Data processing method, apparatus, device, computer-readable medium, and program product

Abstract

This application discloses a data processing method, apparatus, device, computer-readable medium, and program product. The method includes: acquiring user behavior data of a first user and user behavior data of a second user from a user database; determining the target similarity between the first user and the second user based on the user behavior data of the first user and the second user; determining the interaction intensity index between the first user and the second user based on historical interaction data between the first user and the second user; determining the target matching degree between the first user and the second user based on the target similarity and the interaction intensity index; and classifying users in the user database according to the target matching degree. This solution comprehensively characterizes the degree of fit between two users in user behavior data and historical interaction data through the target matching degree, which can provide an accurate basis for subsequent user classification.

Description

Technical Field

[0001] This application belongs to the field of computer technology, and specifically relates to a data processing method, apparatus, device, computer-readable medium, and program product. Background Technology

[0002] With the increasing popularity of online social networking, community operations, and precise user operations, the demand for refined matching and large-scale classification of massive users continues to surge. However, traditional user classification methods usually rely solely on pre-selected tag data, with a simplistic filtering logic and insufficient accuracy in classification and matching. Summary of the Invention

[0003] The purpose of this application is to provide a data processing method, apparatus, device, computer-readable medium, and program product to improve the accuracy of user classification.

[0004] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part by practice of this application.

[0005] According to one aspect of the embodiments of this application, a data processing method is provided, including:

[0006] The user behavior data of a first user and a second user are obtained from the user database, and the target similarity between the first user and the second user is determined based on the user behavior data of the first user and the second user; the target similarity represents the degree of matching between the user behavior data of the first user and the user behavior data of the second user; the first user and the second user are two different users in the user database. Based on the historical interaction data between the first user and the second user, an interaction intensity index is determined between the first user and the second user; the interaction intensity index is used to indicate the frequency of interaction between the first user and the second user. The degree of target matching between the first user and the second user is determined based on the target similarity and the interaction intensity index. Users in the user database are categorized based on the degree of target matching.

[0007] According to one aspect of the embodiments of this application, a data processing apparatus is provided, comprising: The similarity determination module is used to acquire user behavior data of a first user and user behavior data of a second user from the user database, and determine the target similarity between the first user and the second user based on the user behavior data of the first user and the second user; the target similarity represents the degree of matching between the user behavior data of the first user and the user behavior data of the second user; the first user and the second user are two different users in the user database; The intensity index determination module is used to determine the interaction intensity index between the first user and the second user based on the historical interaction data between the first user and the second user; the interaction intensity index is used to indicate the frequency of interaction between the first user and the second user. A matching degree determination module is used to determine the target matching degree between the first user and the second user based on the target similarity and the interaction intensity index; The user category determination module is used to classify users in the user database according to the target matching degree.

[0008] According to one aspect of the embodiments of this application, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the data processing method provided in any embodiment of this application.

[0009] According to one aspect of the embodiments of this application, an electronic device is provided, the electronic device comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor executes the executable instructions to cause the electronic device to perform a data processing method according to any embodiment of the present application.

[0010] According to one aspect of the embodiments of this application, a computer program product or computer program is provided, which includes 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 data processing method of any embodiment of this application.

[0011] In the technical solution provided in this application embodiment, firstly, user behavior data of a first user and user behavior data of a second user are obtained from the user database. Based on these user behavior data, a target similarity is determined between the first user and the second user. The target similarity characterizes the degree of matching between the user behavior data of the first user and the second user. The first user and the second user are two different users in the user database. Subsequently, based on historical interaction data between the first user and the second user, an interaction intensity index is determined. This interaction intensity index indicates the frequency of interaction between the first user and the second user. Next, the target matching degree between the first user and the second user is determined based on the target similarity and the interaction intensity index. Finally, the users in the user database are classified according to the target matching degree to obtain multiple user categories. Thus, by fusing target similarity and interaction intensity index to obtain the target matching degree, the degree of fit between two users in user behavior data and historical interaction data can be comprehensively characterized, thereby providing an accurate basis for subsequent user classification.

[0012] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0013] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0014] Figure 1 A schematic diagram of a data processing system architecture provided in one embodiment of this application is shown.

[0015] Figure 2 A flowchart illustrating the data processing method provided in an embodiment of this application is shown schematically.

[0016] Figure 3A A topology diagram of one embodiment of this application is shown schematically.

[0017] Figure 3B A topology diagram of one embodiment of this application is shown schematically.

[0018] Figure 3C A topology diagram of one embodiment of this application is shown schematically.

[0019] Figure 4 A flowchart illustrating a data processing method provided in one embodiment of this application is shown schematically.

[0020] Figure 5 A schematic block diagram of a data processing apparatus provided in one embodiment of this application is shown.

[0021] Figure 6 A schematic diagram of the computer system architecture used to implement the technical solution of this application is shown. Detailed Implementation

[0022] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0023] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0024] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0025] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0026] It is understood that in the specific implementation of this application, data related to user information (such as age, gender, occupation, chat history) is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0027] Figure 1 A schematic diagram of a data processing system architecture provided in one embodiment of this application is shown.

[0028] like Figure 1 As shown, the data processing system in this embodiment adopts a layered, decoupled, and modular design, comprising four core parts: a user database, a user classification layer, a group construction layer, and an information recommendation layer. Each layer interacts with the others through standardized data interfaces, forming a stable, scalable, and dynamically iterative complete data processing closed loop.

[0029] The user database, as the core data storage unit of the system, is used to persistently store user profile data, user behavior data, and historical interaction data of all registered users on the platform. This database adopts a distributed storage architecture and is sharded and indexed according to user ID to support high-concurrency read / write operations and fast range queries. User profile data includes basic information such as user age, region, and declared interest tags, and may also include dynamic attributes such as user level, registration duration, and number of historical group participations. User behavior data comprehensively reflects users' speaking rhythm, expression habits, and emotional tendencies. Specifically, it can include timestamp sequences of messages sent by users in public channels, groups, or private chats on the platform, as well as text messages, voice messages, emoticons, and punctuation usage. Historical interaction data includes the number of mutual replies between users, the number of proactive interactions, the number of passive responses, the number of mentions, the average response latency, the distribution of interaction intervals, and the duration of a single interaction, which can be used to quantify the closeness of relationships and the stability of interactions between users.

[0030] The user classification layer can classify users based on user data sent from the user database, thereby aggregating originally scattered individual users into category groups with high internal cohesion and low external coupling.

[0031] The group building layer receives user categories output by the user classification layer, filters them, and constructs groups containing multiple users, such as creating an online chat room, voice room, or interest group. It also initializes the group's runtime data structure, including a list of group members and a creation timestamp. After group creation, the group building layer writes metadata such as the group ID, member ID list, and creation time to persistent storage and simultaneously notifies the recommendation layer and the push layer to begin real-time monitoring of the group.

[0032] The information recommendation layer is divided into two sub-layers: the Adjustment Information Push Sub-layer and the Unified Push Sub-layer. The Adjustment Information Push Sub-layer monitors the group operation status output by the group building layer and pushes information based on group interaction data (e.g., inviting new users, pushing buffered topics) to adjust the matching degree between users in the group. The Unified Push Sub-layer is responsible for static content recommendation based on user categories, that is, uniformly pushing the same information (e.g., technology news, emotional articles, entertainment information, learning materials, etc.) to users in a certain user category. Specifically, the Unified Push Sub-layer obtains user category tags output by the user classification layer offline, determines the information type most interesting to each user category through offline analysis (e.g., calling collaborative filtering or statistical methods based on category frequency), and then pushes the corresponding standard content of the same type to all users belonging to the same user category in batches on a daily or weekly basis (e.g., through application notifications, in-app messages, or information feed recommendations). This push does not rely on the real-time status of the group, but is based on long-term user profiles and classification results. Therefore, it can work in parallel with the Adjustment Information Push Sub-layer, complementing each other and jointly improving user activity and retention rates on the social platform.

[0033] The data processing system in this embodiment achieves fully automated processing across the entire chain, including unified user data management, precise user clustering, intelligent group construction, and hierarchical information push, through the collaborative operation of a user database, user classification layer, group construction layer, and information recommendation layer. Each layer interacts via standardized interfaces, ensuring both independent scalability of modules, support for distributed high-concurrency deployment and rapid iteration, and the ability to achieve precise classification, group matching, and information push based on user behavior and interaction characteristics.

[0034] The data processing method provided in this application will be described in detail below with reference to specific implementation methods.

[0035] Figure 2 A flowchart illustrating a data processing method according to an embodiment of this application is shown schematically. The implementation process of the technical solution of this application will be described below using a terminal device as the execution subject. Figure 2 As shown, the smart contract execution method provided in this embodiment includes steps 210 to 240, as detailed below: Step 210: Obtain user behavior data of the first user and user behavior data of the second user from the user database, and determine the target similarity between the first user and the second user based on the user behavior data of the first user and user behavior data of the second user.

[0036] Specifically, user behavior data includes data that can characterize a user's social behavior style, such as the sequence of timestamps of a user's messages in different time windows (reflecting the time pattern of a user's social interactions), the content of text messages sent, the duration, volume, and speech rate of voice messages, and the type and frequency of emoticon usage. This data comprehensively reflects how users interact with others, when they interact, and with what emotions they interact; it is a digital carrier of an individual's social habits and expression style.

[0037] Target similarity measures the degree of matching between the first and second users in terms of social behavioral styles. Higher target similarity means that the two users' behavioral patterns (such as speaking time, speaking habits, punctuation usage, and emotional intensity in language expression) are more coordinated, making smooth communication more likely when they interact together. Here, the first and second users are two different users from the user database.

[0038] Step 220: Determine the interaction intensity index between the first user and the second user based on the historical interaction data between the first user and the second user.

[0039] Specifically, historical interaction data records all direct interaction information that has occurred between pairs of users in the user database. This may include the number of replies between pairs of users, the number of times they actively initiate interactions (such as sending private messages or mentioning others), the number of times they passively respond, the average response latency (the time interval between receiving a message and sending a reply), the interaction interval distribution (the time distribution between consecutive interactions), the duration of a single interaction for voice or video calls, and the total length of the interaction text. This data can comprehensively reflect the frequency and depth of interaction between two users.

[0040] The interaction intensity index indicates the frequency of interaction between the first user and the second user. The higher the interaction intensity index, the more active their historical interactions and the closer their connection, thus truly reflecting the level of trust and interaction patterns that have been formed between the users.

[0041] Step 230: Determine the degree of target matching between the first user and the second user based on the target similarity and interaction intensity index.

[0042] Specifically, the target matching degree integrates the intrinsic fit of social behavior styles between users with the external closeness of historical interactions. It can comprehensively and objectively represent the overall fit level between the first user and the second user, taking into account the consistency of social behavior styles and the effectiveness of historical real interactions, thereby providing an accurate basis for subsequent classification.

[0043] In one embodiment of this application, the method for determining the target matching degree between a first user and a second user based on target similarity and interaction intensity index can be as follows: multiply the target similarity and interaction intensity index, and use the product as the target matching degree between the first user and the second user; or assign different weight coefficients to the target similarity and interaction intensity index respectively, and then use the weighted sum of the two as the target matching degree; or calculate the geometric mean / harmonic mean of the target similarity and interaction intensity index as the target matching degree between the first user and the second user; or directly take the smaller value between the target similarity and interaction intensity index as the target matching degree. This method best reflects the weakest link effect; only when both indicators reach a high level will the matching degree be high, making it suitable for scenarios with extremely high security requirements where both must be considered.

[0044] Step 240: Classify users in the user database according to the degree of target matching.

[0045] Specifically, users with a target matching degree greater than or equal to a first threshold can be grouped into the same user category, while users with a target matching degree less than the first threshold can be grouped into different user categories. In this way, users within the same user category generally have a high degree of target matching and exhibit high consistency in social behavior styles and interaction habits, while users in different categories generally have a low degree of target matching and show significant differences in social behavior styles and interaction habits. It is understandable that the aforementioned user categories can directly serve scenarios such as group building, personalized recommendations, and social matching.

[0046] In the technical solution provided in this application embodiment, user behavior data of a first user and user behavior data of a second user are first obtained from the user database. Based on these two user behavior data, a target similarity is determined between the first user and the second user. The target similarity characterizes the degree of matching between the user behavior data of the first user and the second user. The first user and the second user are two different users in the user database. Subsequently, based on historical interaction data between the first user and the second user, an interaction intensity index is determined. The interaction intensity index indicates the frequency of interaction between the first user and the second user. Next, the target matching degree between the first user and the second user is determined based on the target similarity and the interaction intensity index. Finally, the users in the user database are classified according to the target matching degree, resulting in multiple user categories. Thus, by fusing target similarity and interaction intensity index to obtain the target matching degree, the degree of fit between two users on user behavior data and historical interaction data can be comprehensively characterized, thereby providing an accurate basis for subsequent user classification.

[0047] In one embodiment of this application, the process of determining the target similarity between a first user and a second user includes: extracting features from the user behavior data of the first user to obtain a first time series and a first behavior feature vector; the first time series includes a timestamp sequence of messages sent by the first user; extracting features from the user behavior data of the second user to obtain a second time series and a second behavior feature vector; the second time series includes a timestamp sequence of messages sent by the second user; calculating a first matching degree between the first time series and the second time series, and a second matching degree between the first behavior feature vector and the second behavior feature vector; and determining the target similarity between the first user and the second user based on the first matching degree and the second matching degree.

[0048] Specifically, the first time series is a sequence of data extracted from the user behavior data of the first user, composed of the timestamps of messages sent by the user arranged in chronological order. Each timestamp is accurate to the millisecond, and the length of the sequence depends on the number of times the first user speaks within that time window. This sequence fully preserves the temporal rhythm information of the first user's social behavior, including the density of messages, periodicity, and distribution of active periods. Similarly, the second time series is a sequence of data extracted from the user behavior data of the second user, composed of the timestamps of messages sent by the user arranged in chronological order. The first and second time series can be used to assess whether the two users are "in sync" in terms of time.

[0049] For example, the feature extraction process for the first time series may include: firstly, acquiring all the user's posts within a certain time window (e.g., the past six months, 30 days, or one week), with each post accompanied by a timestamp accurate to milliseconds, and sorting them chronologically to form an original sequence; then, preprocessing the sequence to remove duplicate timestamps, and merging multiple messages within the same millisecond. Next, calculating multiple dimensions of rhythmic features based on the preprocessed original sequence: firstly, statistical features of post intervals, including the mean, median, and standard deviation of the time interval between two adjacent posts, reflecting the density and regularity of the user's posts; secondly, distribution features of active time periods, dividing a 24-hour day into several time slices (e.g., one block every 2 hours), and statistically analyzing the post frequency within each time slice to form a 24-dimensional activity histogram. Finally, concatenating the extracted features into a multi-dimensional numerical vector and normalizing it to obtain the first time series. It is understood that the feature extraction process for the second time series is similar to that for the first time series, and will not be elaborated further here.

[0050] The first behavioral feature vector characterizes the social behavior style of the first user. Dimensions of this vector may include average text message length, frequency of punctuation usage (especially exclamation marks and question marks), duration and average speech rate of voice messages, and frequency of emoji usage. Similarly, the second behavioral feature vector characterizes the social behavior style of the second user. It is understood that the values ​​of each dimension are standardized so that the behavioral feature vectors of different users can be used to calculate distance or similarity in the same geometric space.

[0051] For example, the feature extraction process for the first row of feature vectors may include: First, preprocessing the user behavior data of the first user, including filtering irrelevant messages automatically sent by the system (such as welcome messages for joining a group), removing invalid text that is too short (e.g., less than 2 characters), and segmenting continuous voice data into independent events based on a single speech. Then, the preprocessed user behavior data is divided into four subcategories according to feature type: speech expression features, interaction rhythm features, emotional tendency features, and social preference features. For each subcategory, corresponding quantitative indicators are extracted, such as text length distribution, high-frequency word attributes, proportion of tone marks (such as exclamation marks and question marks), message sending interval, and average response time. Next, the above four types of sub-features are numerically encoded, that is, discrete features are converted into numerical vectors using one-hot encoding or embedding encoding, and all sub-features are uniformly projected into a feature space of the same dimension through a fixed-dimensional mapping. Finally, through various methods such as weighted concatenation, linear mapping, or feature fusion networks, the sub-feature vectors are integrated into a single continuous vector, which serves as the first row of feature vectors. It is understandable that the feature extraction process for the second row of feature vectors is similar to that for the first row of feature vectors, and will not be elaborated here.

[0052] The first degree of matching is used to indicate the synchronicity or similarity between a first user's first time series and a second user's second time series. It can be calculated using dynamic time warping to measure the shape similarity between two time series of unequal lengths, or using cross-correlation analysis to detect the lag correlation between the two time series on the time axis. Alternatively, it can be simply calculated by calculating the overlap ratio of the two users' active time periods. A higher first degree of matching indicates that the first and second users are more similar in their speaking times; for example, both may frequently speak between 8 PM and 10 PM, meaning that the first and second users are more likely to engage in real-time interaction during this period.

[0053] The second matching degree indicates the spatial proximity between the first user's first behavioral feature vector and the second user's second behavioral feature vector. Optionally, the second matching degree can be calculated using cosine similarity (measuring the cosine of the angle between vectors, focusing on direction rather than magnitude), Euclidean distance (measuring absolute distance in geometric space), or Pearson correlation coefficient (measuring the linear correlation of trends in each dimension). A higher second matching degree indicates a greater similarity in the expression styles of the first and second users, making smoother communication and emotional resonance more likely when they interact together.

[0054] In one embodiment of this application, the calculation method of the interaction intensity index between the first user and the second user includes: determining the first number of active interactions and the first number of passive responses corresponding to the first user based on historical interaction data between the first user and the second user; determining the second number of active interactions and the second number of passive responses corresponding to the second user; performing a weighted summation of the first number of active interactions and the first number of passive responses based on a first weight to obtain a first calculation result, and performing a weighted summation of the second number of active interactions and the second number of passive responses based on a second weight to obtain a second calculation result; and determining the interaction intensity index between the first user and the second user based on the first calculation result, the second calculation result, and a time decay factor.

[0055] Specifically, the first proactive interaction count refers to the cumulative number of times the first user initiates an interaction with the second user, including actions explicitly targeting the second user such as sending private messages, directly mentioning the second user in a group chat, and replying to the second user's messages. Conversely, the second proactive interaction count refers to the cumulative number of times the second user initiates the same type of interaction with the first user. These two metrics record the initiative of users in initiating interactions with each other from their respective perspectives.

[0056] The first passive response count refers to the cumulative number of times the first user replies or responds after receiving a message from the second user; that is, the number of interactions where the first user acts as the responder. Similarly, the second passive response count refers to the number of times the second user responds to messages initiated by the first user. The passive response count reflects the user's level of attention to others' messages and their responsiveness, and together with the active interaction count, constitutes a complete two-way record of interactions between the first and second users.

[0057] The first weight is a pre-set weighting coefficient for the number of proactive interactions and passive responses of the first user, used to adjust the relative importance of proactive and passive behaviors in the interaction intensity index calculation. Optionally, a higher weight can be assigned to the number of proactive interactions, as it better reflects the first user's proactive social intentions; while a lower weight is assigned to the number of passive responses, because passive responses depend to some extent on the other party initiating the interaction. It should be noted that the first weight and the second weight can be the same, or different values ​​can be set according to differences in user roles or scenarios; this is not limited here. Similarly, the second weight is a pre-set weighting coefficient for the number of proactive interactions and passive responses of the second user.

[0058] The first calculation result is the sum of the number of active interactions and passive responses of the first user multiplied by their respective first weights. The second calculation result is the sum of the number of active interactions and passive responses of the second user multiplied by their respective second weights.

[0059] The preset time decay factor measures how quickly the influence weight of historical interaction data on the interaction intensity index decays over time. Optionally, the preset time decay factor can be a mathematical function, such as an exponential decay form e. λΔt (Where λ is the preset decay rate parameter, and Δt is the time difference from the current time). For example, an interaction from a year ago contributes far less to the current relationship between two people than an interaction that occurred yesterday. Therefore, by introducing a time decay factor, the system can gradually reduce the weight of earlier historical interaction data, making the interaction intensity index more sensitive to reflect the recent activity level between users, and avoiding being dominated by long-accumulated but cooled historical data.

[0060] For example, suppose user A and user B have interacted multiple times over a period of time. The system needs to calculate the current interaction intensity index based on their historical interaction data to determine the current interaction intensity index between the two. First, the statistical time range is set to the past 30 days, and the following data is obtained: User A actively sent messages to user B 12 times (including private messages, mentioning B in groups, replying to B's messages, etc.); User A passively responded to B 8 times. User B actively sent messages to user A 15 times. User B passively responded to A 10 times. Subsequently, the system sets the weight of the number of active interactions to 0.7, because it better reflects the user's social willingness; the weight of the number of passive responses is 0.3, and user A and user B use the same weight setting (i.e., the first weight and the second weight are the same). Next, the first calculation result (from user A's perspective) is calculated as follows: (User A's number of active interactions × 0.7) + (User A's number of passive interactions × 0.3) = (12 × 0.7) + (8 × 0.3) = 8.4 + 2.4 = 10.8; the second calculation result (from user B's perspective) is calculated as follows: (User B's number of active interactions × 0.7) + (User B's number of passive interactions × 0.3) = (15 × 0.7) + (10 × 0.3) = 10.5 + 3.0 = 13.5. Furthermore, considering that recent interactions better reflect the current relationship's intensity than older interactions, the system introduces a time decay factor e. 0.1Δt Where Δt is the number of days since the most recent interaction. If the query shows that the most recent interaction between user A and user B occurred 5 days ago, then the time decay factor is approximately 0.6065. Therefore, the first calculated result after decay is 10.8 × 0.6065 ≈ 6.55, and the second calculated result after decay is 13.5 × 0.6065 ≈ 8.19. Finally, the first and second calculated results after decay are multiplied together, and the square root is taken to obtain the final interaction strength index. The value is approximately 7.32, indicating that User A and User B still have relatively frequent interactions recently.

[0061] In one embodiment of this application, the decay rate parameter in the aforementioned time decay factor can be set based on historical interaction data between the first user and the second user. Specifically, firstly, historical interaction data between the first user and the second user is acquired, the timestamps of all historical interaction events are counted, the average interval between two adjacent interactions is calculated, and the time difference between the most recent interaction and the current time is calculated. If the average interval is short and the most recent interaction occurs within a preset time, it indicates that the two parties have a high-frequency interaction relationship. The system can automatically set a larger decay coefficient for it, so that the contribution of earlier historical interactions to the weighted summation of the interaction intensity index decreases rapidly, thereby allowing the index to sensitively reflect the recent interaction heat and avoiding the fact that the current relationship has cooled down due to a large amount of outdated data. If the average interval is long or the most recent interaction has exceeded the preset time, it indicates that the two parties have a low-frequency or cooled-down relationship. In this case, a smaller decay coefficient can be set for it, so that the index retains the interaction memory within a longer time window, preventing the index from suddenly dropping to zero due to a natural silence period and completely negating the remaining matching basis between the two parties.

[0062] Simultaneously, this adaptive mechanism can utilize interaction statistics from all users in the user database to determine the threshold for distinguishing between high and low frequencies using the quantile method, thereby enabling dynamic and personalized settings for the attenuation coefficient. In this way, the interaction intensity index can accurately quantify the recent interaction activity between any two users, providing a reliable basis for subsequent target matching calculations.

[0063] It should be noted that the execution order of the steps in the above embodiments is merely an exemplary combination of this application and does not constitute the sole limitation on the implementation of the technical solution of this application. In other embodiments, the order of the steps can be adaptively adjusted according to the actual application scenario. For example, the operation of setting the first preset weight, the second preset weight, and the preset time decay factor does not necessarily depend on the user data acquisition step being executed after it; in specific implementations, the configuration steps of the weight parameters and decay factors in the above embodiments can be completely independent of the data acquisition step, and can be initialized or dynamically updated in advance. That is, such flexible adjustment of the step order also falls within the protection scope covered by this application.

[0064] In one embodiment of this application, after calculating the target similarity and the interaction intensity index, the target matching degree between the first user and the second user can be determined based on the two. Specifically, if the target similarity is greater than or equal to a first preset value and the interaction intensity index is greater than or equal to a second preset value, then the product between the target similarity and the interaction intensity index is taken as the target matching degree between the first user and the second user.

[0065] Specifically, the first preset value is a pre-set value used to measure whether the target similarity between the first user and the second user has reached the minimum standard for subsequent matching. This value can be set based on historical data or experience values ​​of the platform. When the target similarity is greater than or equal to the first preset value, it indicates that the first user and the second user have a high degree of matching in social behavior style; conversely, if the target similarity is less than the first preset value, it indicates that the social behavior styles of the two are too different, and it is difficult to produce smooth communication even if they are forcibly combined.

[0066] The second preset value is pre-set and used to measure whether the interaction intensity index between the first user and the second user meets the numerical requirements for a valid match. This value can also be set based on platform experience. When the interaction intensity index is greater than or equal to the second preset value, it indicates that the first user and the second user have had sufficiently frequent and timely interactions recently, and their relationship is active. Conversely, if the interaction intensity index is less than the second preset value, it indicates that although the two users may have interacted in the past, their relationship has cooled down recently and their contact has become sparse. In this way, by filtering through the second preset value, the system can exclude user combinations that were "once close but have now drifted apart".

[0067] Optionally, when the target similarity is greater than or equal to a first preset value and the interaction intensity index is greater than or equal to a second preset value, the product of the target similarity and the interaction intensity index is used as the target matching degree between the first user and the second user. When the target similarity is less than the first preset value or the interaction intensity index is less than the second preset value, the target matching degree will no longer be calculated using the "product" method, but will be directly set to a minimum effective value (usually 0 or a constant much smaller than the normal matching threshold). This means that the user combination (the first user and the second user) is determined to not meet the basic conditions for being in the same user category. In this way, the threshold judgment avoids the situation where one of the target similarity and the interaction intensity index is extremely high and the other is extremely low (for example, the target similarity is extremely high but the interaction intensity index is extremely low, i.e., the two people have similar social behavior styles but never interact), resulting in a high product of the two, and thus classifying them into the same category, thereby improving the accuracy and flexibility of user classification.

[0068] In one embodiment of this application, users in the user database are classified according to the degree of target matching to obtain multiple user categories, including: constructing a first weighted graph with users in the user database as nodes, constructing edges based on the interaction behavior between adjacent users, and using the degree of target matching between adjacent users as edge weights; dividing the first weighted graph to obtain multiple subgraphs; wherein each subgraph corresponds to a user category; the degree of target matching between users within a user category is greater than or equal to a preset matching value, and the degree of target matching between users in different user categories is less than the preset matching value.

[0069] Specifically, the first weighted graph is a topological structure used to represent the matching relationships among all users in a user database. In this graph, all users are abstracted as vertices, and the relationships between users are abstracted as edges connecting the vertices. As long as two users are queried in the user database and the target matching degree (regardless of its value) is calculated, the system will establish an edge between the corresponding two nodes, and the target matching degree between the two users is the edge weight. The higher the edge weight, the more compatible the two users are in terms of social behavior style and interaction activity, and the more likely they are to be classified into the same user category; an edge weight of zero or a very small value indicates that the two do not have a valid matching basis.

[0070] A subgraph is a topological structure obtained by partitioning the first weighted graph. Each subgraph consists of a subset of nodes from the original graph, along with the edges and their weights. There are no connecting edges between different subgraphs, or the weights of the connecting edges are extremely low, below the preset matching value. Each subgraph corresponds to a user category. The nodes within a subgraph have high edge weights, meaning the target matching degree is generally greater than or equal to the preset matching value. This indicates that these users are highly matched in their social behavior styles and actual interactions, making them suitable for being grouped together or receiving similar content recommendations.

[0071] The preset matching value is a pre-defined numerical threshold used to determine whether two users should be grouped into the same user category during user classification. This threshold is typically set based on historical platform data, experimental optimization, or business needs. When the target matching degree between two users is greater than or equal to the preset matching value, it indicates that their matching degree in social behavior style and actual interaction is acceptable, and they can coexist in the same subgraph. Conversely, if the target matching degree is less than the preset matching value, it indicates that the matching degree between the two in social behavior style and actual interaction is insufficient, and they should be separated into different subgraphs.

[0072] For example, spectral clustering can be used for user category segmentation. First, each user in the user database is treated as a node, and a symmetric similarity matrix is ​​constructed based on the calculated target matching degree between any two users. The rows and columns of this matrix correspond to all users, and each element represents the edge weight (target matching degree) between the corresponding two users. Then, a degree matrix is ​​constructed based on this similarity matrix. Each element on the diagonal of this matrix equals the sum of all elements in the corresponding row of the similarity matrix, representing the total connection strength of each node to all other nodes. Next, the Laplacian matrix is ​​calculated using the similarity matrix and the degree matrix. For example, subtracting the similarity matrix from the degree matrix yields a denormalized Laplacian matrix, or a normalized Laplacian matrix is ​​used to improve the robustness of the clustering effect to outliers. Next, the Laplacian matrix is ​​subjected to eigenvalue decomposition. The eigenvalues ​​are sorted from smallest to largest, and the k smallest eigenvalues ​​are selected (k is the preset number of target categories, which can be automatically determined by a heuristic method such as the eigenvalue gap method). The corresponding eigenvectors are then extracted, representing each user as a k-dimensional vector composed of the values ​​of these k eigenvectors at the user's index. Finally, using these reduced k-dimensional vectors as the new data representation, the classic K-means clustering algorithm is run to divide the users into k clusters, each cluster corresponding to a subgraph, i.e., a user category.

[0073] For example, the Leuven community detection algorithm can also be used for user category classification. First, in the initial state, each node in the first weighted graph is assigned a separate category (i.e., each user forms their own category). Then, for each node, the change in weighted modularity when moving it from its current category to the categories of all its neighboring nodes is calculated (the calculation of weighted modularity directly depends on edge weights): edges with high target matching degrees contribute significantly if they connect nodes within the same category; if they connect nodes from different categories, they incur a negative penalty. The system selects the move operation that maximizes the modularity gain. The first iteration ends when no further improvement in modularity can be achieved through movement. Then, each currently formed category is shrunk into a "supernode," and the edge weights between supernodes are defined as the sum of the target matching degrees of all user pairs between the original two categories, thus constructing a smaller weighted graph. The above "move-shrink" process is repeated on this new graph until no further improvement in modularity occurs. Finally, the set of users corresponding to each supernode constitutes a user category.

[0074] In one embodiment of this application, after dividing the first weighted graph into multiple subgraphs, the sum of the weights of all edges in each subgraph can be calculated; the ratio between the sum of weights and the maximum number of edges corresponding to the subgraph is determined as the edge weight density of the subgraph; the edge weight density is used to indicate the average matching degree between every two nodes in the subgraph; the user corresponding to the subgraph with the highest edge weight density is taken as the target user, and a target group is constructed based on the target user.

[0075] Specifically, the sum of all edge weights refers to the total value obtained by adding up the weights (i.e., the target matching degree) of all edges connecting two nodes within a subgraph. This sum reflects the overall cumulative matching degree between all user pairs within the subgraph.

[0076] The maximum number of edges refers to the theoretically maximum number of edges that can exist in a subgraph. For a subgraph containing n nodes, the maximum number of edges is... This means that every pair of nodes is connected by an edge. In this embodiment, the maximum number of edges is used as the normalized denominator to eliminate the absolute numerical influence of the subgraph size (number of nodes) on the sum of edge weights, allowing subgraphs of different sizes to be compared on the same scale.

[0077] Edge weight density measures the average matching degree carried by each possible edge in a subgraph. A higher edge weight density indicates a generally higher degree of target matching among user combinations within the subgraph, meaning stronger overall social compatibility among users. Conversely, a lower edge weight density indicates more user combinations with low matching degrees within the subgraph, or sparser connections between users. The subgraph with the highest edge weight density corresponds to the best overall compatibility of users' social behavior styles and interaction activity.

[0078] Target users refer to all users contained in the subgraph with the highest edge weight density. Target groups are social group entities actually established based on target users, such as an online chat room, voice chat room, interest group, or multiplayer game team. The system will call the group creation interface of the instant messaging or social platform based on the target user list to generate a unique group identifier for these users and initialize the data structures required for group operation, such as member list, creation time, and sentiment baseline.

[0079] In another embodiment of this application, the method of constructing the target group can also be to determine the user group corresponding to each user category; the user group includes two different users in the user category; calculate the sum of the target matching degree of all user groups in the user category, and determine the average matching degree of the user category by the ratio between the sum of the target matching degree and the number of user groups; take the user in the user category with the highest average matching degree as the target user, and construct the target group based on the target user.

[0080] Specifically, a user group refers to a binary ordered or unordered pair consisting of two different users within a user category. For example, if a user category contains three users, A, B, and C, then the possible user groups include (A,B), (A,C), and (B,C), for a total of three groups, each corresponding to a previously calculated target match degree.

[0081] The number of user groups refers to the total number of all possible distinct user groups within a given user category. For a user category containing n users, the number of user groups is... .

[0082] In one embodiment of this application, after constructing a target group, the group can be monitored in real time, that is, the interaction data generated by each target user in the target group within a preset time period is acquired at preset time intervals; a second weighted graph corresponding to the target group is constructed based on the interaction data generated within the preset time period; features are extracted from the second weighted graph to obtain multiple interaction features, and the interaction score of the target group is calculated based on the multiple interaction features; the interaction score characterizes the frequency of interaction between each target user in the target group; adjustment information is pushed to the target group based on the interaction score; the adjustment information is used to improve the interaction score of the target group.

[0083] Specifically, the preset time refers to the fixed time interval set by the system when periodically monitoring the target group, such as every 30 seconds, every minute, or every 5 minutes. The length of this time window can be dynamically adjusted according to the real-time requirements of the group: for high-frequency interactive voice chat rooms, the preset time is usually shorter to quickly capture changes in atmosphere; for low-frequency interactive interest groups, the preset time can be appropriately extended to reduce computational overhead. At the end of each preset time window, the system uniformly collects all interaction data generated by the target group within that window, which serves as the basis for constructing the second weighted graph.

[0084] Interaction features are three core quantitative indicators extracted from the second weighted graph, which can characterize the group's interaction from different dimensions. Optionally, interaction features may specifically include interaction density, interaction centrality, and splitting index. Interaction density reflects the overall activity level of the target group (the frequency with which each target user sends or replies to messages). Higher interaction density indicates more frequent and effective interactions between users in the target group. Therefore, the second weighted graph corresponding to this group can present a graph as follows: Figure 3A The diagram shows a normal topology; the interaction centrality is used to measure whether interactive behavior within a group is excessively concentrated in a few members. A higher interaction centrality indicates that the interaction of the target group is as follows: Figure 3B The "star-shaped concentration" structure shown indicates that most interactions originate from a small number of core members, with peripheral members having limited participation. The splitting index indicates the existence of "cliques" within the target group, and the minimal interaction between different cliques results in the second weighted graph corresponding to this group exhibiting the following characteristics. Figure 3C The separation state is shown.

[0085] For example, the interaction density can be obtained by calculating the sum of the edge weights of all directed edges in the second weighted graph and then obtaining the ratio between the sum of edge weights and the maximum number of edges in the second weighted graph.

[0086] For each node in the second weighted graph, the sum of the weights of all directed edges pointing to that user can be calculated; this is called the weighted in-degree. A higher weighted in-degree indicates a greater overall interaction intensity from other members, suggesting higher influence or attention within the group. Next, the maximum weighted in-degree of all nodes and the sum of all node weighted in-degrees are identified. Then, the difference between the weighted in-degree of each node and the maximum value is calculated, and these differences are summed. Finally, this sum is divided by the maximum sum achievable in the most extreme theoretical case (where only one node's weighted in-degree equals the maximum possible value, and the weighted in-degrees of all other nodes are zero). The result is the interaction central potential.

[0087] The subgraph structure in the second weighted graph can be identified by running the Leuven community discovery algorithm on the second weighted graph. Then, the degree of difference between the average edge weight inside the subgraph and the average edge weight between subgraphs can be calculated. For example, the average weight inside the subgraph is subtracted from the average weight outside the subgraph and then divided by the sum to obtain the split index.

[0088] The second weighted graph is a directed weighted graph structure dynamically constructed by the system during the real-time monitoring phase based on the interaction data generated by the target group within a preset time window. A directed edge is a line connecting two nodes in the second weighted graph. Unlike undirected edges, directed edges have a clear direction, represented by an arrow. Whenever a target user interacts with another target user (e.g., sending a message, replying directly, or responding via voice), a directed edge is established between the nodes corresponding to the two users, with the arrow pointing from the party initiating the interaction to the party receiving and responding. If two users interact with each other within the time window, there will be two directed edges in opposite directions.

[0089] For example, the construction process of the second weighted graph may include: taking each target user in the target group as a node in the second weighted graph; if there is an interaction between the first target user and the second target user, constructing a directed edge between the node where the first target user is located and the node where the second target user is located in the second weighted graph; the direction of the directed edge is from the initiator of the interaction to the receiver of the interaction; determining a first delay time based on the interaction data generated by the first target user within a preset time; determining a second delay time based on the interaction data generated by the second target user within the preset time; determining the weight of the edge from the node where the first target user is located to the node where the second target user is located based on the first delay time, and determining the weight of the edge from the node where the second target user is located to the node where the first target user is located based on the second delay time.

[0090] Specifically, the first target user and the second target user are two different target users within the target group. The first delay time refers to the time elapsed from the moment the first target user receives a message from the second target user until the first target user responds. The length of the first delay time directly reflects the first target user's level of attention to the message sent by the second target user and their responsiveness: the shorter the delay time, the more the first target user values ​​the interaction and the more timely their response; conversely, a longer delay time may indicate lower attention or a lack of willingness to interact.

[0091] The second delay time refers to the length of time elapsed from the moment the message arrives at the second target user's address until the second target user responds. By combining the first and second delay times, the system can comprehensively assess the directionality and asymmetry of the interaction between the two users: for example, if the first delay time is significantly shorter than the second delay time, it indicates that the first user is more responsive to the second user, and there is a one-way dependency tendency in their interaction.

[0092] The first and second preset weights are weight values ​​obtained by looking up the first and second delay times in a preset mapping table. The shorter the delay time, the higher the weight. For example, when the first delay time is less than 1 second, the first preset weight can be set to a higher value close to 1, indicating that this is a very timely response; when the first delay time exceeds 10 minutes, the first preset weight can be set to a lower value close to 0, indicating that this interaction has lost its timeliness.

[0093] In one embodiment of this application, to address the problem that newly constructed target groups lack real interaction data during the initial operation phase and cannot directly construct an effective second weighted graph, a virtual edge initialization mechanism based on historical prior information can be introduced. Specifically, at the initial moment after the target group is created, the historical interaction data of each member in the user database is accessed, including prior statistical characteristics such as the member's average response latency, speaking frequency, and response rate in other groups. Then, for any two members in the group, a prior virtual weight is calculated based on their historical interaction data. This weight represents the best estimate of the potential interaction strength between the two members when no real interaction data exists. Next, using group members as nodes and the prior virtual weights between all member pairs as edge weights, a fully connected second weighted graph is constructed, meaning that there is a virtual directed edge (bidirectional) between all nodes. Finally, this initial second weighted graph is used as an alternative topology for the cold start phase, enabling the system to immediately calculate indicators such as interaction density and interaction central potential without waiting for real interactions to occur. During the subsequent grace period (e.g., 10 or 30 minutes), real-time interaction events are collected. Whenever a real directed interaction occurs between a pair of members (such as message reply, mention, etc.), the transient weight calculated from the real interaction data overwrites or replaces the prior virtual weight in that direction. If no real interaction occurs between a pair of members after the grace period ends, the prior virtual weight in the corresponding direction is gradually decayed to zero by the system, and the edge is eventually removed from the second weighted graph.

[0094] In this way, the cold start mechanism based on prior virtual edges effectively avoids misjudging new groups with high activity potential as "dead groups" due to the lack of initial data. It also prevents inappropriate adjustment information pushes from being triggered due to data sparsity during the grace period, such as prematurely introducing new external users or new topic information, thus ensuring the stability and accuracy of group evaluation in the early stage of its life cycle.

[0095] In one embodiment of this application, assume a target group called "Late Night Book Club" has been created on a social platform, with four members: A, B, C, and D. The system sets a preset time window of 30 minutes. At the end of each window, an interaction score is calculated for the group, including: First, collecting interaction data within the "Late Night Book Club" within the 30-minute window. It is found that during these 30 minutes, Member A sent a text message to Member B, 80 characters long, using an exclamation mark. Member B replied after 2 seconds, with a reply of 30 characters. Member C sent a text message to Member D, 120 characters long, with an emoticon. Member D did not reply immediately but replied with a short message of 5 characters after 3 minutes. Member A sent a voice message to Member C, lasting 8 seconds. Member C replied with a text message of 40 characters after 1 second. Member B sent a 50-character message to Member A without punctuation. Member A replied with an emoji (no text) 5 seconds later. Additionally, Member D sent a mention message to Member A, but Member A did not reply within that time window.

[0096] Subsequently, a second weighted graph was constructed based on the aforementioned interaction data. This graph contains four nodes and four directed edges (A→B, B→A, C→D, A→C), each with its own weight. Since there is no interaction between B and C, or between D and B, there are no directed edges such as B→C or D→B. Next, the interaction density of the second weighted graph was calculated as 0.208 (indicating that most possible interactions within the group did not occur, and the overall activity level is low), the interaction centrality was 0.407 (indicating that interactions are not highly concentrated on any one node, but the in-degree of nodes C and B is relatively high, showing a certain concentration trend), and the splitting index was 0.5 (indicating that the group has a certain degree of splitting trend, with sparse interactions between D and other members, forming an isolated subgroup).

[0097] Finally, the preset weights corresponding to each interaction feature are obtained: interaction density weight is 0.4, interaction centrality potential weight is 0.3, and splitting index weight is 0.3. The interaction score is calculated as: Interaction density weight × Interaction density - Interaction centrality potential weight × Interaction centrality potential - Community splitting index weight × Community splitting index = -0.1913. Since the original score may be negative, the system normalizes it to the 0-100 range: first, the theoretical highest score (e.g., 0.4 when interaction density = 1, interaction centrality potential = 0, splitting index = 0) is set to correspond to 100 points, and the theoretical lowest score (-0.6 when interaction density = 0, interaction centrality potential = 1, splitting index = 1) is set to correspond to 0 points. Then, a linear mapping is performed to obtain the interaction score corresponding to -0.1913, which is approximately 100 × (-0.1913 - (-0.6)) / (0.4 - (-0.6)) = 100 × 0.4087 / 1.0 = 40.87 points.

[0098] In one embodiment of this application, after calculating the interaction score of the target group, corresponding adjustment information can be pushed. That is, if the interaction score is less than the first preset score, at least one new user information is pushed to the target group based on the interaction data generated within a preset time; if the interaction score is greater than or equal to the first preset score and less than the second preset score, new topic information is pushed to the target group based on the interaction data generated within a preset time.

[0099] Specifically, when the interaction score is lower than the first preset score, it indicates sparse interaction, long response delays, or narrow coverage within the group, with significantly insufficient overall activity, indicating a lull or inefficient state. The second preset score defines the group's interaction quality as "below average but not yet good." When the interaction score is greater than or equal to the first preset score but less than the second preset score, it indicates that while the group has a certain level of interaction, there is a tendency towards centralization (a few members dominating the discourse) or slight signs of fragmentation, and the interaction quality needs improvement. The first preset score is lower than the second preset score.

[0100] This embodiment integrates three dimensions—interaction density, interaction centrality, and community splitting index—into a single interaction score. The system can comprehensively and evenly evaluate the overall interaction quality of the target group, thereby adjusting information push notifications. Specifically, pushing new user information aims to increase interaction sources within the group by introducing fresh external nodes, thus improving interaction density in subsequent time windows. Pushing new topic information aims to promote more balanced and broader interaction within the group by changing the discussion prompts, thereby reducing interaction centrality and the splitting index. It is understood that the information pushed in this application is not limited to text information but should also include video, audio, and combinations thereof.

[0101] For example, using the data from the aforementioned embodiment, the interaction score of the Late Night Reading Club is approximately 41 points. If the first preset score is 45 points and the second preset score is 70 points, at least one new user who can quickly elicit a rapid response from other users can be selected and the information of this new user can be pushed to the Late Night Reading Club to encourage interaction among the original members, thereby improving the overall interaction score of the group.

[0102] In one embodiment of this application, the type of adjustment information to be pushed can also be determined based on a single interaction feature, that is, by directly monitoring the values ​​of interaction density, interaction central potential and splitting index, without calculating the interaction score.

[0103] For example, a voice chat room labeled "Midnight Healing" is created as the target group, initially containing 6 members. The system sets a healthy threshold for interaction density of 0.2 (ranging from 0 to 1) and calculates the interaction density within the current window every 30 minutes. After the group has been running for 10 minutes, the system calculates that the interaction density in the last 30 minutes is only 0.05. Further analysis of the second weighted graph reveals that although members occasionally speak, no directed edge has a weight exceeding 0.1 (i.e., all responses are too delayed or too brief), and the graph structure presents 6 isolated nodes (i.e., no two nodes form a directed edge). This indicates that the voice chat room is in a severe state of silence, with a lack of effective interactive responses among members. At this point, new users can be selected from the user database who "have an average response delay of less than 2 seconds after their first speech in previously participated unfamiliar groups, and whose speech elicits responses from at least 3 other members," while also having social behavior styles that match the overall style of the voice chat room.

[0104] For example, the target group is a 50-person digital product exchange group. The system sets a health threshold of 0.4 for the split index (range 0-1), and calculates the split index within the current window every 30 minutes. If, within 30 minutes of a brand launch event, the system detects a sharp increase in the group's split index from 0.2 to 0.85, further analysis of the second weighted graph reveals that the group members are divided into two highly dense subgroups. Members within each subgroup frequently interact with each other, but interactions between the two subgroups are very sparse. Furthermore, the existing cross-group interactions all contain extremely opposing expressions (such as the use of strong negative words, frequent exclamation marks, and lengthy arguments). In this case, the core users in each subgroup who trigger cross-group arguments can be identified. To prevent further escalation of conflict, the system moderately collapses the continuous messages from these core users at the information flow distribution layer, so that other members can only see the first two lines of the message by default and need to click "expand" to read the full content. This reduces the visual impact of extreme statements and cuts off the rapid propagation chain of negative interactions. Then, the system uses a large language model to extract high-frequency words and common concerns from the two users' historical conversations over the past 10 minutes. For example, although the two users have opposing attitudes towards the brand, they are both frequently discussing "heat dissipation performance" and "material craftsmanship." Therefore, the system automatically generates a neutral, buffer topic that can elicit positive responses from both sides and pushes it into the group chat under the name of the group assistant: "Putting the brand aside, how much do you think the superconducting heat dissipation material used in this new model improves the actual gaming experience?" This shifts the focus of the debate within the group, avoiding the risk of disbanding or a large number of members leaving due to heated arguments, and preventing the second weighted graph edge from breaking down in the group.

[0105] This embodiment determines which adjustment information to push based on a detected interaction feature. It has advantages such as low computational overhead, fast response speed, and strong strategy targeting, and is especially suitable for voice chat rooms or live broadcast bullet screen scenarios with extremely high real-time requirements.

[0106] In one embodiment of this application, a deep reinforcement learning model can also be constructed as the execution subject for adjusting information push to the target group. Specifically, an end-to-end adaptive decision-making mechanism is constructed using a proximal policy optimization algorithm. First, multi-dimensional information such as interaction density, interaction centrality potential, splitting exponent, corresponding historical time window change rate, maximum connected subgraph size, isolated node ratio, negative response ratio, group size, survival time, and current time period are uniformly encoded into a continuous state vector, forming a state space that comprehensively reflects the operational status of the target group. Then, using this state vector as input, the policy network outputs a hybrid action space containing discrete policy selection and continuous intervention intensity. Discrete actions include no intervention, pushing new user information, performing visibility adjustment, and pushing new topic information, while continuous actions are used to finely control the intervention magnitude. Next, the selected action is applied to the current target group, and a comprehensive reward function is calculated based on the group state in the next preset time window. This function takes into account the increase in interaction score, the penalty for intervention cost, and the reward for group retention, which can achieve the optimization goal of being imperceptible, low-intrusive, and high-return. Finally, based on the training samples composed of state, action, and reward, the policy network and value network are updated iteratively in sync, so that the model can continuously learn the optimal decision mode. This enables it to automatically match the most suitable adjustment strategy (such as pushing new user information, performing visibility adjustment, or pushing new topic information) for different pathological structures of the target group, forming a complete closed-loop adjustment capability of self-awareness, self-decision, self-execution, and self-optimization.

[0107] In one embodiment of this application, after dividing the user into multiple categories, a preset information type matching each user category can be determined, and information of the corresponding preset information type can be pushed to users belonging to each user category.

[0108] Specifically, preset information types refer to predefined sets of information with clear content themes or format categories, such as technology news, relationship articles, entertainment information, learning materials, life tips, and product recommendations. Each information type corresponds to a category of content resources, typically set by the platform operator based on the common needs of the user group or business goals. Information types can be further subdivided into subcategories (e.g., technology news can be divided into artificial intelligence, consumer electronics, and space exploration) to achieve more precise content matching.

[0109] For example, the basis for determining the preset information type that matches each user category could be the type of content that users in the user category most frequently read or interact with in their historical behavior; the social behavior style of the category, such as user categories with shorter text lengths and more enthusiastic tone being more suitable for pushing trending news, while user categories with longer text lengths and more neutral tone are more suitable for pushing in-depth analysis articles; or the click-through rate or conversion rate of different information types in each user category obtained from offline testing.

[0110] Pushing information of a pre-defined type to users within a user category can be achieved by selecting one or more specific pieces of information belonging to that type from the content library and sending them to each user in that category in an appropriate manner (such as in-app notifications, site messages, news feed recommendations, push notifications, etc.). Push notifications can be batch-based, scheduled, or triggered in real-time based on user online status. It's important to note that this information push does not depend on group activity or real-time interaction ratings, but rather on the user category itself, thus remaining independent of the information push adjustment layer. Through this mechanism, all users within the same category will receive similar information, achieving both "categorized operation" and "scale-personalized" content distribution effects.

[0111] Figure 4 A flowchart illustrating the data processing method provided in an embodiment of this application is shown schematically. Figure 4 As shown, the data processing method provided in this embodiment specifically includes the following steps.

[0112] Step 401: Obtain user behavior data for the first user and the second user respectively. Read user behavior data of the first user and the second user within a preset historical time window (e.g., the past 90 days) from the user database, such as the timestamp sequence of speech, text message content (including length and punctuation usage), acoustic characteristics of voice messages (duration, volume variance, speech rate), and the type and frequency of emoticon usage.

[0113] Step 402: Determine the target similarity between the first user and the second user based on user behavior data. Feature extraction is performed on the user behavior data of the first and second users respectively, resulting in a first time series, a second time series, a first behavior feature vector, and a second behavior feature vector. The dynamic time-normalized distance between the first and second time series is calculated as the first matching degree, and the cosine similarity between the first and second behavior feature vectors is calculated as the second matching degree. The first and second matching degrees are weighted and fused according to preset weights to obtain the target similarity. The higher the similarity value, the more compatible the two users are in terms of social behavior styles.

[0114] Step 403: Determine the interaction intensity index between the first user and the second user based on the historical interaction data between them. Obtain historical interaction data from the user database and count the following: the number of times the first user actively initiates an interaction with the second user (as the first active interaction count), and the number of times the first user responds to messages received from the second user (as the first passive response count); similarly, obtain the second active interaction count and the second passive response count. Weight the first set of data based on a first preset weight to obtain the first calculation result, and perform similar processing on the second set of data based on a second preset weight to obtain the second calculation result. Then, introduce a preset time decay factor, multiply the first and second calculation results by the decay factor respectively, and take the geometric or arithmetic mean to obtain the interaction intensity index, quantifying the recent interaction activity between the first user and the second user.

[0115] Step 404: Determine the target matching degree between the first user and the second user based on the target similarity and interaction intensity index. First, determine whether the target similarity is greater than or equal to a preset first value and whether the interaction intensity index is greater than or equal to a preset second value. If both are satisfied, it indicates that the two users have reached an acceptable level in terms of behavioral style compatibility and recent interaction activity, so the product of the two is taken as the target matching degree; if either condition is not satisfied, the target matching degree is set to 0 or a very small value, indicating that the two users do not have a valid matching basis.

[0116] Step 405: Classify users in the user database according to their target matching degree, resulting in multiple user categories. Using all users in the user database as nodes and the target matching degree between each pair of users as edge weights, construct a first weighted graph. Divide this weighted graph using a spectral clustering algorithm or the Leuven community detection algorithm, resulting in several subgraphs, each corresponding to a user category. Within the same user category, the target matching degree between users is generally greater than or equal to a preset matching value; between users from different user categories, the target matching degree is generally less than the preset matching value.

[0117] Step 406: Construct target groups based on user categories and / or push information to users within user categories. On one hand, for each user category, calculate the sum of the target matching scores of all user pairs within that category, divided by the total number of user pairs, to obtain the average matching score for that category. Select users from the user category with the highest average matching score as target users, physically create one or more target groups based on these target users, and initialize the group runtime data structure. On the other hand, determine the preset information types matched for each user category. For example, by analyzing the historical reading preferences of users within the category or by using statistics based on interest tags, generate a category-information type mapping table, and then push the corresponding type of information (such as technology news, emotional articles, entertainment information, etc.) to all users belonging to that category on a daily or weekly basis. It should be noted that these two operations can be performed simultaneously or selectively.

[0118] Step 407: Monitor the interaction data of the target group within a preset time period and adjust the information push accordingly. After the target group is running, continuously collect multimodal interaction data (text, voice, reply delay, emoticons, etc.) of all members in the group at preset time intervals. Based on this data, construct a second weighted graph (directed weighted graph) in real time and extract three interaction features from the second weighted graph: interaction density, interaction centrality, and splitting index. The three interaction features are weighted and fused to obtain an interaction score. When the interaction score is less than the first preset score, the system determines that the group is in a state of inactivity, retrieves new users with high interaction stimulation capabilities from the user database, and pushes new user information to them; when the interaction score is greater than or equal to the first preset score and less than the second preset score, the system determines that the group has a centralization or slight splitting tendency. At this time, a neutral topic is generated using a large language model and a new topic information is pushed in the name of the group assistant; when the interaction score reaches or exceeds the second preset score, no intervention is taken. In addition, if the splitting index is too high alone, visibility adjustment can be triggered to collapse extreme statements.

[0119] This embodiment sets dual thresholds for target similarity and interaction intensity index, ensuring that only user groups with matching social behavior styles and high interaction activity are included in the same user category or group, thus improving the accuracy of user classification and the initial quality of groups from the source. Through the real-time construction of the second weighted graph and the quantitative evaluation of interaction scores, it can sensitively capture pathological structures such as group silence, centralization, and splitting, and then adjust information push in a targeted manner, significantly extending the group life cycle, improving user retention rate and interaction quality, and providing social platforms with quantifiable, early warning, and self-healing intelligent group management capabilities.

[0120] It should be noted that although the steps of the method in this application are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0121] The following describes an apparatus embodiment of this application, which can be used to execute the data processing method described in the above embodiments of this application. Figure 5 A schematic block diagram of the data processing apparatus provided in an embodiment of this application is shown. Figure 5 As shown, the data processing apparatus provided in this application embodiment includes: The similarity determination module 510 is used to acquire user behavior data of a first user and user behavior data of a second user from the user database, and determine the target similarity between the first user and the second user based on the user behavior data of the first user and the second user; the user behavior data is used to indicate the user's social behavior style; the target similarity characterizes the degree of matching between the first user and the second user based on the social behavior style; the first user and the second user are two different users in the user database; The intensity index determination module 520 is used to determine the interaction intensity index between the first user and the second user based on the historical interaction data between the first user and the second user; the interaction intensity index is used to indicate the frequency of interaction between the first user and the second user. The matching degree determination module 530 is used to determine the target matching degree between the first user and the second user based on the target similarity and the interaction intensity index; User category determination module 540 is used to classify users in the user database according to the target matching degree to obtain multiple user categories.

[0122] In one embodiment of this application, the similarity determination module 510 is specifically used for: Feature extraction is performed on the user behavior data of the first user to obtain a first time series and a first profile feature vector; the first time series includes the timestamp sequence of messages sent by the first user; the first profile feature vector represents the social behavior style of the first user. Feature extraction is performed on the user behavior data of the second user to obtain a second time series and a second profile feature vector; the second time series includes the timestamp sequence of messages sent by the second user; the second profile feature vector represents the social behavior style of the second user. Calculate the first degree of matching between the first time series and the second time series, and the second degree of matching between the first portrait feature vector and the second portrait feature vector; The target similarity between the first user and the second user is determined based on the first matching degree and the second matching degree.

[0123] In one embodiment of this application, the intensity index determination module 520 is specifically used for: Based on the historical interaction data between the first user and the second user, the first number of active interactions and the first number of passive responses corresponding to the first user are determined; the first number of active interactions includes the number of times the first user sends information to the second user; the first number of passive responses includes the number of times the first user receives information sent by the second user. Determine the second number of active interactions and the second number of passive responses for the second user; the second number of active interactions includes the number of times the second user sends information to the first user; the second number of passive responses includes the number of times the second user receives information from the first user. The first active interaction count and the first passive response count are weighted and summed based on the first weight to obtain the first calculation result, and the second active interaction count and the second passive response count are weighted and summed based on the second weight to obtain the second calculation result; Based on the first calculation result, the second calculation result, and the preset time decay factor, the interaction intensity index between the first user and the second user is determined; the preset time decay factor is used to measure the rate at which the influence weight of the historical interaction data on the interaction intensity index decays over time.

[0124] In one embodiment of this application, the matching degree determination module 530 is specifically used for: If the target similarity is greater than or equal to a first preset value, and the interaction intensity index is greater than or equal to a second preset value, then the product of the target similarity and the interaction intensity index is taken as the target matching degree between the first user and the second user.

[0125] In one embodiment of this application, the user category determination module 540 is specifically used for: Using users in the user database as nodes, constructing edges based on the interaction behavior between adjacent users, and using the target matching degree between adjacent users as edge weights, a first weighted graph is constructed. The first weighted graph is divided into multiple subgraphs; each subgraph corresponds to a user category; the target matching degree between users within a user category is greater than or equal to a preset matching value, and the target matching degree between users in different user categories is less than the preset matching value.

[0126] In one embodiment of this application, the data processing apparatus further includes a target group establishment module, which includes: For each user category, determine the corresponding user group; the user group includes two different users from the user category. Calculate the sum of the target matching degree of all user groups in the user category, and determine the average matching degree of the user category as the ratio between the sum of the target matching degree and the number of user groups; Users in the user category with the highest average matching degree are selected as target users, and target groups are constructed based on the target users.

[0127] In one embodiment of this application, the data processing apparatus further includes a group management module, which includes: At preset time intervals, acquire the interaction data generated by each target user in the target group within the preset time interval; Construct a second weighted graph corresponding to the target group based on the interaction data generated within the preset time period; Feature extraction is performed on the second weighted graph to obtain multiple interaction features, and the interaction score of the target group is calculated based on the multiple interaction features; the interaction score represents the frequency of interaction between each target user in the target group. Adjustment information is pushed to the target group based on the interaction score; the adjustment information is used to improve the interaction score of the target group.

[0128] In one embodiment of this application, the group management module includes: Based on the interaction data generated by the first target user and the second target user in the target group within the preset time, a first delay time and a second delay time are determined; the first delay time includes the interval time for the first target user to reply to the message sent by the second target user; the second delay time includes the interval time for the second target user to reply to the message sent by the first target user; the first target user and the second target user are two different target users in the target group; A first preset weight is determined based on the first delay time, and a second preset weight is determined based on the second delay time; Using each target user in the target group as a node, directed edges are constructed based on the interaction behavior between the first target user and the second target user to form a second weighted graph corresponding to the target group; wherein, the arrow direction of the directed edge points to the passive responder of the interaction behavior; the first preset weight is the weight of the directed edge where the first target user initiates an interaction behavior with the second target user, and the second preset weight is the weight of the directed edge where the second target user initiates an interaction behavior with the first target user.

[0129] In one embodiment of this application, the group management module includes: If the interaction score is less than the first preset score, then at least one new user information is pushed to the target group based on the interaction data generated within the preset time period; If the interaction score is greater than or equal to the first preset score and less than the second preset score, then the new topic information is pushed to the target group based on the interaction data generated within the preset time period; the first preset score is less than the second preset score.

[0130] In one embodiment of this application, the data processing apparatus further includes an information push module, which includes: Determine a preset information type that matches each user category, and push information of the corresponding preset information type to users belonging to each user category.

[0131] The specific details of the data processing apparatus provided in the various embodiments of this application have been described in detail in the corresponding method embodiments, and will not be repeated here.

[0132] Figure 6 A schematic diagram of the computer system architecture used to implement the technical solution of this application is shown.

[0133] It should be noted that, Figure 6 The computer system 600 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0134] like Figure 6As shown, the computer system 600 includes a central processing unit (CPU) 601, which performs various appropriate actions and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 607 into random access memory (RAM). The RAM 603 also stores various programs and data required for system operation. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output interface 605 (I / O interface) is also connected to the bus 604.

[0135] The following components are connected to the input / output interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a local area network card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0136] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0137] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0138] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0139] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the method according to the embodiments of this application.

[0140] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0141] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A data processing method, characterized in that, include: Obtain user behavior data of a first user and user behavior data of a second user from the user database, and determine the target similarity between the first user and the second user based on the user behavior data of the first user and user behavior data of the second user; The target similarity represents the degree of matching between the user behavior data of the first user and the user behavior data of the second user; The first user and the second user are two different users in the user database; Based on the historical interaction data between the first user and the second user, an interaction intensity index is determined between the first user and the second user; the interaction intensity index is used to indicate the frequency of interaction between the first user and the second user. The degree of target matching between the first user and the second user is determined based on the target similarity and the interaction intensity index. Users in the user database are categorized based on the degree of target matching.

2. The data processing method according to claim 1, characterized in that, The step of determining the target similarity between the first user and the second user based on the user behavior data of the first user and the user behavior data of the second user includes: Feature extraction is performed on the user behavior data of the first user to obtain a first time series and a first behavior feature vector; the first time series includes the timestamp sequence of messages sent by the first user. Feature extraction is performed on the user behavior data of the second user to obtain a second time series and a second behavior feature vector; the second time series includes the timestamp sequence of messages sent by the second user. Calculate the first degree of matching between the first time series and the second time series, and the second degree of matching between the first behavioral feature vector and the second behavioral feature vector; The target similarity between the first user and the second user is determined based on the first matching degree and the second matching degree.

3. The data processing method according to claim 1, characterized in that, The step of determining the interaction intensity index between the first user and the second user based on historical interaction data between the first user and the second user includes: Based on the historical interaction data between the first user and the second user, the first number of active interactions and the first number of passive responses corresponding to the first user are determined; the first number of active interactions includes the number of times the first user sends information to the second user; the first number of passive responses includes the number of times the first user receives information sent by the second user. Determine the second number of active interactions and the second number of passive responses for the second user; the second number of active interactions includes the number of times the second user sends information to the first user; the second number of passive responses includes the number of times the second user receives information from the first user. The first active interaction count and the first passive response count are weighted and summed based on the first weight to obtain the first calculation result, and the second active interaction count and the second passive response count are weighted and summed based on the second weight to obtain the second calculation result; Based on the first calculation result, the second calculation result, and the preset time decay factor, the interaction intensity index between the first user and the second user is determined; the preset time decay factor is used to measure the rate at which the influence weight of the historical interaction data on the interaction intensity index decays over time.

4. The data processing method according to claim 1, characterized in that, Determining the target matching degree between the first user and the second user based on the target similarity and the interaction intensity index includes: If the target similarity is greater than or equal to a first preset value, and the interaction intensity index is greater than or equal to a second preset value, then the product of the target similarity and the interaction intensity index is taken as the target matching degree between the first user and the second user.

5. The data processing method according to claim 1, characterized in that, The step of classifying users in the user database according to the target matching degree includes: Using users in the user database as nodes, constructing edges based on the interaction behavior between adjacent users, and using the target matching degree between adjacent users as edge weights, a first weighted graph is constructed. The first weighted graph is divided into multiple subgraphs; each subgraph corresponds to a user category; the target matching degree between users within the same user category is greater than or equal to a preset matching value, and the target matching degree between users in different user categories is less than the preset matching value.

6. The data processing method according to claim 1, characterized in that, After classifying users in the user database according to the target matching degree, the method further includes: For each user category determined by the classification, any two users corresponding to the user category are grouped together as a user group; Calculate the sum of the target matching degree corresponding to all user groups belonging to the user category, and determine the average matching degree of the user category as the ratio between the sum of the target matching degree and the number of user groups; Users in the user category with the highest average matching degree are selected as target users, and target groups are constructed based on the target users.

7. The data processing method according to claim 6, characterized in that, After constructing the target group based on the target user, the method further includes: At preset time intervals, acquire the interaction data generated by each target user in the target group within the preset time interval; Construct a second weighted graph corresponding to the target group based on the interaction data generated within the preset time period; Feature extraction is performed on the second weighted graph to obtain multiple interaction features, and the interaction score of the target group is calculated based on the multiple interaction features; the interaction score represents the frequency of interaction between each target user in the target group. Adjustment information is pushed to the target group based on the interaction score; the adjustment information is used to improve the interaction score of the target group.

8. The data processing method according to claim 7, characterized in that, The step of constructing a second weighted graph based on the interaction data generated within the preset time period includes: Each target user in the target group is taken as a node in the second weighted graph. If there is an interaction between the first target user and the second target user, a directed edge is constructed between the node where the first target user is located and the node where the second target user is located in the second weighted graph. The first target user and the second target user are two different target users in the target group, and the direction of the directed edge is from the initiator of the interaction to the receiver of the interaction. A first delay time is determined based on the interaction data generated by the first target user within the preset time period; the first delay time includes the interval between the first target user responding to the message sent by the second target user; The second delay time is determined based on the interaction data generated by the second target user within the preset time period; the second delay time includes the interval between the second target user responding to the message sent by the first target user; The weight of the edge pointing from the node where the first target user is located to the node where the second target user is located is determined based on the first delay time, and the weight of the edge pointing from the node where the second target user is located to the node where the first target user is located is determined based on the second delay time.

9. The data processing method according to claim 7, characterized in that, The adjustment information includes new user information and new topic information; The step of adjusting the information push to the target group based on the interaction score includes: If the interaction score is less than the first preset score, then at least one new user information is pushed to the target group based on the interaction data generated within the preset time period; If the interaction score is greater than or equal to the first preset score and less than the second preset score, then the new topic information is pushed to the target group based on the interaction data generated within the preset time period; the first preset score is less than the second preset score.

10. The data processing method according to claim 1, characterized in that, After classifying users in the user database according to the target matching degree to obtain multiple user categories, the method further includes: Determine a preset information type that matches each user category, and push information of the corresponding preset information type to users belonging to each user category.

11. A data processing apparatus, characterized in that, include: The similarity determination module is used to obtain user behavior data of a first user and user behavior data of a second user from the user database, and determine the target similarity between the first user and the second user based on the user behavior data of the first user and the user behavior data of the second user. The target similarity represents the degree of matching between the user behavior data of the first user and the user behavior data of the second user; The first user and the second user are two different users in the user database; The intensity index determination module is used to determine the interaction intensity index between the first user and the second user based on the historical interaction data between the first user and the second user; the interaction intensity index is used to indicate the frequency of interaction between the first user and the second user. A matching degree determination module is used to determine the target matching degree between the first user and the second user based on the target similarity and the interaction intensity index; The user category determination module is used to classify users in the user database according to the target matching degree.

12. An electronic device, characterized in that, include: processor; as well as A memory for storing executable instructions of the processor; wherein the processor executes the executable instructions to cause the electronic device to perform the data processing method according to any one of claims 1 to 10.

13. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the data processing method according to any one of claims 1 to 10.

14. A computer program product, characterized in that, The computer program product includes computer instructions, which are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from a computer-readable storage medium, and executes the computer instructions to cause the computer device to perform the data processing method according to any one of claims 1 to 10.

Images