An e-commerce platform member life cycle value stratification and differential marketing method and system
By constructing member feature vectors and dynamic clustering, and combining Euclidean distance and gradient descent algorithms to optimize weights, the problem of dynamic changes in member value in existing technologies is solved, enabling precise segmentation and differentiated marketing of e-commerce platform members.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BANGLIDE TECHNOLOGY (SHANGHAI) CO LTD
- Filing Date
- 2026-03-04
- Publication Date
- 2026-06-09
AI Technical Summary
The existing membership management system relies on a static RFM statistical model, which cannot dynamically capture changes in user value. This makes it difficult to distinguish between members in the value growth phase and those in the decline phase, and the segmentation logic remains fixed and cannot be adaptively adjusted based on marketing feedback.
By acquiring raw transaction data to construct member feature vectors, using the K-Means++ algorithm for clustering, dynamically monitoring member behavior shifts, and combining Euclidean distance and gradient descent algorithms to optimize weights, dynamic stratification of member value and differentiated marketing can be achieved.
It enables precise and dynamic monitoring and segmentation of member value, improves the accuracy of identifying high-value members, provides self-evolution capabilities, and ensures that the timing and effectiveness of marketing interventions are superior to traditional static models.
Smart Images
Figure CN122175638A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, and in particular to a method and system for segmenting and differentiating marketing based on the lifetime value of members on an e-commerce platform. Background Technology
[0002] In the fields of e-commerce and big data processing, existing membership management systems mainly rely on static RFM statistical models to define user value, which is often regarded as a common and conventional technical approach.
[0003] However, in real-world applications, when market competition intensifies or user behavior fluctuates slightly, this approach, which focuses solely on historical transaction snapshots, reveals significant lag and limitations. Existing technologies typically treat technical problems in isolation, failing to consider them within a dynamically changing lifecycle context. This results in systems that can only identify established facts at the "what" level, but cannot reach the evolutionary logic at the "how" level.
[0004] Traditional models, lacking the ability to analyze dynamic characteristics such as user activity slopes, struggle to differentiate between members of the same spending level in their value growth and decline phases. Furthermore, their segmentation logic is often fixed and cannot be adaptively adjusted based on marketing feedback. This lack of unique optimization or innovative combinations for specific problems makes existing comparative documents highly susceptible to "teaching" or "inspiration" of conventional solutions, thus limiting the creative scope for patent protection. Therefore, there is an urgent need for a method and system for segmenting and differentiating marketing based on the member lifecycle value of e-commerce platforms. Summary of the Invention
[0005] A method for segmenting and differentiating marketing based on the lifetime value of members on an e-commerce platform, the method comprising the following steps: Obtain raw transaction data and preprocess the raw transaction data to construct a member feature vector, which includes the most recent consumption time, consumption frequency, consumption amount, and activity slope. The K-Means++ algorithm is used to perform clustering calculations on the member feature vectors to divide all members into multiple value-stratified clusters, and the cluster center corresponding to each value-stratified cluster is determined. Based on the cluster center of each value stratification cluster, associate a corresponding marketing script with each value stratification cluster; Periodically acquire real-time behavioral data of the member to be evaluated and update the corresponding real-time feature vector, and calculate the offset between the real-time feature vector and the cluster center of the value hierarchy cluster to which the member to be evaluated currently belongs; When the offset exceeds a preset offset threshold, the member to be evaluated is dynamically assigned to a target value stratification cluster based on the offset, and the marketing script associated with the target value stratification cluster is invoked.
[0006] Furthermore, the step of constructing the activity slope includes: determining a preset sliding time window; Calculate the sequence of member behavior frequencies within the sliding time window; calculate the first derivative of the sequence to obtain the behavior change rate, and determine the behavior change rate as the activity slope.
[0007] Furthermore, the step of using the K-Means++ algorithm to perform clustering calculations on the member feature vectors includes: randomly selecting a vector from the member feature vectors as the first cluster center; Calculate the shortest distance between each of the remaining member feature vectors and the currently selected cluster center; The next cluster center is selected based on the squared proportional probability of the shortest distance, until the number of selected cluster centers reaches the preset number K. Iterative clustering is performed using the selected cluster number as the initial center point to output the value-stratified cluster.
[0008] Further, the step of calculating the offset between the real-time feature vector and the cluster center of the value hierarchy cluster to which the member to be evaluated currently belongs includes: Extract the components of each dimension of the real-time feature vector and the components of each dimension of the cluster center; Calculate the Euclidean distance between the real-time feature vector and the cluster center in multidimensional space, and determine the Euclidean distance as the offset.
[0009] Furthermore, the method further includes: constructing a state transition matrix, the state transition matrix containing the conditional probabilities of a member transitioning between different value hierarchical clusters; Based on the state transition matrix, calculate the instantaneous transfer probability of the member to be evaluated from the current value stratification cluster to the preset churn risk value stratification cluster; When the instant transfer probability exceeds a preset probability threshold ranging from 0.7 to 1.0, an early warning marketing script is triggered for the member to be evaluated.
[0010] Furthermore, the method also includes: real-time statistical analysis of changes in member repurchase rate after the marketing script is executed; Based on the change in the member repurchase rate, the weight allocation matrix of each dimension in the member feature vector is corrected using the gradient descent algorithm; The corrected weight allocation matrix is fed back into the clustering calculation step to update the value hierarchical cluster.
[0011] Furthermore, the step of invoking the marketing script associated with the target value hierarchy cluster includes: The marketing rule base is retrieved based on the identifier of the target value hierarchical cluster; the corresponding coupon issuance instruction or push copy instruction in the marketing rule base is matched; The instructions are sent to downstream business systems via an application programming interface (API).
[0012] A customer lifetime value stratification and differentiated marketing system for e-commerce platforms, the system comprising: The data preprocessing module is configured to acquire raw transaction data and preprocess the raw transaction data to construct a member feature vector, which includes the most recent consumption time, consumption frequency, consumption amount, and activity slope. The clustering engine module is configured to use the K-Means++ algorithm to perform clustering calculations on the member feature vectors, so as to divide all members into multiple value-stratified clusters and determine the cluster center corresponding to each value-stratified cluster. The rule configuration module is configured to associate a corresponding marketing script with each value stratification cluster based on the cluster center of each value stratification cluster; The dynamic monitoring module is configured to periodically acquire real-time behavioral data of the member to be evaluated and update the corresponding real-time feature vector, and calculate the offset between the real-time feature vector and the cluster center of the value hierarchy cluster to which the member to be evaluated currently belongs. The execution module is configured to dynamically classify the member to be evaluated into a target value stratification cluster according to the offset when the offset exceeds a preset offset threshold, and call the marketing script associated with the target value stratification cluster.
[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: By precisely defining essential technical features and constructing a tiered protection hierarchy, this solution represents a shift from general-purpose technologies to unique optimizations tailored to specific marketing problems. By introducing activity slopes, the system can capture dynamic trends in member value at a micro-level, providing technical details that are more difficult to "teach" using existing technologies. Secondly, this solution employs a dynamic offset monitoring mechanism based on Euclidean distance, enabling the discovery of new, non-obvious, groundbreaking application directions. It achieves precise reclassification at the early stages of member value shifts, with intervention timing significantly superior to traditional static models. Furthermore, by using marketing effectiveness as feedback gain and employing gradient descent algorithms to correct weights, a complete, non-simple, interconnected technical solution is constructed, giving the system self-evolutionary capabilities. This elevates the system from a "what" level to a concrete execution level of "how," greatly improving the accuracy of high-value member identification. Moreover, by minimizing the scope of the main claim and enriching dependent claims, the solution ensures strong inventiveness and a stable prospect for grant. Attached Figure Description
[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.
[0015] Figure 1 A flowchart illustrating the overall logic of a method for segmenting and differentiating marketing based on the lifetime value of members on an e-commerce platform, as provided in this embodiment of the invention. Figure 2 A schematic diagram illustrating the composition structure of an e-commerce platform member lifecycle value stratification and differentiated marketing system provided in this embodiment of the invention; Figure 3 The schematic diagram of the dynamic offset triggering principle provided in the embodiment of the present invention shows the physical trajectory process of the real-time feature vector deviating from the cluster center and crossing the threshold boundary. Detailed Implementation
[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0017] This invention provides a method and system for segmenting and differentiating marketing based on the lifetime value of members on e-commerce platforms.
[0018] like Figure 1-3 A method for segmenting and differentiating marketing based on the lifetime value of e-commerce platform members, with the following specific steps: The first step involves acquiring raw transaction data and preprocessing it to construct member feature vectors. In this step, the system's backend server extracts the historical consumption logs of all members from the e-commerce platform's distributed database via a pre-defined data interface. This logs include raw fields such as order number, payment time, settlement amount, and unique user identifier. The preprocessing specifically includes denoising duplicate order data, outlier removal for abnormally large orders, and dimension conversion of data with different dimensions using a normalization algorithm.
[0019] Subsequently, based on the preprocessed data, the system calculates each member's most recent consumption time, consumption frequency, consumption amount, and activity slope, and encapsulates these into a multi-dimensional member feature vector. These feature vectors serve as the foundational data for subsequent algorithms, fully representing a member's historical value contribution and recent activity trend on the platform.
[0020] In a preferred embodiment, the activity slope is constructed by determining a preset sliding time window, calculating the sequence of member behavior frequencies within the window, and calculating the first derivative of the sequence to obtain the rate of change of behavior, thereby capturing the potential evolution of member value through dynamic trends.
[0021] Next, the system performs clustering calculations on the member feature vectors using the K-Means++ algorithm to divide all members into multiple value-stratified clusters and determine the cluster center corresponding to each value-stratified cluster. During the operation of this algorithm, in order to overcome the shortcomings of traditional clustering algorithms that are sensitive to the initial point, the system first randomly selects a vector from all member feature vectors as the first cluster center. Then, it calculates the shortest distance between each remaining vector and the currently selected cluster center, and selects subsequent cluster centers according to the squared probability of the shortest distance, until the preset number of clusters K is met.
[0022] After determining the initial center point, the system performs iterative convergence calculations in the multi-dimensional feature space, aggregating member feature vectors that are close in distance into specific value-stratified clusters. Each value-stratified cluster has a cluster center representing the average feature level of that group. The system can identify micro-level technical issues from macro-level problems, accurately aggregating members with similar consumption habits and lifecycle characteristics, thereby providing data support for subsequent targeted optimization or innovative combinations.
[0023] Furthermore, after completing the member segmentation, the system executes the step of associating a corresponding marketing script with each value segmentation cluster based on the cluster center of each value segmentation cluster. This is used to transform the algorithm output after clustering into business-executable logical instructions. The system analyzes the characteristic dimension components of each cluster center. For example, cluster centers with recent consumption in the distant past and negative activity slopes are automatically identified as "churn risk groups"; cluster centers with high consumption frequency and large amounts are identified as "core high-value groups".
[0024] The system pre-stores multiple sets of marketing scripts in its rules engine. Each set of scripts is optimized for a specific level of problem. By establishing a mapping relationship between value stratification cluster identifiers and marketing script identifiers, the system can achieve a direct correspondence between technical solutions and beneficial effects. In actual deployment, the marketing scripts include specific algorithm parameters, push priorities, and benefit allocation quotas, ensuring the integrity and protection levels of the technical solution.
[0025] Based on the above embodiments, the system performs the following steps: periodically acquiring real-time behavioral data of the member to be evaluated and updating the corresponding real-time feature vector, and calculating the offset between the real-time feature vector and the cluster center of the value hierarchy cluster to which the member to be evaluated currently belongs.
[0026] The dynamic monitoring module collects the latest clicks, add-to-cart, and payment data of the members to be evaluated at a preset time frequency (e.g., hourly or daily), and overlays it onto the original feature vector to generate a real-time feature vector. The process of calculating the offset is specifically manifested in extracting the various dimensional components of the real-time feature vector and the various dimensional components of the cluster center, and using the multidimensional Euclidean distance formula to calculate the absolute distance between the two in space. The real-time offset monitoring mechanism can promptly detect the deviation of member behavior from its original level, moving from "what it is" to a more specific level of "how it is", solving the problem that traditional static hierarchical models cannot respond to fluctuations in member value in a timely manner.
[0027] Finally, when the offset exceeds a preset offset threshold, the system dynamically assigns the member to be evaluated to a target value stratification cluster based on the offset, and calls the marketing script associated with the target value stratification cluster. When the real-time feature vector shifts in the feature space, causing the offset to exceed the set boundary standard deviation, the execution module determines that the member has undergone a stratification jump, calculates the Euclidean distance between the real-time feature vector and the cluster centers of the remaining value stratification clusters, and selects the cluster with the smallest distance value as the target value stratification cluster.
[0028] Subsequently, the system retrieves marketing rule base in real time through application programming interface, matches corresponding coupon issuance instructions or copywriting push instructions, and sends them to downstream business systems for execution. In addition, to enhance the creativity of the solution, the system constructs a state transition matrix based on the migration frequency of members between different value stratification clusters in historical time series data to calculate the instantaneous transition probability, and triggers an alert when the probability exceeds a preset range (e.g., 0.7 to 1.0).
[0029] Meanwhile, the system uses the change in repurchase rate after the script is executed as the gain target and the gradient descent algorithm to minimize the deviation between the predicted stratification and the actual marketing feedback, thereby correcting the weight allocation matrix and forming a complete related technical solution and a self-evolving closed-loop process.
[0030] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for segmenting and differentiating marketing based on the lifetime value of members on an e-commerce platform, characterized in that, The method includes the following steps: Obtain raw transaction data and preprocess the raw transaction data to construct a member feature vector, which includes the most recent consumption time, consumption frequency, consumption amount, and activity slope. The K-Means++ algorithm is used to perform clustering calculations on the member feature vectors to divide all members into multiple value-stratified clusters, and the cluster center corresponding to each value-stratified cluster is determined. Based on the cluster center of each value stratification cluster, associate a corresponding marketing script with each value stratification cluster; Periodically acquire real-time behavioral data of the member to be evaluated and update the corresponding real-time feature vector, and calculate the offset between the real-time feature vector and the cluster center of the value hierarchy cluster to which the member to be evaluated currently belongs; When the offset exceeds a preset offset threshold, the member to be evaluated is dynamically assigned to a target value stratification cluster based on the offset, and the marketing script associated with the target value stratification cluster is invoked.
2. The method for segmenting and differentiating marketing based on the lifetime value of e-commerce platform members according to claim 1, characterized in that, The steps for constructing the activity slope include: determining a preset sliding time window; Calculate the sequence of member behavior frequencies within the sliding time window; calculate the first derivative of the sequence to obtain the behavior change rate, and determine the behavior change rate as the activity slope.
3. The method for segmenting and differentiating marketing based on the lifetime value of e-commerce platform members according to claim 1, characterized in that, The step of using the K-Means++ algorithm to perform clustering calculation on the member feature vector includes: randomly selecting a vector from the member feature vector as the first cluster center; Calculate the shortest distance between each of the remaining member feature vectors and the currently selected cluster center; The next cluster center is selected based on the squared proportional probability of the shortest distance, until the number of selected cluster centers reaches the preset number K. Iterative clustering is performed using the selected cluster number as the initial center point to output the value-stratified cluster.
4. The method for segmenting and differentiating marketing based on the lifetime value of e-commerce platform members according to claim 1, characterized in that, The step of calculating the offset between the real-time feature vector and the cluster center of the value stratification cluster to which the member to be evaluated currently belongs includes: Extract the components of each dimension of the real-time feature vector and the components of each dimension of the cluster center; Calculate the Euclidean distance between the real-time feature vector and the cluster center in multidimensional space, and determine the Euclidean distance as the offset.
5. The method for segmenting and differentiating marketing based on the lifetime value of e-commerce platform members according to claim 1, characterized in that, The method further includes: constructing a state transition matrix, the state transition matrix containing the conditional probabilities of a member transitioning between different value hierarchical clusters; Based on the state transition matrix, calculate the instantaneous transfer probability of the member to be evaluated from the current value stratification cluster to the preset churn risk value stratification cluster; When the instant transfer probability exceeds a preset probability threshold ranging from 0.7 to 1.0, a marketing script is triggered for the member to be evaluated.
6. The method for segmenting and differentiating marketing based on the lifetime value of e-commerce platform members according to claim 1, characterized in that, The method also includes: real-time statistical analysis of changes in member repurchase rate after executing the marketing script; Based on the change in the member repurchase rate, the weight allocation matrix of each dimension in the member feature vector is corrected using the gradient descent algorithm; The corrected weight allocation matrix is fed back into the clustering calculation step to update the value hierarchical cluster.
7. The method for segmenting and differentiating marketing based on the lifetime value of e-commerce platform members according to claim 1, characterized in that, The step of invoking the marketing script associated with the target value hierarchy cluster includes: The marketing rule base is retrieved based on the identifier of the target value hierarchical cluster; the corresponding coupon issuance instruction or push copy instruction in the marketing rule base is matched; The instructions are sent to downstream business systems via an application programming interface (API).
8. A segmented and differentiated marketing system for the lifetime value of members on an e-commerce platform, characterized in that, The system includes: The data preprocessing module is configured to perform the steps described in claim 1: acquiring raw transaction data and preprocessing the raw transaction data to construct a member feature vector. The clustering engine module is configured to use the K-Means++ algorithm to perform clustering calculations on the member feature vectors, so as to divide all members into multiple value-stratified clusters and determine the cluster center corresponding to each value-stratified cluster. The rule configuration module is configured to associate a corresponding marketing script with each value stratification cluster based on the cluster center of each value stratification cluster; The dynamic monitoring module is configured to periodically acquire real-time behavioral data of the member to be evaluated and update the corresponding real-time feature vector, and calculate the offset between the real-time feature vector and the cluster center of the value hierarchy cluster to which the member to be evaluated currently belongs. The execution module is configured to dynamically classify the member to be evaluated into a target value stratification cluster according to the offset when the offset exceeds a preset offset threshold, and call the marketing script associated with the target value stratification cluster.