User care method and system based on real-time user portrait

Abstract

The invention relates to the technical field of information, in particular to a user care method and system based on a real-time user portrait. The method comprises the following steps: firstly, according to real-time acquisition and analysis calculation of user credit rating, product use behaviors and consumption behaviors, through an optimized user behavior data real-time calculation method, efficiently calculating to obtain user behavior characteristics and credit characteristics, generating a user characteristic vector matrix, and further performing similarity calculation with a typical user template matrix; therefore, a user scoring model combining multiple dimensions of user credit, consumption behaviors and the like is obtained. According to the user scoring model, a multi-dimensional user type can be abstracted from behaviors of a single user, and a plurality of user care strategies are matched to be applied to a plurality of actual scenes, so that the pain point problems of product consultation, product complaint, ordering and unsubscribing and the like of the user are solved comprehensively; and meanwhile, related requirements of product risk control, precision marketing, user retention and the like of the business platform are met.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a user care method and system based on real-time user portraits. Background Art

[0002] In order to meet the needs of precise operation of Internet product platforms, it has become a necessary technical means for major platforms to collect user behavior data, conduct big data analysis, generate user tags of multiple dimensions, and match personalized product recommendations based on user portraits through the background collection of user behavior data. Introducing a natural language processing model based on artificial intelligence on the basis of personalized recommendations can realize intelligent customer service functions, complete automated and personalized interactive feedback between the platform and users, further reduce the workload of platform marketing and customer service, and improve user experience.

[0003] In terms of technical implementation, the platform collects user behavior logs from the log server in real time, transmits the logs to the log processing process through the message queue for cleaning and filtering, and uses big data to perform real-time or scheduled analysis, classifies the data, and saves it in the user tag library to complete a multi-dimensional user portrait related to product features. Based on the user portrait, the user's preferences can be analyzed to achieve personalized recommendations for the user; or based on the user tag, the user's business needs can be predicted in advance, and the customer service staff can be reminded to use the corresponding words to guide in time, or the intelligent customer service can realize automated Q&A and consulting services.

[0004] When the platform scale grows to more than 10 million users, it is necessary to process massive amounts of user data in real time and respond efficiently in order to meet the requirements of business timeliness and refined operations. Relevant technical methods have been published for real-time user portraits (such as Chinese patent application CN114663132A), but they do not involve how to optimize and implement efficient real-time calculation methods for massive data, nor do they combine with actual business characteristics. At the same time, from the perspective of business development, relatively single personalized recommendations and intelligent customer service cannot meet the requirements of service quality standardization. It is necessary to consider the establishment of a user care system from a holistic perspective to comprehensively solve users' customer service consultation needs such as product consultation, product complaints, subscription cancellations, and platform operators' product risk control, precision marketing, user retention and other related needs. Summary of the invention

[0005] The purpose of the present invention is to address the problems existing in the background technology and propose a user care method and system based on real-time user portraits, which is suitable for 2C application scenarios of content-based and capability-based business platforms of telecom operators. First, based on the real-time collection and analysis of user credit ratings, product usage behaviors and consumption behaviors, an optimized real-time calculation method for user behavior data is used to efficiently calculate user behavior characteristics and credit characteristics, generate a user feature vector matrix, and further perform similarity calculation with a typical user template matrix, thereby obtaining a user scoring model that combines multiple dimensions such as user credit and consumption behavior. According to the user scoring model, multi-dimensional user types can be abstracted from the behavior of a single user, and a variety of user care strategies can be matched and applied to a variety of actual scenarios, solving the user's pain points such as product consultation, product complaints, and subscription cancellations in an all-round way, while also meeting the product risk control, precision marketing, user retention and other related needs of the business platform.

[0006] This user care system adapts to the business characteristics of content-based and capability-based 2C products, such as business and technology integration, personalized user needs, and flexible payment methods. It provides the basic conditions for the subsequent data-driven multi-channel marketing and personalized services, and can better focus on the needs of key users, high-value users and potential customers.

[0007] The technical solution of the present invention, in a first aspect, provides a user care method based on real-time user portrait, comprising the following specific steps:

[0008] S1. Data collection: collect user behavior logs and user attributes in real time through the log server cluster;

[0009] In step S1, the user behavior logs and user attributes uploaded to the platform by the user through the product client are collected by the log server cluster in real time through the business interface, background service, real-time printing or external platform business interface transmission;

[0010] User behavior logs specifically include: product usage, ordering / payment, cancellation, and consultation and complaint behavior;

[0011] User attributes include the length of time the user has been online, the level of their call package, and whether they have any outstanding payments.

[0012] For user attribute data that is not time-sensitive, it is obtained by extracting files at regular intervals to reduce system load; some user attribute data obtained from third-party platforms such as operators are non-real-time and updated regularly, such as the user's cumulative online time (days) and call package level (months), which are time-insensitive user attributes.

[0013] S2. Clean the collected raw data;

[0014] The data cleaning steps in step S2 include data consistency check and deduplication, outlier processing, and normalization.

[0015] S3, data processing: processing the cleaned data; the processing process includes real-time data processing and offline data processing;

[0016] Step S3 also includes the following steps:

[0017] S31. Save all the user's ordering / payment behaviors in the last 24 hours and sort them by time. Each element in the list contains a timestamp (t n ) and an integer (a n ), records the startup data of the ordering / payment behavior at a certain moment in the past 24 hours; and is expressed as:

[0018] [(t1,a1),(t2,a2),…,(t n , a n )]

[0019] S32, cumulative order / payment data: indicates the cumulative number of times up to the last order / payment, expressed as Sum;

[0020] S321, add a key hash process, and send the data with the same key to the same Reducer in MapReduce for processing, to ensure that the same user can fall into the same processing process;

[0021] S322, as the user ordering / payment data collected by the log server in real time enters the message queue, the processing process will read this information in real time and modify the corresponding user's data;

[0022] Get the current time t c ;

[0023] Read historical data

[0024] Read user history data from the local cache: [(t1, a1), (t2, a2), …, (t n , a n )], the cumulative value is SumA; if there is no local cache information, read the user data from the database;

[0025] S33, cumulative value increases: SumA = SumA + 1;

[0026] S34, historical data update;

[0027] t c The user behavior at that moment is written into the historical ordering / payment behavior list, that is, the list is updated to:

[0028] [(t1,a1),(t2,a2),…,(t n , a n ), (tc, 1)]

[0029] If you encounter c =t n In the special case, it is updated to:

[0030] [(t1,a1),(t2,a2),…,(t n , a n +1)]

[0031] Because the calculation is within 24 hours, it is also necessary to delete the timestamp less than t in the list. c -24 hour element, and subtract the accumulated value corresponding to that moment 24 hours ago from SumA;

[0032] S4, real-time storage and real-time calculation of user behavior data to obtain user behavior feature labels;

[0033] S5. Offline storage and calculation of user attribute data to obtain user credit feature labels

[0034] S6. Build a user scoring model based on the user behavior feature labels and the user credit feature labels; and process the data to obtain a user portrait based on the calculation results of the user scoring model;

[0035] S61, after respectively performing real-time and offline calculations on the user behavior data and the user attribute data, obtaining the user's behavior feature label and credit feature label;

[0036] S62, merging and normalizing the behavior feature labels and the credit feature labels to obtain a feature vector matrix F of multiple dimensions for a specific user, where each row represents a feature vector of one dimension;

[0037] For a certain user, the feature vector matrix with n categories of business features is expressed as:

[0038]

[0039] S63, updating the typical user template matrix T according to the user behavior feature label and the credit feature label; the typical user template matrix is ​​obtained by offline calculation under preset conditions, and each row represents a typical user of a certain type of business feature; each row of the typical user template matrix corresponds to each other with respect to the feature vector matrix;

[0040]

[0041] S64, calculate the cosine similarity between the feature vector matrix and the typical user template matrix to obtain the user rating vector S:

[0042] S=[s1,s2,…,s n ]

[0043] The value of the element Sn in the vector S represents the degree of match between a specific user and the typical user of the nth category;

[0044] S7. Select a variety of corresponding user care strategies for the user based on the obtained user portrait.

[0045] A second aspect of the present invention provides a user care system based on real-time user portraits, which executes the above method and includes a data collection layer, a data processing layer, a data application layer and a business logic layer;

[0046] The data collection layer is used to collect and clean the raw data;

[0047] The data processing layer includes a computing module for real-time data processing and offline data processing of the cleaned data;

[0048] The data application layer aggregates, converts, and loads relevant data based on the calculation results of user behavior characteristics and user credit characteristics to realize user profiling and provide data services for the business logic layer.

[0049] The business logic layer automatically selects the corresponding user care strategy based on the data provided by the data application layer, realizing the user care function that integrates customer service and precise operation.

[0050] Preferably, the user care strategy includes product consultation, product complaints, subscription and cancellation plans for users, as well as the platform operator's product risk control, precision marketing, and user retention plans.

[0051] Preferably, the business logic layer updates the relevant log data of this user care to the data collection layer to realize a business closed loop.

[0052] Compared with the prior art, the present invention has the following beneficial technical effects:

[0053] 1. For the real-time data calculation module of the data processing layer, a real-time calculation optimization method for user behavior data for massive users is proposed, which parallelizes the calculation operation to avoid the overhead of database locking; at the same time, different persistence strategies are adopted for historical data and cumulative data, and both are cached in the calculation process. The cumulative data is updated in real time, and the historical data is updated to the database regularly. In this way, in a high-stability system, resource competition can be reduced, and the database pressure, I / O and serialization / deserialization times can be effectively reduced, thereby improving calculation efficiency.

[0054] 2. Based on user feature extraction, a user rating model combining multiple dimensions such as user credit, usage behavior and consumption behavior is proposed. The multi-dimensional labels of users are calculated in real time from user attributes and user behavior data to generate feature vectors. After similarity calculation, multi-dimensional user types can be abstracted for a single user, and multiple user care strategies can be further matched. This model is more suitable for the business characteristics of content-based and capability-based 2C products, and provides basic conditions for carrying out data-driven personalized services and multi-channel marketing.

[0055] 3. The user care system that runs through the entire business operation process realizes the integration of customer service and precise operation, and comprehensively solves users' pain points such as product consultation, product complaints, subscription cancellations, etc., while also meeting the platform operator's product risk control, precision marketing, user retention and other related needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A schematic diagram of the system structure of an embodiment of the present invention;

[0057] Figure 2 A schematic diagram of a method for real-time calculation of user behavior data optimized in an embodiment of the present invention;

[0058] Figure 3 The data processing flow chart in the embodiment of the present invention. DETAILED DESCRIPTION

[0059] Example 1

[0060] The present invention proposes a user care method based on real-time user portrait, which includes the following specific steps:

[0061] Data collection: collect user behavior logs and user attributes in real time through the log server cluster;

[0062] Collect user behavior logs and user attributes uploaded to the platform through product clients through real-time collection of business interfaces, backend services, real-time printing, or external platform business interface transmission via log server clusters;

[0063] User behavior logs specifically include: product usage, ordering / payment, cancellation, and consultation and complaint behavior;

[0064] User attributes include the length of time the user has been online, the level of their call package, and whether they have any outstanding payments.

[0065] For user attribute data that is not time-sensitive, it is obtained by extracting files at a scheduled time to reduce the system load.

[0066] Clean the collected raw data;

[0067] Data processing: Process the cleaned data; the processing process includes two parts: real-time data processing and offline data processing; real-time storage and real-time calculation of user behavior data are performed to obtain user behavior characteristics; user attribute data is stored and calculated offline to obtain user credit characteristics; specifically, offline storage and calculation are relatively common methods, and the commonly used MapReduce framework can also be used for calculation. If the data volume is small, it can also be calculated directly using the database. This solution does not limit the specific calculation method.

[0068] The following example illustrates the real-time calculation method of user behavior data:

[0069] Figure 2 This represents a scenario where three users (User1, User2, and User3) parallelly calculate ordering / payment data within 24 hours. The processing processes (or threads) are P1, P2, and P3 respectively. Real-time calculation of other types of behavioral data is similar.

[0070] The platform saves a copy of the ordering / payment data for each user. The data structure is as follows:

[0071] (1) Historical ordering / payment behavior list (also called historical data): saves all ordering / payment behaviors of the user in the last 24 hours, sorted by time. Each element in the list contains a timestamp (t n ) and an integer (a n ), records the startup data of the ordering / payment behavior at a certain moment (accurate to ms) in the past 24 hours. It is expressed as:

[0072] [(t1,a1),(t2,a2),…,(t n ,a n )];

[0073] (2) Cumulative order / payment data (also called cumulative value): indicates the cumulative number of times up to the last order / payment, expressed as Sum.

[0074] In the traditional processing mode, in order to ensure data consistency, the database is generally locked when reading user historical data, and the lock is released after the new historical data and accumulated values ​​are output to the database. To improve real-time processing efficiency and concurrent throughput, and avoid locking the database, the following optimization is performed:

[0075] (1) Add a key hash process, and hand over the data with the same key to the same Reducer in MapReduce for processing, to ensure that the same user can fall into the same processing process; after log data collection and preprocessing, a key field can be extracted or generated from it to identify a specific user; hash processing is to ensure the integrity of key data by calculating the hash value of key data, that is, to ensure that it is the data of a specific user. Subsequently, the specific user data is processed by the same process.

[0076] The purpose of MapReduce processing is to use the big data processing framework to batch process log data, corresponding to Figure 2 P1 P2 P3 in order to obtain the historical ordering / payment behavior list and cumulative value.

[0077] (2) Considering that in actual applications, the main concern when counting user behavior is real-time cumulative values, different persistence strategies are adopted for historical data and cumulative values, that is, both are cached in the calculation process, but only the cumulative values ​​are persisted in real time. The cumulative values ​​update the value database in real time, while the historical data are updated to the database regularly.

[0078] As the user ordering / payment data collected by the log server in real time enters the message queue, the processing process will read this information in real time and modify the corresponding user's data. For example, when P1 reads the payment record of user A at time t, the specific operation process is as follows:

[0079] (1) Get the current time t c ;

[0080] (2) Read historical data

[0081] Read user history data from the local cache: [(t1,a1), (t2,a 2) ,…,(t n ,a n )], the cumulative value is SumA. If there is no local cache information, read the user data from the database;

[0082] (3) Cumulative value increases

[0083] SumA=SumA+1;

[0084] (4) Historical data update

[0085] Write the user behavior at time tc into the historical ordering / payment behavior list, that is, update the list to:

[0086] [(t1,a1),(t2,a2),…,(t n ,a n ),(tc ,1)]

[0087] If you encounter c =t n In the special case, it is updated to:

[0088] [(t1,a1),(t2,a2),…,(t n ,a n +1)]

[0089] Because the calculation is within 24 hours, it is also necessary to delete the timestamp less than t in the list. c -24 hour element, and subtract the accumulated value corresponding to that moment 24 hours ago from SumA.

[0090] In scenarios with a large number of users and high-frequency access, the above solution can significantly reduce database pressure, I / O, and serialization / deserialization times, thereby improving processing efficiency.

[0091] The data application layer aggregates, converts, and loads relevant data based on the calculation results of user behavior characteristics and user credit characteristics to realize user profiling (including user tags and user ratings) and provide data services for the business logic layer.

[0092] Based on real-time log collection and data calculation, real-time monitoring of business can also be achieved at the data application layer, such as real-time user access times, order times, daily cumulative revenue and other operational indicators, as well as concurrent connections of important business interfaces and other operation and maintenance indicators. In actual applications, the real-time statistical data packets are stored in redis, and a web monitoring system is built to graphically display multi-dimensional real-time data.

[0093] The specific data processing steps are as follows: Figure 3 As shown:

[0094] (1) After real-time and offline calculations of user behavior data and user attribute data, the user's behavior feature label and credit feature label are obtained;

[0095] (2) The behavior feature labels and credit feature labels are merged and normalized to obtain a feature vector matrix F of multiple dimensions for a specific user, where each row represents a feature vector of one dimension (corresponding to a type of business feature). For a certain user, the feature vector matrix with n types of business features is expressed as:

[0096]

[0097] (3) Update the typical user template matrix T according to the user behavior feature label and credit feature label. The typical user template matrix is ​​calculated offline under preset conditions, and each row represents a typical user of a certain type of business feature. The typical user template matrix corresponds to each row of the feature vector matrix;

[0098]

[0099] For example: the first row of matrix T refers to credit, t11, t12, t13 respectively represent the user's package level code (1 / 2 / 3... different values ​​represent different levels), whether there is arrears (0 / 1), and the length of time online (expressed in days, a natural number); the second row of matrix T refers to service loyalty, t21, t22, t23 respectively represent the user's number of uses (natural numbers), usage time (in minutes), cumulative payment amount (accurate to 0.01), and so on.

[0100] The preset conditions refer to a typical user characteristic determined according to the needs of business operation. For example, the characteristics of high-credit users can be expressed as: package level t11>=3; whether there is arrears t12=0; online time t13>=1800 days. The specific conditions can be adjusted by the business operator.

[0101] (4) Calculate the cosine similarity between the feature vector matrix and the typical user template matrix to obtain the user rating vector S:

[0102] S=[s1,s2,…,s n ]

[0103] The element S in vector S n The value of indicates the matching degree between a particular user and the nth type of typical user. Furthermore, a plurality of corresponding user care strategies can be automatically selected for the user in the background.

[0104] Taking user care of a content-based monthly subscription service as an example, the specific implementation steps are as follows:

[0105] First, according to the actual business needs, determine the types of user characteristics (the number and meaning of specific types can be adjusted by the platform operator), including:

[0106] Feature 1, credit, is represented by the user's package level, whether they are in arrears, and the length of time they have been online;

[0107] Feature 2, service loyalty, is represented by the number of times a user uses the service, the duration of use, the number of services ordered, the number of payments, and the cumulative payment amount;

[0108] Feature 3: Customer service level, expressed in terms of the number of consultations, cumulative number of cancellations, and cumulative number of complaints.

[0109] Step 1: According to the proposed user feature type, the behaviors of all users in the past period of time are counted from the database to obtain a typical user template matrix T containing three rows;

[0110] Step 2: Collect user behavior logs and background interface service logs in real time, and obtain user attribute data such as user package level and whether the user is in arrears from the external platform through the interface at regular intervals;

[0111] Step 3: After real-time calculation, the user's behavior feature label and user credit feature label are obtained, with the user ID as the unique identifier;

[0112] Step 4: Merge and normalize the behavior feature labels and credit feature labels to obtain a feature vector matrix F with the user ID as the index, which represents the characteristics of the user in three dimensions: credit, business loyalty, and customer service level;

[0113] Step 5: Calculate the cosine similarity between the feature vector matrix F identified by the user ID and the typical user template matrix T obtained in step 1 to obtain a user rating vector S containing three elements;

[0114] Step 6: According to the user rating value contained in S, select the corresponding care strategy, which can be implemented through automatic system processing or manual service, for example:

[0115]

[0116] The business logic layer automatically selects the corresponding user care strategy based on the user rating model provided by the data application layer to realize the user care function that integrates customer service and precise operation, including: product consultation, product complaints, subscription and cancellation for users, as well as product risk control, precision marketing, user retention and other related specific functions of the platform operator.

[0117] The business logic layer is also responsible for updating the relevant log data of this user care to the data collection layer to achieve a closed business loop.

[0118] In this embodiment, user portrait: based on big data technology, the user's specific information is abstracted into tags, and these tags are used to concretize the user's image, so that the user's attributes and behaviors can be linked with the expected services or data conversion, so as to provide users with targeted services.

[0119] User care: By focusing on user needs and interactive feedback, we provide users with personalized services and solutions to improve customer satisfaction and loyalty and promote business growth. User care is a basic way to standardize service quality, covering customer analysis, customer service, personalized marketing and other aspects.

[0120] Example 2

[0121] The present invention provides a user care system based on real-time user portraits, which executes the steps of the method in Example 1, including a data collection layer, a data processing layer, a data application layer and a business logic layer;

[0122] The data collection layer is used to collect and clean the raw data;

[0123] The data processing layer includes a computing module for real-time data processing and offline data processing of the cleaned data;

[0124] The data application layer aggregates, converts, and loads relevant data based on the calculation results of user behavior characteristics and user credit characteristics to realize user profiling and provide data services for the business logic layer.

[0125] The business logic layer automatically selects the corresponding user care strategy based on the data provided by the data application layer, realizing the user care function that integrates customer service and precise operation.

[0126] User care strategies include product consultation, product complaints, subscription and cancellation plans for users, as well as product risk control, precision marketing, and user retention plans by platform operators.

[0127] The business logic layer updates the relevant log data of this user care to the data collection layer to achieve a closed business loop.

[0128] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A user care method based on real-time user portrait, characterized in that: The specific steps include: S1. Data collection: collect user behavior logs and user attributes in real time through the log server cluster; S2. Clean the collected raw data; S3, data processing: processing the cleaned data; the processing process includes real-time data processing and offline data processing; S4, real-time storage and real-time calculation of user behavior data to obtain user behavior feature labels; S5. Offline storage and calculation of user attribute data to obtain user credit tags; S6. Build a user scoring model based on the user behavior feature label and the user credit feature label; and process the data to obtain a user portrait based on the calculation result of the user scoring model; S7. Select a variety of corresponding user care strategies for the user based on the obtained user portrait.

2. The user care method based on real-time user portrait according to claim 1 is characterized in that: In step S1, the user behavior logs and user attributes uploaded to the platform by the user through the product client are collected by the log server cluster in real time through the business interface, background service, real-time printing or external platform business interface transmission.

3. The user care method based on real-time user portrait according to claim 2 is characterized in that: The user behavior log in step S1 specifically includes: product use, ordering / payment, unsubscription consumption behavior, and consultation and complaint customer service behavior; The user attributes in S1 include the user's online time, call package level, and whether the user is in arrears.

4. The user care method based on real-time user portrait according to claim 2 is characterized in that: For user attribute data that is not time-sensitive, it is obtained by extracting files at a scheduled time to reduce the system load.

5. The user care method based on real-time user portrait according to claim 1 is characterized in that: The data cleaning steps in step S2 include data consistency check and deduplication, outlier processing, and normalization.

6. The user care method based on real-time user portrait according to claim 1 is characterized in that: Step S3 also includes the following steps: S31. Save all the user's ordering / payment behaviors in the last 24 hours and sort them by time. Each element in the list contains a timestamp (t n ) and an integer (a n ), records the startup data of the ordering / payment behavior at a certain moment in the past 24 hours; and is expressed as: <h2 style=";text-align:left;direction:ltr">[(t1, a1),(t2, a2),…,(t<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ,a<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> )] S32, cumulative order / payment data: indicates the cumulative number of times up to the last order / payment, expressed as Sum; S321, add a key hash process, and send the data with the same key to the same Reducer in MapReduce for processing, to ensure that the same user can fall into the same processing process; S322, as the user ordering / payment data collected by the log server in real time enters the message queue, the processing process will read this information in real time and modify the corresponding user's data; Get the current time t c ; Read historical data Read user history data from the local cache: [(t1, a1), (t2, a2), …, (t n , a n )], the cumulative value is SumA; if there is no local cache information, read the user data from the database; S33, cumulative value increases: SumA = SumA + 1; S34, historical data update; t c The user behavior at that moment is written into the historical ordering / payment behavior list, that is, the list is updated to: [(t1,a1),(t2,a2),…,(t n ,a n ),(tc,1)] If you encounter c =t n In the special case, it is updated to: <h2 style=";text-align:left;direction:ltr">[(t1, a1),(t2, a2),…,(t<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ,a<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> +1)] Because the calculation is within 24 hours, it is also necessary to delete the timestamp less than t in the list. c -24 hour element, and subtract the accumulated value corresponding to that moment 24 hours ago from SumA.

7. The user care method based on real-time user portrait according to claim 1 is characterized in that: The specific steps of step S6 are as follows: S61, after respectively performing real-time and offline calculations on the user behavior data and the user attribute data, obtaining the user's behavior feature label and credit feature label; S62, merging and normalizing the behavior feature labels and the credit feature labels to obtain a feature vector matrix F of multiple dimensions for a specific user, where each row represents a feature vector of one dimension; For a certain user, the feature vector matrix with n categories of business features is expressed as: S63, updating the typical user template matrix T according to the user behavior feature label and the credit feature label; the typical user template matrix is ​​obtained by offline calculation under preset conditions, and each row represents a typical user of a certain type of business feature; each row of the typical user template matrix corresponds to each other with respect to the feature vector matrix; S64, calculate the cosine similarity between the feature vector matrix and the typical user template matrix to obtain the user rating vector S: S=[s1,s2,…,s n ] The value of the element Sn in the vector S represents the degree of matching between a specific user and the nth type of typical users.

8. A user care system based on real-time user portrait, executing the method according to any one of claims 1 to 7, characterized in that: Includes data collection layer, data processing layer, data application layer and business logic layer; The data collection layer is used to collect and clean the raw data; The data processing layer includes a computing module for real-time data processing and offline data processing of the cleaned data; The data application layer aggregates, converts, and loads relevant data based on the calculation results of user behavior characteristics and user credit characteristics to realize user profiling and provide data services for the business logic layer. The business logic layer automatically selects the corresponding user care strategy based on the data provided by the data application layer, realizing the user care function that integrates customer service and precise operation.

9. The user care system based on real-time user portrait according to claim 8, characterized in that: User care strategies include product consultation, product complaints, subscription and cancellation plans for users, as well as product risk control, precision marketing, and user retention plans by platform operators.

10. The user care system based on real-time user portrait according to claim 8, characterized in that: The business logic layer updates the relevant log data of this user care to the data collection layer to achieve a closed business loop.

Images