisition module 902 is configured to obtain a target user group to be classified; the target user group includes multiple target users. The divided user group set acquisition module 904 is configured to divide the target user group based on candidate feature dimensions in a candidate feature dimension set, thereby obtaining a set of divided user groups corresponding to the candidate feature dimensions. The user group feature acquisition module 906 is configured to obtain user group features corresponding to each divided user group in the divided user group set. The feature difference acquisition module 908 is configured to obtain feature differences between user group features and, based on the feature differences, obtain feature discrimination corresponding to the candidate feature dimensions. The feature discrimination is positively correlated with the feature differences. The target feature dimension obtaining module 910 is used to select a target feature dimension from a set of candidate feature dimensions based on feature discrimination, and the target feature dimension is used to classify the target user group.

[0179] In the above-mentioned user data processing device, a target user group to be classified is obtained, the target user group includes multiple target users, and the target user group is divided based on the candidate feature dimensions in the candidate feature dimension set to obtain a set of divided user groups corresponding to the candidate feature dimensions, and the user group features corresponding to each divided user group in the divided user group set are obtained. The feature differences between the user group features are obtained, and the feature discrimination corresponding to the candidate feature dimension is obtained based on the feature differences. The feature discrimination is positively correlated with the feature difference. The target feature dimension is selected from the candidate feature dimension set based on the feature discrimination, and the target feature dimension is used to classify the target user group. Since the feature discrimination is positively correlated with the feature difference, and the feature difference is the difference between the user group features, the greater the difference between the user group features, the greater the feature discrimination. Therefore, based on the feature discrimination, the feature dimension that can distinguish the user groups can be selected, thereby improving the accuracy of the obtained feature dimension. Therefore, when the obtained feature dimension is used to classify the user group, the accuracy of dividing the user group can be improved.

[0180] In some embodiments, the user group feature acquisition module 906 includes a target statistical value acquisition unit configured to perform statistics on multiple user features corresponding to the user groups to obtain target statistical values ​​corresponding to each user feature. The user group feature acquisition unit is configured to obtain target feature values ​​corresponding to each target statistical value, and to arrange the target feature values ​​to obtain user group features.

[0181] In some embodiments, the user features include interactive behavior features, and the target statistical value obtaining unit is further used to count users in the divided user group who have an interactive correlation relationship with the first interactive content to obtain a target statistical value corresponding to the interactive behavior feature.

[0182] In some embodiments, the user group feature obtaining unit is also used to obtain a first reference statistical value, which is a statistical value obtained by statistically analyzing the user features corresponding to the reference user group; and obtain the corresponding target feature value based on the numerical relationship between the target statistical value and the first reference statistical value.

[0183] In some embodiments, the numerical relationship includes a ratio relationship, and the user group feature obtaining unit is also used to perform a ratio operation on the target statistical value and the first reference statistical value to obtain a target ratio; obtain the target ratio range where the target ratio is located, and use the range characteristic value corresponding to the target ratio range as the target characteristic value.

[0184] In some embodiments, the target feature dimension obtaining module 910 is also used to select a candidate feature dimension that meets the discrimination condition from the candidate feature dimension set based on the feature discrimination, and use the candidate feature dimension that meets the discrimination condition as the target feature dimension; the discrimination condition includes at least one of the feature discrimination being greater than the discrimination threshold or the discrimination ranking being before the ranking threshold.

[0185] In some embodiments, the target feature dimension obtaining module 910 is further used to select candidate feature dimensions that meet the discrimination conditions from the candidate feature dimension set based on feature discrimination, and use each candidate feature dimension that meets the discrimination conditions as a selected feature dimension to form a selected feature dimension set; the discrimination condition includes at least one of the feature discrimination being greater than the discrimination threshold or the discrimination ranking being before the ranking threshold; obtain the feature correlation between the selected feature dimensions in the selected feature dimension set; obtain a selected feature dimension combination whose feature correlation is greater than the correlation threshold, the selected feature dimension combination includes a selected feature dimension whose feature correlation is greater than the correlation threshold, determine the selected feature dimension to be removed from the selected feature dimension combination, remove the selected feature dimension to be removed from the selected feature dimension set, and use at least one of the remaining selected feature dimensions in the selected feature dimension set as the target feature dimension.

[0186] In some embodiments, the target feature dimension obtaining module 910 is also used to obtain the feature coverage of each selected feature dimension in the selected feature dimension combination in the target user group; and the selected feature dimension with smaller feature coverage in the selected feature dimension combination is used as the selected feature dimension to be removed.

[0187] In some embodiments, the feature difference acquisition module 908 includes: a feature distance calculation unit configured to calculate the feature distance between user group features and use the calculated feature distance as the feature difference between the user group features; and a feature discrimination degree acquisition unit configured to use the feature distance as the feature discrimination degree corresponding to the candidate feature dimension.

[0188] In some embodiments, the device also includes: a target sub-user group obtaining module, which is used to classify the target user group based on the target feature dimension to obtain the target sub-user group; a user label obtaining module, which is used to count the user characteristics of users in the target sub-user group, and obtain the user label corresponding to the target sub-user group based on the statistically obtained user statistical characteristics.

[0189] In some embodiments, the user tag acquisition module includes: a user statistics acquisition unit for collecting statistics of users in the target sub-user group that have an interactive association relationship with the second interactive content to obtain user statistics corresponding to the second interactive content; a statistical difference value acquisition unit for obtaining a second reference statistical value corresponding to the second interactive content and obtaining a statistical difference value between the user statistics and the second reference statistical value; and a user tag acquisition unit for obtaining a user tag corresponding to the target sub-user group based on content attributes corresponding to the second interactive content when the statistical difference value is greater than a difference threshold, so as to push content based on the user tag.

[0190] In some embodiments, the user tag acquisition module also includes: an attribute statistical feature acquisition unit, which is used to count the attribute features of users in the target sub-user group, obtain the attribute statistical features corresponding to the target sub-user group, and use the attribute statistical features as the user tags corresponding to the target sub-user group; an attribute statistical feature sending unit, which is used to send the attribute statistical features to the device that sends the attribute feature acquisition request when receiving an attribute feature acquisition request for users in the target sub-user group.

[0191] The specific definition of the user data processing device can be found in the definition of the user data processing method above and will not be repeated here. Each module in the user data processing device described above may be implemented in whole or in part through software, hardware, or a combination thereof. Each of the modules described above may be embedded in or independent of a processor in a computer device in hardware form, or may be stored in a memory in the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0192] In some embodiments, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 10As shown. The computer device includes a processor, a memory, and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store data involved in the user data processing method. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a user data processing method is implemented.

[0193] In some embodiments, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 11 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a user data processing method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse.

[0194] Those skilled in the art will understand that Figure 10 and Figure 11 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0195] In some embodiments, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0196] In some embodiments, a computer-readable storage medium is provided, storing a computer program, which implements the steps in the above-mentioned method embodiments when executed by a processor.

[0197] In some embodiments, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of each of the above method embodiments.

[0198] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0199] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0200] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A user data processing method, characterized in that: The method comprises: Obtaining a target user group to be classified; the target user group includes multiple target users; Dividing the target user group based on the candidate feature dimensions in the candidate feature dimension set, respectively, to obtain a set of divided user groups corresponding to the candidate feature dimensions; Obtaining user group characteristics corresponding to each divided user group in the divided user group set; Obtaining feature differences between the user group features, and obtaining feature discrimination corresponding to the candidate feature dimension based on the feature differences, wherein the feature discrimination is positively correlated with the feature differences; Selecting a target feature dimension from the candidate feature dimension set based on the feature discrimination; Classifying the target user group based on the target feature dimension to obtain a target sub-user group; Collecting statistics on users in the target sub-user group who have an interactive association relationship with the second interactive content to obtain user statistics corresponding to the second interactive content; Obtaining a second reference statistical value corresponding to the second interactive content, calculating a ratio of the user statistical value to the second reference statistical value, and using the calculated ratio as a statistical difference value; When the statistical difference value is greater than a difference threshold, a user tag corresponding to the target sub-user group is obtained based on a content attribute corresponding to the second interactive content, so as to push content based on the user tag.

2. The method according to claim 1, characterized in that The acquiring of user group features corresponding to each divided user group in the divided user group set includes: Performing statistics on a plurality of user features corresponding to the divided user groups to obtain a target statistical value corresponding to each of the user features; The target characteristic values ​​corresponding to the target statistical values ​​are obtained, and the target characteristic values ​​are arranged to obtain the user group characteristics.

3. The method according to claim 2, characterized in that The user characteristics include interactive behavior characteristics, and the statistics of the multiple user characteristics corresponding to the divided user groups are performed to obtain the target statistical value corresponding to each of the user characteristics. Statistics are collected on users in the divided user group who have an interactive association relationship with the first interactive content to obtain a target statistical value corresponding to the interactive behavior feature.

4. The method according to claim 2, characterized in that The obtaining of target characteristic values ​​corresponding to the target statistical values ​​includes: Obtain a first reference statistical value, where the first reference statistical value is a statistical value obtained by performing statistics on the user characteristics corresponding to the reference user group; A corresponding target characteristic value is obtained based on a numerical relationship between the target statistical value and the first reference statistical value.

5. The method according to claim 4, characterized in that The numerical relationship includes a ratio relationship, and obtaining the corresponding target characteristic value based on the numerical relationship between the target statistical value and the first reference statistical value includes: Performing a ratio operation on the target statistical value and the first reference statistical value to obtain a target ratio; A target ratio range in which the target ratio lies is obtained, and a range characteristic value corresponding to the target ratio range is used as the target characteristic value.

6. The method according to claim 1, characterized in that The step of selecting a target feature dimension from the candidate feature dimension set based on the feature discrimination comprises: Selecting a candidate feature dimension that meets a discrimination condition from the candidate feature dimension set based on the feature discrimination, and using the candidate feature dimension that meets the discrimination condition as a target feature dimension; The discrimination condition includes at least one of: the feature discrimination is greater than a discrimination threshold; or the discrimination ranking is before a ranking threshold.

7. The method according to claim 1, characterized in that The step of selecting a target feature dimension from the candidate feature dimension set based on the feature discrimination comprises: Selecting candidate feature dimensions that meet a discrimination condition from the candidate feature dimension set based on the feature discrimination, and using each candidate feature dimension that meets the discrimination condition as a selected feature dimension to form a selected feature dimension set; the discrimination condition includes at least one of the feature discrimination being greater than a discrimination threshold or the discrimination ranking being before a ranking threshold; Obtaining feature correlations between selected feature dimensions in the selected feature dimension set; Obtain a selected feature dimension combination whose feature correlation is greater than a correlation threshold, wherein the selected feature dimension combination includes selected feature dimensions whose feature correlation is greater than the correlation threshold, determine the selected feature dimensions to be removed from the selected feature dimension combination, remove the selected feature dimensions to be removed from the selected feature dimension set, and use at least one of the remaining selected feature dimensions in the selected feature dimension set as a target feature dimension.

8. The method according to claim 7, characterized in that The step of determining the selected feature dimensions to be removed from the selected feature dimension combination includes: Obtaining feature coverage of each selected feature dimension in the selected feature dimension combination in the target user group; The selected feature dimensions with smaller feature coverage in the selected feature dimension combination are used as the selected feature dimensions to be removed.

9. The method according to claim 1, characterized in that The acquiring of feature differences between the user group features and obtaining feature discrimination corresponding to the candidate feature dimensions based on the feature differences includes: Calculating the characteristic distance between the user group characteristics, and using the calculated characteristic distance as the characteristic difference between the user group characteristics; The feature distance is used as the feature discrimination corresponding to the candidate feature dimension.

10. The method according to claim 1, characterized in that The method further comprises: Performing statistics on the attribute characteristics of users in the target sub-user group to obtain attribute statistical characteristics corresponding to the target sub-user group, and using the attribute statistical characteristics as user labels corresponding to the target sub-user group; When an attribute feature acquisition request for users in the target sub-user group is received, the attribute statistical feature is sent to a device that sends the attribute feature acquisition request.

11. A user data processing device, characterized in that: The device comprises: A target user group acquisition module is used to acquire a target user group to be classified; the target user group includes multiple target users; A module for obtaining a divided user group set is configured to divide the target user group based on the candidate feature dimensions in the candidate feature dimension set, and obtain a divided user group set corresponding to the candidate feature dimensions; A user group feature acquisition module, configured to acquire user group features corresponding to each divided user group in the divided user group set; a feature difference acquisition module, configured to acquire feature differences between the features of the user groups, and obtain feature discrimination corresponding to the candidate feature dimensions based on the feature differences, wherein the feature discrimination is positively correlated with the feature differences; A target feature dimension obtaining module, configured to select a target feature dimension from the candidate feature dimension set based on the feature discrimination; a target sub-user group obtaining module, configured to classify the target user group based on the target feature dimension to obtain a target sub-user group; a user statistical value obtaining unit, configured to collect statistics on users in the target sub-user group who have an interactive association relationship with the second interactive content, and obtain user statistical values ​​corresponding to the second interactive content; a statistical difference value obtaining unit, configured to obtain a second reference statistical value corresponding to the second interactive content, calculate a ratio of the user statistical value to the second reference statistical value, and use the calculated ratio as a statistical difference value; The user tag obtaining unit is configured to obtain the user tag corresponding to the target sub-user group based on the content attribute corresponding to the second interactive content when the statistical difference value is greater than a difference threshold, so as to push content based on the user tag.

12. The device according to claim 11, characterized in that The user group feature acquisition module includes: a target statistical value obtaining unit, configured to perform statistics on a plurality of user features corresponding to the divided user groups, and obtain a target statistical value corresponding to each of the user features; The user group feature obtaining unit is configured to obtain target feature values ​​corresponding to the target statistical values, and arrange the target feature values ​​to obtain the user group features.

13. The device according to claim 12, characterized in that The user characteristics include interactive behavior characteristics. The target statistical value obtaining unit is further configured to collect statistics on users in the divided user groups who have an interactive association relationship with the first interactive content to obtain a target statistical value corresponding to the interactive behavior characteristics.

14. The device according to claim 12, characterized in that The user group feature obtaining unit is further used to obtain a first reference statistical value, where the first reference statistical value is a statistical value obtained by performing statistics on the user features corresponding to the reference user group; and obtain a corresponding target feature value based on a numerical relationship between the target statistical value and the first reference statistical value.

15. The device according to claim 14, characterized in that The numerical relationship includes a ratio relationship. The user group feature obtaining unit is also used to perform a ratio operation on the target statistical value and the first reference statistical value to obtain a target ratio; obtain the target ratio range where the target ratio is located, and use the range characteristic value corresponding to the target ratio range as the target characteristic value.

16. The device according to claim 11, characterized in that The target feature dimension obtaining module is further configured to select a candidate feature dimension that meets a discrimination condition from the candidate feature dimension set based on the feature discrimination, and use the candidate feature dimension that meets the discrimination condition as the target feature dimension; The discrimination condition includes at least one of: the feature discrimination is greater than a discrimination threshold; or the discrimination ranking is before a ranking threshold.

17. The device according to claim 11, characterized in that The target feature dimension obtaining module is further configured to select, based on the feature discrimination, candidate feature dimensions that meet a discrimination condition from the candidate feature dimension set, and use each candidate feature dimension that meets the discrimination condition as a selected feature dimension to form a selected feature dimension set; the discrimination condition includes at least one of a feature discrimination greater than a discrimination threshold or a discrimination ranking before a ranking threshold; and obtain feature correlations between the selected feature dimensions in the selected feature dimension set; Obtain a selected feature dimension combination whose feature correlation is greater than a correlation threshold, wherein the selected feature dimension combination includes selected feature dimensions whose feature correlation is greater than the correlation threshold, determine the selected feature dimensions to be removed from the selected feature dimension combination, remove the selected feature dimensions to be removed from the selected feature dimension set, and use at least one of the remaining selected feature dimensions in the selected feature dimension set as a target feature dimension.

18. The device according to claim 17, characterized in that The target feature dimension obtaining module is further used to obtain the feature coverage of each selected feature dimension in the selected feature dimension combination in the target user group; and to use the selected feature dimension with smaller feature coverage in the selected feature dimension combination as the selected feature dimension to be removed.

19. The device according to claim 11, characterized in that The feature difference acquisition module includes: a feature distance calculation unit, configured to calculate the feature distance between the features of the user groups, and use the calculated feature distance as the feature difference between the features of the user groups; The feature discrimination obtaining unit is configured to use the feature distance as the feature discrimination corresponding to the candidate feature dimension.

20. The device according to claim 11, characterized in that The device further comprises: an attribute statistical feature obtaining unit, configured to collect statistics on the attribute features of users in the target sub-user group, obtain the attribute statistical features corresponding to the target sub-user group, and use the attribute statistical features as user labels corresponding to the target sub-user group; The attribute statistical feature sending unit is configured to send the attribute statistical feature to a device that sends the attribute statistical feature obtaining request when receiving an attribute characteristic obtaining request for a user in the target sub-user group.

21. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 10 is implemented.

22. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

23. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the method according to any one of claims 1 to 10 is implemented.

Images