Dynamic customer group division method and system

Abstract

The invention provides a customer group dynamic division method and system, and relates to the technical field of customer group management.The method comprises the steps that firstly, commodity related data are collected in real time through a multi-channel data collection module, commodity behavior analysis is conducted in combination with service requirements, features are extracted according to analysis results, and commodities are divided into multiple groups; and then, dynamically dividing customers by combining service requirements, generating initial customer group information, and optimizing and adjusting the customer group information according to the updated user portrait information. And finally, performing preference calculation according to the optimized customer group information, and generating a personalized push strategy. According to the method, through dynamic division and optimization of customer group information, unique requirements and behavior characteristics of each customer can be better met, and personalized services are provided.

Description

Technical Field

[0001] The present invention relates to the technical field of customer group management, and particularly to a method and system for dynamically dividing customer groups. Background Art

[0002] With the booming development of the e-commerce industry, the Internet has generated a large amount of user data, including data such as user browsing behavior, click behavior, and purchase behavior. A large amount of user demand and user behavior pattern information is contained in these data, which plays an important role in business decision-making, product optimization, and personalized recommendation. Achieving accurate division and timely dynamic adjustment of customer groups has become an important means for industries such as e-commerce, advertising, and finance to pursue efficient operation, control marketing costs, and improve user experience and satisfaction.

[0003] Traditional customer group division methods generally divide according to users' demographic information such as age, gender, income, etc., or according to users' behavior data such as browsing records and purchase records. However, such methods have some problems. For example, the accuracy of division by demographic information is limited by the detail of data collection, the real-time nature of updates, and the true impact of demographic factors on specific business decisions. The problem with behavior data is that due to the variability of user behavior data and its strong influence by factors such as time and environment, using a few fixed behavior characteristics for division often fails to accurately reflect users' actual needs and behavior habits, resulting in a deviation of the prediction of users' needs from reality.

[0004] With the development of big data technology, the use of machine learning means for customer group division has been widely studied. Such methods mine hidden patterns from a large amount of user behavior data through machine learning algorithms to make a more refined division of customer groups. However, such methods also have some problems. For example, on the one hand, machine learning algorithms require a large amount of computing resources and have high requirements for computing power; on the other hand, machine learning algorithms have a high dependence on the quality and integrity of the data set and the accuracy of feature selection. Data missing, incomplete, or inaccurate feature selection will have a negative impact on the results of the algorithm. Therefore, how to use enhanced customer dynamic data and machine learning algorithms to perform accurate and real-time customer group division is an urgent problem to be solved.

[0005] The method with the prior art publication number: CN114493686A mainly focuses on the process of generating push content. Although it uses a pre-constructed customer portrait model, it has the following disadvantages: At the static level: Its customer portrait model does not have the ability to self-update and adjust, and it is difficult to track and reflect changes in customer behavior in real time. This may lead to the push content not accurately meeting the latest needs and interests of customers, thereby reducing the marketing effect. Summary of the Invention

[0006] This application provides a method and system for dynamically dividing customer groups, aiming to solve the problem that traditional customer division methods usually rely on simple manual rules or static analysis of historical data, and to achieve precise dynamic division and optimization of customer groups to meet the personalized service needs of online e-commerce and other businesses, and improve customer satisfaction and business operation efficiency.

[0007] To solve the above problems, the present invention is implemented by the following technical solutions: The method includes:

[0008] Real-time collection of product-related data through a multi-channel data collection module to obtain a multi-source product data set;

[0009] Based on the multi-source product data set and combined with business requirement information, conduct product behavior analysis to generate behavior analysis data, and extract features according to the behavior analysis data to determine multiple product features;

[0010] According to the product features, use a clustering algorithm to divide products into multiple groups; through the clustering algorithm, products with similar click frequencies, click time distributions, attribute associations, etc. are grouped into the same group;

[0011] Based on the multi-dimensional user portrait information and combined with the business requirement information, dynamically divide the customer groups to generate multiple initial customer group information, and achieve dynamic analysis and personalized services;

[0012] Regularly update the multi-dimensional user portrait information according to the multi-source customer data set, and feedback the updated portrait information to the multiple initial customer group information for adjustment and optimization to generate multiple optimized customer group information, and optimize for the unique needs and behavior characteristics of each customer;

[0013] According to the multiple optimized customer group information, calculate preferences according to the business requirement information to generate multiple preference push strategies, and design personalized push strategies for different customer groups by analyzing customer behavior characteristics.

[0014] In one solution, for online data in the data collection, use API interfaces, web crawlers, and log collection to obtain user behavior data in real time;

[0015] For offline data, connect to the POS system, CRM system, and other internal databases, and use ETL tools to transfer data regularly or in real time;

[0016] Third-party data is regularly obtained and updated through API interfaces or data import tools.

[0017] In one solution, the multi-source product dataset combines business requirement information for product behavior analysis, calculates weighted scores for the customer's recent purchase time R, consumption frequency F, and consumption amount M based on the RFM model, and divides high-value, potential, and lost customer groups by combining the K-means clustering algorithm;

[0018] Use principal component analysis (PCA) to perform dimensionality reduction on the feature vectors, and use the Spearman correlation coefficient to screen features strongly related to the business objective to construct an extended feature set including the variance of the purchase cycle and the proportion of cross-category consumption.

[0019] In one solution, the portrait construction includes:

[0020] Based on the extracted multiple behavior features, use the clustering algorithm to divide the multi-source customer dataset to generate K non-overlapping customer clusters, maximizing the similarity of customer behavior features within the clusters and the difference between the clusters;

[0021] Map the customer clusters to a customer group data set with similar behavior features;

[0022] Based on the behavior features of each data set, integrate basic attributes, psychological features, and pain point information of needs to generate a multi-dimensional user portrait;

[0023] According to the demographic, behavior, psychological, and technical dimension information in the user portrait, formulate differentiated marketing strategies and service plans.

[0024] In one solution, the dynamic division includes:

[0025] Data integration and preparation, including multi-source customer data, behavior features, and their clustering results. Ensure the accurate attribution of each customer's data set and uniformly encode and format various features; in the S402 stage, integrate the basic demographic information of customers;

[0026] Basic information integration, each customer portrait contains basic demographic information, age, gender, geographical location, occupation, income level; calculate the average age, gender ratio, and geographical distribution density indicators of the customer group;

[0027] Behavior feature analysis and integration, deeply analyze and describe the behavior patterns of each customer group; purchase frequency F, purchase amount M, recent purchase time R, purchase channel preference C, product preference P;

[0028] Lifecycle stage determination, divide customers into different lifecycle stages, potential customers, first-time purchase customers, active customers, dormant customers, and loyal customers; the determination of the lifecycle stage is based on the dynamic changes of behavior features;

[0029] Personalized demand analysis, by analyzing the behavioral characteristics and life cycle stages of each customer group, identify their personalized demands; adopt the Apriori algorithm of association rule mining to discover the demands and preferences of customers in different scenarios;

[0030] Multi-dimensional user portrait construction, combining the above analysis results, construct the multi-dimensional user portraits of each customer group.

[0031] In one solution, the regular update includes:

[0032] Set the update cycle according to the behavioral patterns of the multiple cycle stages, perform interaction behavior analysis on the multi-source customer data set according to the update cycle, and generate interaction behavior parameters;

[0033] Perform reverse capture according to the interaction behavior parameters combined with the multiple behavioral characteristics to obtain behavior capture data;

[0034] Regularly update the multi-dimensional user portrait information according to the behavior capture data according to the update cycle to generate the updated portrait information.

[0035] In one solution, the generation of multiple preference push strategies includes:

[0036] Clarify and define business requirements, transform them into specific marketing goals and key performance indicators, and build an effective basis for push strategies;

[0037] Use correlation analysis to associate the business requirements with each dimension of the customer portrait to identify customer characteristics associated with the business goals;

[0038] According to the identified customer characteristics, calculate the preference scores for each customer group, which are calculated through a weighted scoring model to reflect their matching degree to the business requirements;

[0039] Rank the customer groups according to the calculated preference scores to determine the customer groups to be pushed preferentially;

[0040] According to the preference scores and priority rankings, design personalized push strategies for different customer groups. The push strategy includes customizing message content according to customer preferences and interests, selecting suitable push channels, optimizing the push time, and designing an interaction mechanism to enhance customer participation and interactivity; at the same time, determine the best combination of different strategy elements through an optimization function to achieve the maximization of expected business indicators;

[0041] Implement the designed push strategy on the premise of meeting constraints such as budget limitations, push frequency limitations, and compliance requirements;

[0042] After implementing the push strategy, the effectiveness of the push strategy is evaluated by monitoring key metrics and collecting customer feedback, thereby providing a basis for optimizing subsequent push strategies.

[0043] On the other hand, a customer group dynamic partitioning system for the described customer group dynamic partitioning method, the system includes:

[0044] A real-time data acquisition module for performing real-time data acquisition on product-related data through a multi-channel data acquisition module to obtain a multi-source customer data set;

[0045] A customer behavior analysis module for performing product behavior analysis based on the multi-source product data set combined with business requirement information, generating behavior analysis data, and performing feature extraction based on the behavior analysis data to determine multiple behavior features;

[0046] A portrait construction module for partitioning the multi-source customer data set according to the multiple behavior features, generating multiple data groups, and performing portrait construction based on the multiple data groups to obtain multi-dimensional user portrait information;

[0047] A dynamic partitioning module for dynamically partitioning the customer group according to the multi-dimensional user portrait information combined with the business requirement information to generate multiple initial customer group information;

[0048] An adjustment and optimization module for regularly updating the multi-dimensional user portrait information according to the multi-source customer data set, and feedbacking the updated portrait information to the multiple initial customer group information for adjustment and optimization to generate multiple optimized customer group information;

[0049] A preference calculation module for performing preference calculation according to the multiple optimized customer group information according to the business requirement information to generate multiple preference push strategies.

[0050] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0051] By collecting multi-source product and customer data in real time, deeply analyzing customer behavior and extracting features to partition customer groups, it can more accurately identify customer characteristics and needs, thereby improving the accuracy and effectiveness of push strategies. This method realizes the dynamic partitioning and adjustment of customer groups, can timely reflect and adapt to changes in market and user needs, and increases the possibility of business operation profitability.

[0052] Based on the updated portrait information, feedback of the optimization strategy is carried out, making the strategy push more targeted and real-time, further improving customer satisfaction and brand loyalty. By customizing push strategies, the marketing effect is improved, the marketing cost is reduced, and a higher return on investment is achieved.

[0053] The present invention can be widely applied to industries such as e-commerce, advertising, and finance, avoiding the problem in traditional methods that customer groups cannot be accurately divided due to demographic information and fixed behavioral characteristics, and improving the satisfaction of business requirements and industry competitiveness.

[0054] The collection, processing, and use of user personal information and related data involved in the technical solution provided in this application strictly comply with the provisions of relevant laws and regulations, and the authorization and consent of relevant users have been obtained. During the implementation process, all personal information involved is protected by encryption technology, and necessary security measures are taken to ensure data privacy and user rights and interests. The collection, storage, and use of all data follow the principles of transparency, legality, and compliance to maximize the protection of user personal privacy and information security. The above description is only an overview of the technical solution of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of this application more obvious and understandable, the following specifically gives the specific implementation manners of this application. Brief Description of the Drawings

[0055] Figure 1 It is a schematic flowchart of a method for dynamically dividing customer groups provided by an embodiment of this application.

[0056] Figure 2 It is a schematic structural diagram of a system for dynamically dividing customer groups provided by an embodiment of this application.

[0057] Figure 3 It is a flowchart for dividing a multi-source customer data set by multiple behavioral characteristics of this application.

[0058] Figure 4 It is the construction process of the user portrait of this application.

[0059] Figure 5 It is a flowchart for constructing a personalized portrait for each group of customers of this application.

[0060] Figure 6 It is a flowchart for generating multiple preference push strategies of this application.

[0061] Explanation of Reference Numerals: Real-time data collection module 10, customer behavior analysis module 20, portrait construction module 30, dynamic division module 40, adjustment and optimization module 50, preference calculation module 60. Detailed Description of the Preferred Embodiments

[0062] Embodiments of the present application provide a method and system for dynamically dividing customer groups, which solve the technical problem that traditional customer division methods are usually based on simple manual setting rules or static analysis of historical data. Such a static division method cannot reflect the changes in customer behavior in real time, resulting in the enterprise's inability to quickly respond when customer needs change.

[0063] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be specifically introduced below in conjunction with the accompanying drawings of the specification. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0064] Embodiment 1, as Figure 1 shown, embodiments of the present application provide a method for dynamically dividing customer groups, and the method includes:

[0065] S1. Real-time collect data related to commodities through a multi-channel data collection module to obtain a multi-source commodity data set.

[0066] For online data, technical means such as API interfaces, web crawlers, and log collection are used to obtain data such as the frequency and click time of commodities in real time; for offline data, by connecting to commodity management systems, supply chain systems, and other internal databases, ETL (Extract, Transform, Load) tools are used to regularly or real-time transmit commodity attribute data (such as category, price, inventory, promotion status, etc.); third-party data is regularly obtained and updated through API interfaces or data import tools. To ensure the efficiency and real-time nature of data collection, a distributed data collection architecture is usually adopted, and Apache Kafka is used for real-time data stream processing. The collected data is subjected to unified format conversion and standardization processing, and ETL tools are used to clean, transform, and load the data from various data sources into a centralized storage system. Apache NiFi is used for automated management of data streams to ensure the accurate transmission and storage of data. For real-time data, streaming processing technology is used for immediate integration; for batch data, batch processing integration is performed regularly. At the same time, a perfect data quality management mechanism must be established to ensure the integrity, accuracy, and timeliness of data. By setting up data verification rules and cleaning processes, duplicate, missing, or abnormal data can be eliminated, data verification tools are used for format checking and error detection, and machine learning algorithms are applied to automatically identify and correct outliers in the data.

[0067] S2. Based on the multi-source commodity data set, combine business requirement information to perform commodity behavior analysis, generate behavior analysis data, and extract features according to the behavior analysis data to determine multiple commodity features.

[0068] Clarify business requirements, which usually come from the product operation team or the supply chain team, such as increasing the sales volume of a certain type of product, optimizing inventory management, improving product exposure, etc. Combine the business requirement information with the multi-source product dataset for product behavior analysis, including product click trend analysis. By analyzing the change in click frequency of products at different time periods, identify product behavior patterns. For example, the click volume of a product surges during a specific time period, or there is a click peak during holidays. These data are used to evaluate the effectiveness of product display strategies and provide a basis for subsequent optimization. A / B testing can be used to compare the effects of different strategies.

[0069] The specific implementation process is as follows:

[0070] Clean and integrate the multi-source data collected from the S1 module to ensure data integrity and consistency. Data cleaning includes handling missing values, outliers, and duplicate data. For missing values, interpolation methods (such as mean filling, regression imputation) are used for filling; for outliers, statistical detection methods (such as Z-score) are used for identification and removal.

[0071] Product behavior analysis combines business requirements and uses statistical and machine learning methods to mine product behavior patterns. Suppose we want to analyze customers' purchase behavior. First, define relevant behavior metrics: the purchase frequency F, purchase amount M, and the time since the last purchase R of customers. The metrics can be calculated by the following formulas:

[0072]

[0073] R = current time - the time of the last purchase

[0074] By calculating these metrics, customers' purchase behavior can be quantified. In addition, clustering analysis can be used to group customers and identify different types of customer groups. Using the K-means algorithm, customers are divided into K clusters, with the goal of minimizing the squared error between samples within the clusters:

[0075]

[0076] where Ck represents the k-th cluster, μ k is the centroid of the k-th cluster, and x i is the feature vector of the customer. Through clustering analysis, customer groups with similar purchase behaviors can be discovered, providing a basis for precision marketing.

[0077] In the feature extraction stage, based on the behavior analysis data, statistical metrics and machine learning are used to extract features that can effectively represent customer behavior. The features include purchase frequency, purchase amount, purchase channel preference, product preference, activity response rate, etc. To further refine the features, principal component analysis (PCA) is adopted for dimensionality reduction, extracting the main components and reducing feature redundancy. PCA performs eigenvalue decomposition on the covariance matrix and selects the top k principal components. The formula is as follows:

[0078] X = WY

[0079] where X is the original data matrix, W is the principal component weight matrix, and Y is the data matrix after dimensionality reduction. Through PCA, the high-dimensional data is mapped to a low-dimensional space, retaining most of the important information.

[0080] S3. According to the commodity features, use the clustering algorithm to divide the commodities into multiple groups. Through the clustering algorithm, the commodities with similar click frequencies, click time distributions, attribute associations, etc. are grouped into the same group. As the commodity click data changes, the commodity group division results are updated regularly to ensure the real-time and accuracy of the division. The results of the commodity group division will be used as the basis for subsequent preference calculation and display strategy generation.

[0081] Customers are divided into several different groups according to multiple behavior features. For example, based on the consumption amount and purchase frequency, customers can be divided into groups such as high-value high-frequency customers, low-value low-frequency customers, high-frequency low-value customers, etc. The clustering algorithm, such as K-means, can be used to cluster the customer data according to multiple features, thus generating multiple data groups, and each data group represents a group of customers with similar behavior features.

[0082] As Figure 3 shown, the main goal of S3 is to reasonably divide the customer data set according to the behavior features, form multiple data groups, and build a multi-dimensional user portrait by analyzing the features of these data groups, providing a basis for precision marketing and personalized services. The entire implementation process includes customer data division, generation of data groups, construction of user portraits, and integration of multi-dimensional information. The specific steps are as follows:

[0083] S301. Customer data division: According to the multiple behavior features extracted in S2, divide the multi-source customer data set. This process uses the clustering algorithm to gather the customers with similar behavior features together, forming several non-overlapping data groups. The goal is to divide the customer data set X = {x1, x2,..., x n} into K clusters {C1, C2,..., C K}, maximizing the similarity of the behavioral characteristics of customers within the cluster and maximizing the differences between clusters. The core of the K-means algorithm is to minimize the Within-Cluster Sum of Squares (WCSS).

[0084]

[0085] Among them, μ k represents the centroid of the k-th cluster, and ||x i -μ k || 2 represents the square of the Euclidean distance between the customer x i and the cluster centroid μ k .

[0086] S302. Generate multiple data groups, and divide the customer dataset into multiple data groups {C1, C2,..., C K} through the clustering algorithm. Each data group represents a group of customers with similar behavioral characteristics. For example, on an e-commerce platform, customers can be divided into high-frequency low-amount purchase groups, high-frequency high-amount purchase groups, low-frequency high-amount purchase groups, etc. The generated data groups should have the following characteristics:

[0087] S303. User persona construction. After the customer data is divided, based on the behavioral characteristics of each data group, construct the corresponding User Persona. A user persona is a comprehensive description of a specific customer group in multiple dimensions, including the following aspects: Basic information: Demographic characteristics such as age, gender, geographical location, etc. Behavioral characteristics: including purchase frequency, purchase amount, recent purchase time, product preference, purchase channel preference, etc. Psychological characteristics: such as brand loyalty, price sensitivity, preference type (such as functional or emotional), etc. Needs and pain points: The problems and needs encountered by customers during the use of products or services.

[0088] As Figure 4 shown, the construction process of the user persona includes the following steps:

[0089] S3031. Feature summarization and statistical analysis: Summarize the behavioral characteristics within each data group, and calculate statistical indicators such as mean, standard deviation, and distribution. For example, for the purchase frequency F, purchase amount M, and recent purchase time R, the average value of each data group can be calculated to understand the overall behavioral characteristics of the group.

[0090] S3032. Multi-dimensional Feature Integration: Integrate multi-dimensional data such as basic information, behavioral features, and psychological features to form a comprehensive user portrait. The user portrait of a certain data group may include "aged 25 - 34, female, living in first-tier cities, high-frequency and low-amount purchases, preferring fashion products, price-sensitive, and paying attention to promotional activities".

[0091] S3033. Visualization and Description: Intuitively display the user portrait through charts, descriptive texts, etc., for easy understanding and application by business departments. Use a radar chart to show the distribution of different data groups in each behavioral feature, or summarize the main features and needs of each user group through detailed descriptions.

[0092] S304. Obtaining Multi-dimensional User Portrait Information. By constructing user portraits, the following multi-dimensional information can be obtained: Demographic dimension: such as age, gender, income level, educational background, etc. Behavioral dimension: including purchase behavior (frequency, amount, channel), browsing behavior (number of visited pages, stay time), interaction behavior (participating in activities, giving feedback), etc. Psychological dimension: such as brand preference, loyalty, price sensitivity, purchase motivation, etc. Technical dimension: the type of device used by the customer, operating system, access time period, etc.

[0093] This multi-dimensional information can help enterprises deeply understand the characteristics and needs of different customer groups, so as to formulate more accurate marketing strategies and personalized service plans. For example, for the high-frequency and low-amount purchase group, more small-scale promotional activities can be launched; while for the high-frequency and high-amount purchase group, exclusive customer services and high-end product recommendations can be provided.

[0094] Through the above steps and measures, multi-source customer data sets can be efficiently and accurately divided according to multiple behavioral features, generating multiple representative data groups, and constructing multi-dimensional user portrait information based on these data groups. These portraits can not only comprehensively reflect the characteristics and needs of different customer groups, but also provide a scientific basis for the enterprise's market strategy formulation, product development, and customer relationship management, thereby enhancing the enterprise's market competitiveness and customer satisfaction.

[0095] S4. According to these divided data groups, construct a personalized portrait for each group of customers. The portrait is a comprehensive description of customer characteristics, usually including basic information, behavioral features, life cycle stage, personalized needs, etc. Finally, after portrait construction, multi-dimensional user portrait information is obtained. The portrait information not only reflects the basic demographic information of the customer, but also covers multi-dimensional features such as the customer's behavior pattern, consumption habit, loyalty, interest and hobby, etc.

[0096] Dynamically divide the customer group according to the multi-dimensional user portrait information combined with the business requirement information to generate multiple initial customer group information.

[0097] Combine business requirement information with multi-dimensional user portrait information for dynamic segmentation. The dynamic segmentation is based on changes in customer behavior, activities, market trends, and business requirements, and continuously optimizes the segmentation of customer groups. Exemplarily, it is segmented from the behavioral level. For example, if the current business requirement is to increase the sales volume of a certain product, then customers who show relevant interests and purchase potential in the multi-dimensional user portrait will be segmented into the target customer group. Through dynamic segmentation, the customer group is segmented into different data groups, and preliminary customer groups are constructed according to the characteristics and needs of each group, generating multiple initial customer group information. These initial customer groups provide a basis for subsequent activities. Through targeted push strategies for different groups, different activity goals can be achieved.

[0098] As Figure 5 shown, based on the construction of multi-source customer data sets, customer behavior analysis and feature extraction, and customer data segmentation and preliminary user portrait construction completed in S1 to S3, further in-depth personalized portrait construction is carried out for each group of customers. The personalized portrait aims to comprehensively and meticulously describe the characteristics of customers, covering multiple dimensions such as basic information, behavioral characteristics, life cycle stages, and personalized needs. Through the implementation of S4, multi-dimensional user portrait information can be obtained. These information not only reflect the basic demographic characteristics of customers, but also cover various characteristics such as customer behavior patterns, consumption habits, loyalty, and hobbies. The specific implementation process is as follows:

[0099] S401. Data integration and preparation. First, integrate the data generated in the S1 to S3 stages, including multi-source customer data, behavioral characteristics, and their clustering results. Ensure the accurate attribution of each customer's data group, and uniformly encode and format various features (such as basic information, behavioral characteristics, life cycle stages, etc.). The goal of data integration is to provide a comprehensive and structured data foundation for subsequent portrait construction.

[0100] S402. Basic information integration. Each customer portrait first includes basic demographic information, such as age, gender, geographical location, occupation, income level, etc. Calculate the average age gender ratio geographical distribution density and other indicators:

[0101]

[0102] These indicators help identify the differences and commonalities in demographics among different customer groups.

[0103] S403. Behavioral Feature Analysis and Integration: Based on the behavioral features extracted in stages S2 and S3, conduct an in-depth analysis and description of the behavioral patterns of each customer group. Purchase frequency F, purchase amount M, time since last purchase R, purchase channel preference C, and product preference P. These features are integrated through methods such as statistical description (e.g., mean, standard deviation), distribution analysis, and correlation analysis. The purchase amount distribution of a certain group can be expressed as:

[0104]

[0105] where and are the mean and standard deviation of the purchase amount of the k-th customer group, respectively.

[0106] S404. Life Cycle Stage Determination: Classify customers into different life cycle stages, such as potential customers, first-time purchase customers, active customers, dormant customers, and loyal customers, etc. The determination of the life cycle stage can be based on the dynamic changes of behavioral features. For example:

[0107]

[0108] where T, F threshold and M threshold are preset time and behavior thresholds.

[0109] S405. Personalized Need Analysis: By analyzing the behavioral features and life cycle stages of each customer group, identify their personalized needs. Use the Apriori algorithm for association rule mining to discover the needs and preferences of customers in different situations. Use association rules to find that customers who purchase a specific product usually have a higher response rate to a certain type of promotion:

[0110]

[0111] Through these rules, the needs of a certain type of customer for specific products or services can be inferred.

[0112] S406. Multidimensional User Portrait Construction: Combine the above analysis results to construct a multidimensional user portrait for each customer group. Specifically include:

[0113] Feature Summary: Summarize the basic information, behavioral features, life cycle stages, and personalized needs to form a multi-dimensional description. A user portrait may include:

[0114] Basic Information: Age range 25 - 34 years old, female, living in first-tier cities, monthly income 5000 - 8000 yuan.

[0115] Behavioral characteristics: The average monthly purchase frequency is 5 times, the average purchase amount per time is 300 yuan, prefers to shop through the mobile terminal, and mainly purchases fashion products.

[0116] Lifecycle stage: Active customers.

[0117] Personalized needs: Responds positively to time-limited discounts and new product recommendations, and prefers personalized customization services.

[0118] Weight allocation and comprehensive scoring: Allocate weights to the characteristics of different dimensions, and calculate the comprehensive score to reflect the overall characteristics of each customer group. Set the weight of basic information as w1, the weight of behavioral characteristics as w2, the weight of lifecycle stage as w3, and the weight of personalized needs as w4, then the comprehensive score S k is expressed as:

[0119] S k = w1·BasicInfo k + w2·BehaviorFeatures k + w3·LifecycleStage k + w4·PersonalizedNeeds k

[0120] Through the comprehensive scoring, the characteristic performance of each customer group in different dimensions can be quantified, which is convenient for subsequent ranking and selection.

[0121] Through the construction of personalized user portraits in the S4 stage, the enterprise can obtain detailed and multi-dimensional customer portrait information. These portraits not only cover the basic demographic information of customers, but also deeply depict the behavioral patterns, consumption habits, loyalty, interests and hobbies of customers. With the help of mathematical models and data analysis methods, S4 effectively transforms multi-source data into user portraits with practical application value, helping the enterprise to achieve the goals of precision marketing, personalized service and customer relationship management, thereby improving the overall business efficiency and market competitiveness.

[0122] S5. Regularly update the multi-dimensional user portrait information according to the multi-source customer data set, and adjust and optimize the multiple initial customer group information based on the updated portrait information feedback, and generate multiple optimized customer group information.

[0123] Customers' behaviors and preferences are dynamically changing, so it is necessary to regularly update the multi-dimensional user portraits. For example, customers may change their purchase habits due to a certain promotion activity, or their preferences may change due to some new interests. Regularly collecting data from multiple channels and updating the portrait information of users can be carried out by setting an update cycle (such as daily, weekly, monthly) or triggering according to specific events (such as when a major behavioral change occurs to a customer).

[0124] Based on the updated portrait information, feedback and adjustment are made to multiple initial customer group information. For example, if there are significant changes in the behavior of certain customer groups, such as a decrease in purchase frequency, a decrease in activity, etc., these changes can be reflected in the preliminary customer group division through an automated feedback mechanism, including adjusting the push strategy for this group, or removing it from certain campaign targets and transferring it to other groups with higher activity, so as to further optimize the customer group division and generate multiple optimized customer group information. The optimized customer group information can provide more accurate targets for the next round of campaigns. This optimization process is a cyclic iterative process. As new data and feedback continuously enter, the customer group division will be continuously adjusted, forming a closed loop of continuous optimization and improved customer conversion.

[0125] S6. Calculate preferences according to the multiple optimized customer group information in accordance with the business requirement information to generate multiple preference push strategies.

[0126] The business requirement information has clear goals, such as increasing the sales volume of a certain type of product, increasing user activity, improving conversion rate, etc. These requirement information provide directions for preference calculation. For example, if the business requirement is to increase the sales volume of a specific product, then the preference calculation will focus on those customer groups with high purchase intent or historical purchases of related products. Based on the preference calculation results, specific push strategies are formulated for each optimized customer group. Each strategy can be customized according to the specific needs, preferences and behaviors of this group, generating multiple preference push strategies. The preference push strategies include product recommendation push, personalized discounts and coupons, timed push, cross-channel push, etc. Finally, through these customized push strategies, the conversion rate, user activity and brand loyalty of the campaign can be effectively improved.

[0127] such as Figure 6 As shown, based on the multi-dimensional user portrait information and the optimized customer group information generated in the previous S3 and S4 stages, calculate the customer preferences according to the business requirements, and accordingly generate multiple targeted push strategies. The main goal of S6 is to transform the accurate user portrait into actual marketing actions and improve customer engagement and conversion rate through personalized push strategies. The specific implementation process is as follows:

[0128] S601. Understand and define business requirements

[0129] Clarify the business goals and requirements, which are the basis for generating effective push strategies. The business requirements include increasing the sales volume of a certain type of product, increasing user activity, promoting new functions or services, improving customer loyalty, etc. After clarifying the business requirements, they need to be transformed into specific marketing goals and key performance indicators (KPIs), such as sales growth rate, click-through rate, conversion rate, etc.

[0130] S602. Mapping of requirements and user portrait

[0131] Correlate business requirements with each dimension of the user profile to identify which user characteristics are associated with business goals. For example, if the business goal is to promote a new product, it is necessary to identify the user groups that may be interested in the new product, and these users may have specific purchasing behavior characteristics, hobbies, or life cycle stages. Use correlation analysis to quantify the relationship between business requirements and user characteristics:

[0132]

[0133] where Y represents an indicator related to business requirements (such as sales), X i represents the i-th feature in the user profile, Cov is the covariance, and σ Y and are the standard deviations of Y and X i respectively.

[0134] S603, Preference Calculation and Scoring Model Construction

[0135] According to the mapping results, calculate the preference score for each customer group to reflect its fit with business requirements. The calculation of the preference score usually adopts a weighted scoring model, which assigns weights to relevant features in the user profile according to their importance, and comprehensively calculates the preference score P k :

[0136]

[0137] where w i is the weight of the i-th feature, x i,k is the score of the k-th customer group on the i-th feature dimension. The weight w i can be determined by expert scoring and statistical analysis.

[0138] S604, Customer Group Priority Ranking

[0139] According to the calculated preference scores, rank the optimized customer groups in order of priority to determine which customer groups should be pushed first. The ranking can be based on the level of preference scores, or other factors can be combined, such as the size of the customer group, potential value, and business urgency. The ranking formula is as follows:

[0140] Priority k = f(P k , V k , U k )

[0141] where P k is the preference score, V k is the potential value of the customer group, and U kFor the business urgency of the customer group, the function f is usually a weighted sum:

[0142] Priority k = αP k + βV k + γU k

[0143] The weight coefficients \((α, β, γ)\) are set according to business requirements.

[0144] S605. Generate personalized push strategies

[0145] Based on the preference scores and priority rankings, design specific push strategies for each customer group. The push strategies should include the following aspects:

[0146] Content customization: Customize personalized message content according to customer preferences and interests. For example, push new product information to customers interested in new products, and push discount offers to price-sensitive customers.

[0147] Push channel selection: Select push channels suitable for the customer group, such as email, SMS, mobile app push notifications, social media ads, etc. The channel selection is based on the customer's preference tendencies and past response data.

[0148] Push time optimization: Optimize the push time points according to the customer's activity time and behavior patterns to increase the message open rate and response rate. For example, push messages during the customer's most active period.

[0149] Interactive mechanism design: Design interactive push content, such as clickable coupons, feedback surveys, membership point rewards, etc., to enhance the customer's sense of participation and interactivity.

[0150] Determine the optimal combination of different strategy elements by optimizing the function to maximize the expected business metrics. For example, set the objective function L to maximize the conversion rate CR:

[0151]

[0152] The constraint conditions include budget limitations, push frequency limitations, compliance requirements, etc.

[0153] S606. Monitoring and feedback

[0154] After implementing the push strategy, it is necessary to monitor key metrics in real time (such as open rate, click-through rate, conversion rate, unsubscribe rate, etc.) and collect customer feedback. These data are used to evaluate the effectiveness of the push strategy and provide a basis for subsequent optimization. A / B testing can be used to compare the effectiveness of different strategies:

[0155]

[0156] Among them, p1 and p2 are the conversion rates of two groups of strategies respectively, and n1 and n2 are the sample sizes.

[0157] Through the preference calculation and push strategy generation in stage S6, the enterprise can transform multi-dimensional user portrait information into specific and personalized marketing actions. This process is not only based on accurate user data and optimized customer group information, but also combines the accurate matching of business requirements to ensure the effectiveness and efficiency of the push strategy. With the help of mathematical models and automated tools, S6 realizes the seamless connection from data-driven user understanding to actual marketing execution, significantly improving the accuracy of marketing activities and customer conversion rate, thus helping the enterprise achieve higher business growth and market competitiveness.

[0158] Embodiment 2, based on the same inventive concept as a customer group dynamic partitioning method in the foregoing embodiment, as Figure 2 shown, the embodiment of the present application provides a customer group dynamic partitioning system, and the system includes:

[0159] A real-time data acquisition module 10, configured to perform real-time data acquisition on product-related data through a multi-channel data acquisition module to obtain a multi-source customer data set; a customer behavior analysis module 20, configured to perform product behavior analysis based on the multi-source product data set in combination with business requirement information, generate behavior analysis data, and perform feature extraction according to the behavior analysis data to determine a plurality of behavior features; a portrait construction module 30, configured to partition the multi-source customer data set according to the plurality of behavior features to generate a plurality of data groups, and perform portrait construction according to the plurality of data groups to obtain multi-dimensional user portrait information; a dynamic partitioning module 40, configured to perform dynamic partitioning on the customer group according to the multi-dimensional user portrait information in combination with the business requirement information to generate a plurality of initial customer group information; an adjustment and optimization module 50, configured to periodically update the multi-dimensional user portrait information according to the multi-source customer data set, and feedback the updated portrait information to the plurality of initial customer group information for adjustment and optimization to generate a plurality of optimized customer group information; a preference calculation module 60, configured to perform preference calculation according to the plurality of optimized customer group information according to the business requirement information to generate a plurality of preference push strategies.

[0160] Furthermore, the system further includes a plurality of behavior feature determination modules to perform the following operation steps:

[0161] Retrieve the target task, traverse the target task for tracking and analysis, and extract the business requirement information according to the task tracking result; perform behavioral trend analysis on the multi-source customer dataset according to the time series, determine multiple behavioral activity levels, and perform correlation analysis based on the multiple behavioral activity levels in combination with the business requirement information to generate multiple correlation coefficients; perform behavioral cycle analysis on the multi-source customer dataset based on the multiple correlation coefficients to generate the behavioral analysis data, where the behavioral analysis data includes multiple periodic stage behavioral patterns; perform extreme value analysis on the multiple periodic stage behavioral patterns according to the multiple behavioral activity levels, and perform feature identification on the multiple periodic stage behavioral patterns according to multiple behavioral activity extremes to determine the multiple behavioral characteristics.

[0162] Furthermore, the system further includes a multi-dimensional user portrait information acquisition module to perform the following operation steps:

[0163] Calculate the similarity of the multiple behavioral characteristics based on the multiple behavioral activity levels to generate multiple similarity coefficients; perform hierarchical clustering on the multiple behavioral characteristics according to the multiple similarity coefficients to determine multiple feature clustering centers; perform multi-dimensional partitioning on the multi-source customer dataset according to the multiple feature clustering centers to generate the multiple data groups; traverse the multiple data groups for group portrait analysis to generate multiple group behavioral characteristics, where the multiple group behavioral characteristics have a corresponding relationship with the multiple data groups; describe the multiple data groups based on the multiple group behavioral characteristics to generate multiple group labels, and perform multi-dimensional evaluation portraits on the multiple data groups according to the multiple group labels to obtain the multi-dimensional user portrait information.

[0164] Furthermore, the system further includes an initial customer group information generation module to perform the following operation steps:

[0165] Perform impact analysis on the behavioral analysis data based on the business requirement information to obtain multiple business impact indicators; perform weight analysis on the multi-dimensional user portrait information according to the multiple group labels in combination with the multiple business impact indicators to generate multiple weight coefficients; sort the multi-dimensional user portrait information in descending order according to the multiple weight coefficients to generate a portrait sequence; perform dynamic behavior prediction on the customer group according to the portrait sequence in combination with the multiple behavioral characteristics to generate multiple dynamic behavior prediction data; perform dynamic partitioning on the customer group according to the multiple dynamic behavior prediction data in combination with the multiple periodic stage behavioral patterns to generate the multiple initial customer group information.

[0166] Furthermore, the system further includes a dynamic behavior prediction data determination module to perform the following operation steps:

[0167] Input the behavioral analysis data into a long short-term memory network according to the image sequence for behavioral prediction, and draw a behavioral prediction trend chart; dynamically select according to the multiple behavioral characteristics based on the behavioral prediction trend chart to generate multiple behavioral prediction nodes; perform multi-task learning based on the multiple behavioral prediction nodes to generate multiple learning effects, and conduct a prediction stability assessment based on the multiple learning effects to generate a data stability score; traverse and screen the behavioral prediction trend chart according to the data stability score to determine the multiple dynamic behavioral prediction data.

[0168] Furthermore, the system further includes a plurality of initial customer group information generation modules to perform the following operation steps:

[0169] Set a behavioral fluctuation interval value based on the multiple behavioral activity extreme values; determine whether the multiple dynamic behavioral prediction data is within the behavioral fluctuation interval value; when the multiple dynamic behavioral prediction data is not within the behavioral fluctuation interval value, generate a dynamic update instruction, and update the multiple dynamic behavioral prediction data through the dynamic update instruction to generate multiple dynamically updated behavioral prediction data; perform an association match between the multiple periodic stage behavioral patterns and the multiple dynamically updated behavioral prediction data to generate multiple stage behavioral prediction data groups; divide the customer group based on the multiple stage behavioral prediction data groups to generate the multiple initial customer group information.

[0170] Furthermore, the system further includes an updated portrait information generation module to perform the following operation steps:

[0171] Set an update period according to the multiple periodic stage behavioral patterns, perform an interactive behavior analysis on the multi-source customer data set according to the update period to generate interactive behavior parameters; perform a reverse capture according to the interactive behavior parameters in combination with the multiple behavioral characteristics to obtain behavior capture data; regularly update the multi-dimensional user portrait information according to the behavior capture data according to the update period to generate the updated portrait information.

[0172] Through the foregoing detailed description of a method for dynamically dividing customer groups in this specification, those skilled in the art can clearly know a system for dynamically dividing customer groups in this embodiment. Since it corresponds to the method disclosed in the embodiment, it is described relatively simply. For related parts, refer to the description in the method section.

[0173] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for dynamically dividing customer groups, characterized in that, The method includes: Real-time collecting commodity-related data through a multi-channel data collection module to obtain a multi-source commodity data set; Based on the multi-source commodity data set and combined with business requirement information, conduct commodity behavior analysis to generate behavior analysis data, extract features according to the behavior analysis data, and determine multiple commodity features; According to the commodity features, use a clustering algorithm to divide commodities into multiple groups; through the clustering algorithm, group commodities with similar click frequencies, click time distributions, attribute associations, etc. into the same group; Dynamically divide the customer groups according to the multi-dimensional user portrait information and combined with the business requirement information to generate multiple initial customer group information, realizing dynamic analysis and personalized services; Regularly update the multi-dimensional user portrait information according to the multi-source customer data set, feedback the updated portrait information to the multiple initial customer group information for adjustment and optimization, generate multiple optimized customer group information, and optimize according to the unique needs and behavior characteristics of each customer; Calculate preference push strategies according to the multiple optimized customer group information in accordance with the business requirement information, and generate multiple preference push strategies. By analyzing customer behavior characteristics, design personalized push strategies for different customer groups.

2. The method for dynamically dividing customer groups according to claim 1, wherein For online data in the data collection, use API interfaces, web crawlers, and log collection to obtain user behavior data in real time; For offline data, connect to the POS system, CRM system, and other internal databases, and use ETL tools to transfer data regularly or in real time; Third-party data is regularly obtained and updated through API interfaces or data import tools.

3. The method for dynamically dividing customer groups according to claim 1, characterized in that, The multi-source commodity data set is combined with business requirement information to conduct commodity behavior analysis, calculate the weighted scores of the customer's recent purchase time R, consumption frequency F, and consumption amount M based on the RFM model, and divide high-value, potential, and lost customer groups by combining the K-means clustering algorithm; Use principal component analysis (PCA) to perform dimensionality reduction on the feature vectors, and use the Spearman correlation coefficient to screen features related to business goals to construct an extended feature set including the variance of the purchase cycle and the proportion of cross-category consumption.

4. The customer group dynamic division method according to claim 1, wherein, The portrait construction includes: Based on the extracted multiple behavior features, use a clustering algorithm to divide the multi-source customer data set to generate K non-overlapping customer clusters, maximizing the similarity of customer behavior features within the clusters and maximizing the differences between the clusters; Map the customer clusters to a customer group data set with similar behavior features; Based on the behavior features of each data set, integrate basic attributes, psychological features, and demand pain point information to generate a multi-dimensional user portrait; According to the demographic, behavioral, psychological, and technical dimension information in the user portrait, formulate differentiated marketing strategies and service plans.

5. The method for dynamically dividing customer groups according to claim 1, wherein, The dynamic division includes: Data integration and preparation, including multi-source customer data, behavior features, and their clustering results; ensure the accurate attribution of each customer's data set, and uniformly encode and format various features; in the S402 stage, integrate the customer's basic demographic information; Basic information integration. Each customer profile includes basic demographic information such as age, gender, geographical location, occupation, and income level. Calculate the average age, gender ratio, and geographical distribution density indicators of the customer group. Behavioral feature analysis and integration. Conduct in-depth analysis and description of the behavioral patterns of each customer group. Purchase frequency F, purchase amount M, recent purchase time R, purchase channel preference C, and product preference P. Determination of the lifecycle stage. Classify customers into different lifecycle stages: potential customers, first-time buyers, active customers, dormant customers, and loyal customers. The determination of the lifecycle stage is based on the dynamic changes in behavioral characteristics. Analysis of personalized needs. By analyzing the behavioral characteristics and lifecycle stages of each customer group, identify their personalized needs. Use the Apriori algorithm of association rule mining to discover the needs and preferences of customers in different scenarios. Construction of multi-dimensional user profiles. Combine the above analysis results to construct multi-dimensional user profiles for each customer group.

6. The method for dynamically dividing customer groups according to claim 1, characterized in that, The regular update includes: Set the update cycle according to the behavioral patterns in multiple cycle stages. Conduct interactive behavior analysis on the multi-source customer dataset according to the update cycle to generate interactive behavior parameters. Perform reverse capture according to the interactive behavior parameters in combination with the multiple behavioral characteristics to obtain behavior capture data. Regularly update the multi-dimensional user profile information according to the behavior capture data according to the update cycle to generate the updated profile information.

7. The customer group dynamic partitioning method according to claim 1, wherein The generation of multiple preference push strategies includes: Clarify and define business requirements, transform them into specific marketing goals and key performance indicators, and build a foundation for effective push strategies. Use correlation analysis to associate the business requirements with each dimension of the customer profile and identify customer characteristics related to the business goals. According to the identified customer characteristics, calculate a preference score for each customer group. The preference score is calculated through a weighted scoring model and is used to reflect its matching degree with the business requirements. Rank the customer groups according to the calculated preference scores to determine the customer groups to be pushed first. According to the preference scores and priority rankings, design personalized push strategies for different customer groups. The push strategy includes customizing message content according to customer preferences and interests, selecting suitable push channels, optimizing push times, and designing an interactive mechanism to enhance customer participation and interactivity. At the same time, use an optimization function to determine the best combination of different strategy elements to achieve the maximization of expected business indicators. Implement the designed push strategy on the premise of meeting budget constraints, push frequency constraints, and compliance requirements. After implementing the push strategy, conduct an evaluation of the effectiveness of the push strategy by monitoring key indicators and collecting customer feedback, so as to provide a basis for the optimization of subsequent push strategies.

8. A customer group dynamic division system, characterized in that, For implementing a customer group dynamic division method according to any one of claims 1-7, the system includes: A real-time data collection module for real-time data collection of commodity-related data through a multi-channel data collection module to obtain a multi-source customer dataset. A customer behavior analysis module, which is used to perform commodity behavior analysis based on the multi-source commodity data set combined with business requirement information, generate behavior analysis data, extract features according to the behavior analysis data, and determine multiple behavior features; A portrait construction module, which is used to divide the multi-source customer data set according to the multiple behavior features, generate multiple data groups, and construct a portrait according to the multiple data groups to obtain multi-dimensional user portrait information; A dynamic division module, which is used to dynamically divide the customer group according to the multi-dimensional user portrait information combined with the business requirement information to generate multiple initial customer group information; An adjustment and optimization module, which is used to regularly update the multi-dimensional user portrait information according to the multi-source customer data set, and feedback the updated portrait information to the multiple initial customer group information for adjustment and optimization to generate multiple optimized customer group information; A preference calculation module, which is used to calculate preferences according to the multiple optimized customer group information according to the business requirement information to generate multiple preference push strategies.

Images