Promotion template calling method based on member loss rate prediction

Abstract

The invention relates to the technical field of marketing promotion, and discloses a member churn rate prediction-based promotion template calling method, which comprises the following steps of collecting behavior data of a plurality of members, and performing preprocessing, the behavior data comprising purchase records, browsing history and social interaction of the members; based on the preprocessed data, establishing a wastage rate prediction model, the wastage rate prediction model being used for calculating the wastage probability of each member; classifying the members according to the loss probabilities, wherein the classification comprises members with high loss probabilities, members with medium loss probabilities and members with low loss probabilities; and according to the classification result, intelligently recommending a promotion activity template adapted to each member group. According to different member loss rates, member groups of different grades are divided, a promotion template is recommended in a targeted manner, and corresponding promotion activities are generated. According to the technical scheme, the intelligence degree of promotion activity generation is greatly improved, and activity recommendation is more personalized and targeted.

Description

Technical Field

[0001] The present invention relates to the technical field of marketing promotion, and in particular to a promotion template calling method based on member churn rate prediction. Background Art

[0002] In current member management and marketing promotion activities, many platforms still rely on manual operations to configure promotion activities. Specifically, existing technologies usually require people to manually input configuration parameters in the system and generate promotion activities based on these manual configurations. This method of relying on manual configuration is not only inefficient, but also easily interfered by human factors, making it difficult to ensure the efficiency and quality of the generation of promotion activities.

[0003] In order to improve efficiency, the existing technology provides some commonly used promotion activity templates for manual selection. These templates can help users reduce duplication of work and improve the convenience of configuration when configuring activities. However, the existing template recommendation method relies on manual selection and manual configuration. Although common template options are provided, the personalized needs and behavior patterns of different member groups are not taken into account. Therefore, the accuracy and pertinence of template recommendations are low, and it is impossible to intelligently recommend suitable promotion templates for specific member groups based on factors such as member loss risk and purchasing behavior. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a promotion template calling method based on member churn rate prediction, which solves the problem of low accuracy and pertinence of template recommendations.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a promotion template calling method based on member churn rate prediction comprises the following steps: Collecting and preprocessing the behavior data of multiple members, wherein the behavior data includes members' purchase records, browsing history, and social interactions; Based on the preprocessed data, a churn rate prediction model is established, wherein the churn rate prediction model is used to calculate the churn probability of each member; Classifying members according to the churn probability, wherein the classification includes high churn probability members, medium churn probability members and low churn probability members; According to the classification results, intelligently recommend promotion activity templates suitable for each member group, the promotion activity templates including more-recommend-more-send activities, volume-reaching reward activities and daily ranking activities; Automatically configure personalized promotion activities for each type of member based on the recommended promotion activity template; Provide real-time feedback and adjust recommendation strategies based on the implementation effect of the promotion activities.

[0006] Preferably, the churn rate prediction model is constructed by Bayesian reasoning, and the Bayesian reasoning model uses the following formula to predict the churn rate: Among them, P(Y|X) represents the probability of churn, P(X|Y) is the probability of the behavioral feature in the churn state, P(Y) is the prior probability of churn, and P(X) is the prior probability of the behavioral feature.

[0007] Preferably, the step of classifying members according to the churn probability includes: Members with high churn probability: The churn probability is greater than the first set threshold; Members with medium churn probability: The churn probability is between the first and second thresholds. Members with low churn probability: The churn probability is less than the set second threshold.

[0008] Preferably, the promotion activity template includes the following types: More promotion, more gifts: rewards based on members’ purchase frequency or activity; Volume Reward Activities: rewards for members with high frequency of consumption; Daily ranking activity: Generate rankings and award rankings based on member participation or purchasing behavior.

[0009] Preferably, the step of intelligently recommending promotion activity templates adapted to each member group is performed by a multi-objective optimization method, and the optimization objectives include but are not limited to: Maximize the benefits of member activities; Minimize the risk of losing members; By solving multi-objective optimization problems, the optimal activity template recommendation is obtained.

[0010] Preferably, the multi-objective optimization model is solved by the following objective function: Among them, P i is the probability of loss of the i-th member, L i (x) is the member’s loss when he / she does not participate in the recommendation activity, E i (x) is the benefit after members accept the recommendation activity, λ1 and λ2 are weight coefficients, and N is the total number of members.

[0011] Preferably, the recommendation strategy of the recommendation activity template is optimized by game theory analysis, and the game theory analysis is modeled based on the member's response utility function, and the utility function includes: U i (x) = α·R i (x)-β·C i (x) Among them, U i (x) represents the overall utility of the i-th member for the promotion activity template, R i (x) represents the income of the i-th member after responding to the recommendation activity, C i (x) represents the cost of the ith member participating in the promotion activity, and α and β are weight coefficients.

[0012] Preferably, the real-time feedback mechanism is optimized by optimal control theory, and the recommendation strategy is adjusted in real time to cope with changes in member behavior. The optimal control problem is solved by the following objective function: Among them, P t is the probability of churn at time t, L t (u(t)) is the loss of not participating in the activity at time t, E t (u(t)) is the benefit after participating in the activity at time t, u(t) is the control variable of the recommended activity template at time t, T is the recommendation period, and λ is the weight coefficient.

[0013] The promotion template calling system based on member churn rate prediction includes: Behavior data collection module, used to collect members' behavior data; A data preprocessing module, used for cleaning, feature selection and feature normalization of the member behavior data; A churn rate prediction module, used to predict the probability of member churn and classify member groups based on the member data; Recommendation module, used to recommend suitable promotion activity templates based on churn probability and member groups; The adjustment module is used to make feedback adjustments and optimize the recommendation strategy based on the real-time effects of the activity.

[0014] Preferably, the churn rate prediction module uses a Bayesian reasoning method for prediction, and the recommendation module provides optimal activity template recommendations based on a multi-objective optimization method and game theory analysis.

[0015] The present invention provides a promotion template calling method based on member churn rate prediction. It has the following beneficial effects: 1. The present invention divides member groups into different levels according to the different member churn rates, and recommends promotion templates in a targeted manner to generate corresponding promotion activities. This technical solution greatly improves the intelligence level of promotion activity generation, making activity recommendations more personalized and targeted. Compared with the single push or unified marketing methods in the prior art, the present invention significantly improves the participation rate and effect of activities by accurately identifying churn risks and tailoring activity templates for different member groups, solving the problem of the lack of refined adjustment and intelligent analysis in traditional promotion methods.

[0016] 2. The present invention uses a churn rate prediction model based on Bayesian reasoning, combined with members' purchase behavior data, historical interaction records and other features, to accurately calculate the churn probability of each member. Through multi-dimensional modeling and prediction of member churn, the technical effect of efficiently and accurately identifying members with high churn risk is achieved. Compared with the churn prediction model in the prior art that relies on simple statistical methods or basic classification algorithms, the present invention can consider the influence of multiple factors in a more refined manner, avoiding the problems of over-simplification or biased prediction that may occur in traditional methods.

[0017] 3. The present invention achieves accurate member grouping by combining the churn probability with member behavior data, and recommends personalized promotional activities according to the needs of different groups. This technical solution effectively avoids the inefficient marketing model of "unified push" in traditional technical solutions. By dynamically grouping each member based on multi-dimensional characteristics such as churn probability, purchase frequency, and purchase amount, the present invention can provide more targeted preferential activities for high-risk churn members, thereby improving member retention and activity.

[0018] 4. The promotion template calling system of the present invention, based on member churn rate prediction, not only greatly improves the effect of promotion activities and the economic benefits of the platform, but also effectively reduces the risk of member churn, bringing higher user retention rate and long-term profitability to the platform through intelligent data processing, accurate churn prediction, personalized recommendation strategy, and dynamic adjustment mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flowchart of the steps of the promotion template calling method based on member churn rate prediction of the present invention; Figure 2 This is a diagram of the promotion template calling system architecture based on member churn rate prediction of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Please refer to the attached Figure 1 The embodiment of the present invention provides a promotion template calling method based on member churn rate prediction, comprising the following steps: S1. Collect behavioral data of multiple members, including members' purchase records, browsing history, and social interactions.

[0022] Specifically, step S1 aims to provide sufficient raw data support for the establishment of a churn rate prediction model by collecting behavioral data of multiple members. Specifically, member behavioral data refers to various operations and activities performed by members on the platform, including but not limited to purchase records, browsing history, social interactions, etc. By collecting and analyzing these data, the system can extract information reflecting member behavioral characteristics, providing the necessary data basis for subsequent churn probability prediction, personalized promotion activity recommendation and other functions.

[0023] In this embodiment, the types of data collected are wide, covering various ways of interaction between members and the platform. These data are stored and managed through log systems, database systems, etc. In this way, a feature vector of members can be formed, which is convenient for churn rate prediction and promotion activity design in subsequent steps.

[0024] Specifically, members’ behavioral data mainly includes the following categories: Purchase records: Purchase records refer to the consumption history data of members on the platform. Specifically, these data include but are not limited to information such as purchase time, type of purchased goods, and purchase amount. Purchase history reflects the member's consumption habits, purchase frequency, and amount, and is one of the important indicators for evaluating member loyalty and churn risk. For example, if a member's purchase frequency gradually decreases or the purchase amount decreases significantly, it may be a precursor to his or her churn.

[0025] Browsing history: Browsing history records the browsing behavior of members on the platform, including the pages visited, the frequency of product browsing, browsing time, etc. By analyzing this data, we can determine the member's interest in the platform and participation intention. For example, if a member stays on a certain product page for a long time but does not make further purchases, it is possible that he is interested in the product but has not yet made a purchase decision.

[0026] Social interaction: Social interaction refers to various interactions that members have with other users or the platform itself on the platform, such as comments, likes, sharing, and participation in activities. Social interaction reflects the engagement and activity of members. Active social interaction is usually a sign of high member loyalty and low risk of churn. For example, members who frequently participate in social activities may have high loyalty to the platform and are less likely to churn.

[0027] In some embodiments, the collection of behavior data may also include other indicators related to member behavior, such as login frequency, account active time, post-purchase evaluation behavior, etc. Through these data, the system can fully understand the behavior patterns of members and further provide rich information for the churn prediction model.

[0028] Specifically, the collection of behavioral data can be carried out through a variety of technical means. For example, using a logging system, every operation of a member on the platform can be recorded in detail, and this data can be stored in a data warehouse for subsequent analysis and modeling. In addition, the platform can also connect data with third-party data sources (such as payment platforms, social platforms, etc.) through API interfaces to further supplement members' behavioral data.

[0029] It should be noted that members' behavioral data usually needs to be de-identified to ensure the privacy and security of the data. In the process of collecting data, anonymization technology can be used to prevent the leakage of members' personal information. At the same time, in order to ensure the quality of data, the system should ensure the accuracy and completeness of the data and screen and correct abnormal data.

[0030] The collected behavioral data will be used as input data for subsequent churn rate prediction modeling. In practical applications, the original data may contain noise, missing values, or incompleteness, so data preprocessing is required to improve data quality and the accuracy of model prediction. The main steps of data preprocessing include data cleaning, feature selection, and feature normalization.

[0031] Data cleaning: For the collected raw data, the system first needs to clean it up to remove incomplete, abnormal or erroneous data. For example, if an unrecognizable product code or unreasonable timestamp appears in a member's purchase record, the system needs to correct or remove it. This process helps eliminate interference factors in the data, thereby improving the prediction effect of the model.

[0032] Feature selection: The collected behavioral data may contain some redundant or irrelevant features. To improve the efficiency and accuracy of the model, it is necessary to use feature selection algorithms to select features that are highly relevant to churn prediction. For example, through correlation analysis or model-based feature importance ranking, features with significant impacts such as purchase frequency, browsing time, and social interaction frequency can be selected.

[0033] Feature normalization: There may be large differences between different features, for example, the value ranges of purchase amount and browsing time are quite different. In order to prevent certain features from occupying too large a weight in model training, the features need to be normalized. Common normalization methods include Min-Max scaling and Z-score normalization, which aim to map feature values ​​to a uniform value range to ensure that each feature has a balanced impact on model training.

[0034] Taking purchase records as an example, suppose the system collects all purchase data of members in the past 6 months. By analyzing data such as purchase frequency and purchase amount, a feature vector can be generated for each member, as shown below: X i =(f i ,m i ,a i ,t i ) Among them, f i represents the purchase frequency of the i-th member, m i represents the total purchase amount of the i-th member, a i represents the average purchase amount of the i-th member, t i Indicates the most recent purchase time of the i-th member.

[0035] These features can help the model predict the probability of a member churn in the future. For example, if a member has a low purchase frequency and the last purchase was a long time ago, the model may predict that the member has a high probability of churn and take corresponding intervention measures.

[0036] Summary: This embodiment describes in detail the process of collecting member behavior data, including data types, collection methods, and data preprocessing methods. Through the collection and processing of this data, the system can provide a solid data foundation for subsequent churn rate prediction and promotion activity recommendations. It should be noted that the data collection and processing process is a dynamic and continuously optimized process. As the platform further understands member behavior, the system can continuously adjust the data collection and processing strategy to improve the accuracy of prediction and the effectiveness of activity recommendations.

[0037] By implementing the present invention, the platform can more accurately understand member behaviors and perform intelligent analysis based on these behavioral data, thereby effectively reducing member churn rates and improving the profitability of the platform and user satisfaction.

[0038] S2. Preprocess member behavior data, including data cleaning, feature selection and feature normalization.

[0039] Specifically, step S2 involves the process of preprocessing the collected member behavior data. Preprocessing is a crucial step in data analysis, which can effectively improve data quality and help the subsequent churn rate prediction model obtain more accurate results. Through appropriate data preprocessing methods, the system can remove unnecessary noise, eliminate incomplete or abnormal data, and optimize feature selection, thereby improving the performance of the prediction model.

[0040] In this embodiment, data preprocessing mainly includes the following steps: data cleaning, feature selection, and feature normalization.

[0041] In one possible implementation, data cleaning is the first step, and its purpose is to remove or repair outliers, missing values, and inconsistent data in the data. For member behavior data, there may be some incomplete records, for example, a member's purchase record lacks the purchase amount or timestamp. In this case, the following methods can be used to handle missing values: for missing values, mean filling, interpolation, or inference based on other data sources can be used; for outliers, the system can filter by setting reasonable upper and lower limits to ensure the accuracy and consistency of the data. For example, when a member's purchase amount is significantly higher than its normal range, the system will mark it as an outlier and process it, or directly remove it.

[0042] As an option, feature selection is one of the key steps in data preprocessing. During the feature selection process, the system selects the features that have a higher impact on the prediction results by analyzing the relationship between each feature in the data set and member churn. For example, features such as purchase frequency, total purchase amount, and average purchase amount can reflect the activity and loyalty of members to a certain extent, so these features are considered to be key features closely related to churn prediction. Conversely, some redundant features or features that are irrelevant to churn prediction (such as the geographic location of certain users) may be removed to simplify the data set and improve prediction efficiency.

[0043] Specifically, in the feature selection stage, the system can use some classic feature selection methods, such as information gain, chi-square test, mutual information, L1 regularization, etc. Through these methods, the system can screen out features with high information content for the prediction model, thereby improving the accuracy and robustness of the model. For example, in some embodiments, the system may sort the features according to their relevance or prediction importance and select the top N features for the subsequent modeling process.

[0044] Feature normalization is another important preprocessing operation, which aims to eliminate the difference in dimensions between different features and avoid the influence of too large or too small numerical range of some features on model training. In actual operation, feature normalization often adopts Min-Max normalization or Z-score normalization method. For example, Min-Max normalization scales the feature value to the interval [0,1], while Z-score normalization converts the feature value into a standard normal distribution with a mean of 0 and a variance of 1. After this processing, the values ​​of different features will be unified into a relatively small range, thereby improving the training effect of the model.

[0045] It should be noted that feature normalization not only helps to speed up the training of the model, but also improves the stability of the prediction results, especially when using machine learning models with distance metrics (such as KNN, SVM, etc.). In some embodiments, if the numerical range of certain features is significantly different from that of other features, normalization can effectively prevent a certain feature from occupying too much weight during the training process, affecting the overall performance of the model.

[0046] After data preprocessing is completed, the member behavior data that has been cleaned, screened, and normalized will be passed as input data to the subsequent churn rate prediction model to accurately predict the churn probability. At this point, the system will be able to generate a feature vector for each member based on this data and provide a reliable basis for subsequent churn prediction.

[0047] In summary, this embodiment ensures the quality and validity of input data through a series of preprocessing steps, including data cleaning, feature selection and feature normalization. These steps not only help to improve the accuracy of the churn rate prediction model, but also reduce the training time of the model to a certain extent and enhance the robustness of the model.

[0048] S3. Based on the preprocessed data, a churn rate prediction model is established. The churn rate prediction model is used to calculate the churn probability of each member.

[0049] Specifically, step S3 is a key link, which involves the establishment of a churn rate prediction model. By constructing a suitable prediction model, the present invention can accurately calculate the churn probability of each member and provide a basis for the subsequent promotion strategy formulation. In order to achieve this goal, the present invention selects a Bayesian inference method based on probabilistic reasoning, which can effectively process complex member behavior data and calculate the churn probability of each member based on data characteristics.

[0050] In this embodiment, the churn rate prediction model uses the Bayesian formula to estimate the churn probability of each member. Specifically, assuming that the feature vector of the i-th member is X i =(f1,f2,…,f n ), where f1,f2,…,f n They represent the member's purchase frequency, total purchase amount, average purchase amount, and last purchase time. The Bayesian formula can be expressed as: Among them, P(Y i |X i ) is the feature vector X of a given member i i The conditional probability of churn is the probability of predicting the churn of the member; P(X i |Y) is the observed feature vector X under the condition of member loss i The likelihood of membership loss; P(Y) is the prior probability of membership loss, that is, the overall probability of membership loss among all members; P(X i ) is the observed feature vector X i The prior probability of .

[0051] Specifically, in some embodiments, Bayesian reasoning can estimate the above probability by training historical member behavior data. i |X i ) will be used as the prediction result of the probability of loss for each member.

[0052] It should be noted that the key advantage of the Bayesian model is that it can combine prior information with observational data to provide a flexible way of probabilistic inference. Therefore, the present invention can comprehensively consider the historical behavior of members in multiple dimensions to generate the probability of churn. For example, if a member has a low purchase frequency, a small total purchase amount, and a long recent purchase time, then the member's churn probability may be higher. Through these data features, the Bayesian model can provide personalized churn predictions for each member, thereby helping the platform identify members with a higher potential risk of churn.

[0053] In one possible implementation, the prior probability of churn P(Y i |X i) can be estimated by the churn rate in historical data. For example, the system can calculate the proportion of all members who churned in a certain period of time in the past as the prior probability of churn. This probability can reflect the overall churn trend of the platform in a specific period of time, thereby providing reasonable prior information for the prediction model. For example, if the number of members who churned in a certain period of time accounts for 30% of the total members, the system can calculate P(Y i |X i )=0.3 as the preliminary churn probability. However, in actual application, the platform can further adjust this prior probability according to the specific behavior data of the members, so as to improve the accuracy of the prediction results.

[0054] In order to improve the accuracy of the model, the present invention can further optimize feature selection through feature engineering. For example, by introducing new data dimensions such as member activity features, loyalty features, interaction frequency, etc., the model can make a more refined prediction of the member's churn probability. As an option, the system can also combine other machine learning algorithms (such as support vector machines, random forests, etc.) to further integrate the churn probability to optimize the final prediction results.

[0055] It is understandable that by applying the Bayesian reasoning method to the churn rate prediction model, the system can accurately estimate the churn probability based on each member's historical behavior data. This method has strong flexibility and adaptability, can handle a large amount of complex member data, and provide data support for subsequent promotional activities. In addition, the probabilistic nature of the Bayesian model also enables the platform to design personalized marketing plans for member groups with different churn probabilities, thereby achieving more accurate member retention and activation.

[0056] In summary, in this embodiment, the churn rate prediction model is established through the Bayesian reasoning method, which can not only provide accurate churn probability prediction results, but also effectively combine historical data with prior information to provide decision support for the platform. Through this model, the system can deeply explore the potential rules in member behavior data and provide a scientific basis for subsequent precision marketing.

[0057] S4. Classify members according to the probability of churn, including members with high churn probability, members with medium churn probability and members with low churn probability.

[0058] Specifically, step S4 involves recommending the most suitable promotion activity template for different member groups based on the prediction results of the churn probability. By using an optimization algorithm, this step aims to maximize the benefits of members participating in the promotion activities while minimizing the risk of member churn. Specifically, the core goal of step S4 is to design an optimal activity template recommendation strategy to improve the overall benefits of the platform and reduce the churn rate.

[0059] In this embodiment, the recommendation strategy is based on the churn probability and benefit analysis and is implemented through a multi-objective optimization problem. The objective function of the optimization problem is as follows: Where x is the activity template parameter, such as discount strength, number of gifts, etc.; P i is the probability of loss of the i-th member; L i (x) is the loss when members do not participate in the recommendation activities, reflecting the negative impact of loss on the platform; E i (x) is the benefit value after the i-th member participates in the recommendation activity, which represents the benefits brought by the activity, such as an increase in member purchases, increased platform activity, etc.; λ1 and λ2 are weight coefficients, which are used to adjust the balance between churn risk and activity benefits; N is the total number of members.

[0060] Specifically, the goal of this optimization problem is to minimize the negative effects of churn and maximize the benefits of activities In actual calculations, the churn risk is usually calculated by the member churn probability P obtained through the prediction model. i The benefit value depends on the impact of specific activity plans on member behavior, such as the increase in purchases and interaction frequency after participating in the activity.

[0061] It should be noted that, in some embodiments, the loss L i (x) The expected change in platform revenue can be estimated by taking into account factors such as the loss of revenue due to lost members and the decline in brand value. i (x) can be measured by the increase in consumption after members participate in the event or the user activity of the platform. For example, some members increase their purchase frequency after the event, thereby increasing the overall revenue of the platform.

[0062] For example, suppose the platform recommends a discount promotion for a certain member group. This promotion may make members with a high probability of churn more willing to participate, thereby reducing their risk of churn. However, this may also increase the platform's short-term costs. Therefore, during the optimization process, the system will recommend the most appropriate discount level or number of gifts based on the trade-off between churn risk and benefit, so as to maximize benefits without increasing excessive costs.

[0063] In a possible implementation, the selection of λ1 and λ2 can be adjusted according to the specific needs of the platform. For example, if the platform pays more attention to the control of member churn, the value of λ1 can be appropriately increased to reduce the risk of churn; if the platform pays more attention to the economic benefits brought by the activities, the value of λ2 can be increased.

[0064] It is understandable that by solving this optimization problem, the system can recommend personalized promotion activity templates for different member groups. For example, for members with a higher probability of churn, the system can recommend more attractive activities (such as larger discounts) to reduce their churn risk; while for members with a lower probability of churn, more moderate activities can be recommended to reduce platform costs while maintaining higher benefits.

[0065] In addition, as an option, the present invention can also combine multiple optimization algorithms, such as genetic algorithm, particle swarm optimization algorithm, etc., to further improve the accuracy and efficiency of the recommendation strategy. Through these intelligent optimization algorithms, the system can quickly find the best activity template recommendation solution in large-scale member data.

[0066] In summary, step S4 intelligently recommends the most appropriate promotion activity template by combining the churn probability and activity benefits through a multi-objective optimization algorithm. This method can effectively reduce the member churn rate while improving the overall benefits of the platform and ensuring the optimal allocation of resources. Through this optimization process, the present invention can provide accurate member retention strategies in actual operations and promote the continuous growth and development of the platform.

[0067] S5. Based on the classification results, intelligently recommend promotion activity templates suitable for each member group. The promotion activity templates include more-promote-more-send activities, volume-reaching reward activities and daily ranking activities.

[0068] Specifically, step S5 involves further implementing personalized promotion activities for different member groups based on the prediction results of the churn probability. This step is based on the previous recommendation results and optimization goals, and automatically configures personalized activities to maximize the participation benefits of each member and ensure the efficient use of platform resources. Specifically, step S5 quantifies the utility of each recommended activity by using a utility function, and then evaluates the impact of the activity on member behavior and determines the final promotion activity configuration.

[0069] In this embodiment, the form of the utility function is as follows: Among them, U(a) is the utility value of the activity template parameter a, which reflects the overall benefit of the activity plan; P i is the probability of loss of the i-th member; E i (a) is the benefit of the ith member after participating in the promotion activity, which is usually manifested in the member's purchasing behavior after participating in the activity, the increase in platform activity, etc.; C i (a) is the cost for the i-th member to participate in the activity, including discounts, rewards, subsidies, etc. provided by the platform; N is the total number of members.

[0070] Specifically, the utility function U(a) calculates the overall benefit of the activity template parameter a among all members. For each member i, its utility consists of two parts: on the one hand, the benefit P brought by the activity i ·E i (a) is the contribution of members’ behavioral changes after participating in the activity to the platform; on the other hand, it is the cost C brought by the activity. i (a), which is the platform expenditure required to attract members to participate. Through this utility evaluation, the system can ensure that the recommended activities can not only reduce the risk of churn, but also maximize the overall benefits of the platform.

[0071] For example, suppose a promotional activity is a discount offer for a certain type of member, which stimulates member purchases by reducing product prices. For members with a higher probability of churn, the benefits of the activity may be reflected in their increased purchase frequency and platform interaction behavior, while the cost of the discount will also affect the overall benefits of the platform. Through the utility function, the system can evaluate the impact of the discount on different member groups, thereby adjusting the activity plan to maximize the overall benefits.

[0072] It should be noted that, in some embodiments, the evaluation of the utility function may also take into account other factors, such as member loyalty, activity duration, market competition, etc. These factors may affect the final benefits and costs of the activity, and thus may be further adjusted as part of the optimization goal.

[0073] As an option, during the configuration of personalized promotion activities, the platform can also make dynamic adjustments based on the actual feedback of the activities. For example, some activity plans may show good results in the early stage, but as time goes by, members' enthusiasm for participation may gradually decline. In this case, the system can adjust the recommended strategy and content of the activity based on real-time data and historical participation results.

[0074] In one possible implementation, the utility function can also be further enhanced by adding the member's lifetime value (LTV). The lifetime value of a member refers to the total contribution of a member to the platform during his or her lifetime, including long-term consumption, referrals of new members, social interactions, and other benefits. By incorporating lifetime value into the utility function, the system can further optimize promotional campaigns, better identify members with greater potential, and customize efficient campaign strategies for them.

[0075] It is understandable that through this step, the system can not only recommend personalized activity templates for each member under given resource conditions, but also continuously optimize the activity results. This process ensures the efficient use of the platform's resource investment, while also improving the member experience and the long-term stable development of the platform.

[0076] In summary, step S5 evaluates the benefits of the activity template through the utility function, and then configures personalized promotion activities. This method ensures that the platform's resources can be optimally configured by weighing the probability of member churn and the benefits of the activity, maximizing the platform's revenue while reducing the risk of churn. Through this refined management, the system can provide the most suitable activity plan for each member group and promote the long-term success of the platform.

[0077] S6. Automatically configure personalized promotion activities for each type of member based on the recommended promotion activity template.

[0078] Specifically, in step S6, after recommending the promotion activity template, the system automatically configures personalized promotion activities based on the characteristics and churn probability of each member. This process combines the historical behavior of members, the predicted churn probability, and the activity goals to ensure that the recommended promotion activities can maximize the attraction of high-risk churn members, while also considering cost-effectiveness and avoiding resource waste.

[0079] Specifically, after recommending an activity template, the system will select the most appropriate activity parameters for each member. For example, for members with a higher probability of churn, the system may recommend a higher discount or increase the number of gifts to increase members' willingness to participate. For members with a lower probability of churn, some basic promotions can be recommended to avoid over-subsidy and maximize benefits.

[0080] Exemplarily, the system configures personalized promotion activities through the following steps: Member data analysis: The system first calculates the churn risk of each member based on each member's behavior data, historical purchase records, and churn prediction models. For example, a member may have a higher probability of churn due to factors such as reduced purchase frequency or a long time since the last purchase; Activity recommendation optimization: Based on the member's churn probability and historical behavior, the system will select the appropriate activity type. For example, for high-risk churn members, the system may recommend discount or gift activities; for more active members, it may recommend points redemption activities or other long-term benefits.

[0081] Activity parameter adjustment: In addition to the selection of activity type, the system will also adjust the specific parameters of the activity, such as the discount amount, number of gifts, etc., according to the economic status and participation ability of members, to ensure a balance between the cost and benefit of the activity.

[0082] Event implementation: Ultimately, the system will automatically configure personalized events for each member and push event information through appropriate channels (such as SMS, email, platform push, etc.). These personalized promotional activities will be consistent with the platform's marketing strategy, thereby maximizing member participation and platform revenue.

[0083] It should be noted that this personalized configuration strategy can adapt to member behavior changes in real time. For example, if a member frequently participates in promotional activities in a short period of time, the system may reduce incentives for that member and instead provide more attractive activities to members with a higher risk of churn. In this way, the system ensures that activities can maximize member participation and platform benefits while maintaining low costs.

[0084] As an option, the system can combine other user behavior data (such as platform interaction, participation in the recommendation system, etc.) to further refine member segments and recommend different levels of personalized activities for different segments.

[0085] S7. Provide real-time feedback and adjust recommendation strategies based on the implementation effect of the promotion activities.

[0086] Specifically, step S7 mainly involves adjusting the promotion strategy through a real-time feedback mechanism. This process aims to continuously optimize the design of promotion activities by continuously collecting and analyzing member behavior feedback, platform benefit data, and market dynamics to ensure the continued effectiveness and optimal results of the activities. Through this feedback and adjustment mechanism, the platform can flexibly respond to changes in member needs, changes in the market environment, and the influence of other external factors, thereby achieving maximum benefits and reducing member churn.

[0087] In this embodiment, step S7 uses optimal control theory to adjust the parameters of the promotion activity so that the activity design can be optimized in real time according to member behavior and platform benefits. Specifically, the system dynamically adjusts the recommended activity template by obtaining member participation data and platform benefit information, thereby maximizing revenue and reducing churn risk. This process can be optimized through the following objective function: Among them, u(t) is the control variable of the activity template recommended at time point t, which represents the promotion activity recommended by the system at that moment; P t is the churn probability at time point t, indicating the possibility of member churn at that moment; L t (u(t)) is the loss when the member does not participate in the activity at time point t, reflecting the loss risk that the platform may bear when no intervention is taken; E t (u(t)) is the benefit of members participating in the activity at time point t, which represents the benefits brought by the activity, such as purchases, interactions or activity levels; λ is the weight coefficient, which is used to balance the relationship between the risk of churn and the benefits of the activity; T is the total time period, which represents the time range of the entire promotion activity cycle.

[0088] Specifically, the optimization process of the objective function is mainly based on real-time feedback data, including member behavior data (such as purchase frequency, interaction frequency, click-through rate, etc.) and platform benefit data (such as sales, user activity, etc.). Through these data, the system calculates the risk of member loss and the benefits brought by the activity in real time, and adjusts the activity strategy according to the optimal control theory.

[0089] As an option, in order to ensure the accuracy of optimization results, the system may also use machine learning or deep learning technology to analyze historical data to provide predictive support for future recommendation activities. In this way, the system can learn the behavior patterns of members and their responses to promotional activities in a relatively short period of time, thereby improving the effectiveness of recommendation activities.

[0090] In one possible implementation, the feedback mechanism can be implemented based on a real-time data collection module, which monitors and records member behavior changes in real time, such as each login, purchase, click and other interactive data, and feeds this data back to the system through a data transmission and processing module. At the same time, the system will dynamically adjust the content and form of the recommendation activities based on this feedback information and the set optimization objective function.

[0091] For example, suppose that at the beginning of an activity, the platform finds through analysis of real-time data that certain member groups have a high risk of churn, while the benefits of the activity are low. The system may increase discounts or provide additional incentives to attract these high-risk members to participate, reduce churn and increase benefits. In this way, the system can flexibly adjust activity strategies according to actual conditions to ensure optimal use of platform resources.

[0092] It should be noted that this feedback adjustment mechanism does not rely solely on a single data source, but is a comprehensive dynamic feedback process. The platform can continuously adjust activity plans through multi-channel data collection, analysis and feedback to cope with the changing market environment and member needs. This mechanism ensures that promotional activities can adapt to various internal and external changes in real time, maintaining a high degree of flexibility and responsiveness.

[0093] It is understandable that through the real-time feedback and adjustment of step S7, the platform can design personalized promotion activities for each member group, which can maximize the benefits of the activities and effectively reduce the risk of member loss. In addition, the application of optimal control theory provides a mathematical and systematic adjustment framework for the system, making the optimization process of the entire promotion activity both efficient and scientific.

[0094] In summary, step S7 ensures dynamic adjustment and optimization of promotion activities by combining real-time feedback with optimal control theory. This process enables the platform to quickly respond to member behavior and market changes, maximize platform benefits, and achieve lower churn rates.

[0095] Please refer to the attached Figure 2 The present invention also provides a promotion template calling system based on member churn rate prediction, including: Behavior data collection module, used to collect members' behavior data; A data preprocessing module, used for cleaning, feature selection and feature normalization of the member behavior data; A churn rate prediction module, used to predict the probability of member churn and classify member groups based on the member data; Recommendation module, used to recommend suitable promotion activity templates based on churn probability and member groups; The adjustment module is used to make feedback adjustments and optimize the recommendation strategy based on the real-time effects of the activity.

[0096] Specifically, the present invention provides a promotion template calling system based on member churn rate prediction. The system comprehensively analyzes and processes member behavior data through the collaborative work of multiple modules, thereby achieving accurate churn prediction and personalized promotion activity recommendations. Specifically, the system includes the following modules: 1. Behavioral data collection module Specifically, this module is responsible for collecting various behavioral data of members on the platform in real time. The behavioral data of members include but are not limited to purchase frequency, total purchase amount, single purchase amount, last purchase time, platform visit frequency, etc. Through these data, the system can fully understand the behavioral characteristics of members and provide basic data for subsequent analysis.

[0097] As an option, the behavioral data collection module can also enrich the data source by integrating other data sources (such as social platform data, customer service records, etc.), thereby improving the accuracy of the churn prediction model.

[0098] 2. Data preprocessing module In this embodiment, the data preprocessing module performs a series of cleaning, feature selection and feature normalization processes on the collected member behavior data.

[0099] Data cleaning: First, missing values ​​and outliers are processed to ensure the integrity and consistency of the data.

[0100] Feature selection: Based on business needs and data analysis results, select features that have a significant impact on churn prediction, such as purchase frequency, purchase amount, and last purchase time, perform feature screening, and reduce redundant features.

[0101] Feature normalization: By standardizing or normalizing the feature data, the impact of different feature dimensions is eliminated, making the data more unified, which is conducive to the subsequent churn prediction model training.

[0102] It should be noted that the data preprocessing module provides high-quality data input for the churn rate prediction module, thereby improving the accuracy and reliability of subsequent analysis results.

[0103] 3. Churn rate prediction module Specifically, the churn rate prediction module uses preprocessed data to predict the churn probability of each member by building a suitable machine learning model (such as decision tree, support vector machine, neural network, etc.).

[0104] This module can not only predict the churn probability of a single member, but also classify members into groups based on these probabilities. Specifically, members can be divided into high churn probability groups, medium churn probability groups, and low churn probability groups.

[0105] For example, for a group with a high probability of churn, the system may focus on recommending incentive activities, such as substantial discounts or gifts; whereas for a group with a low probability of churn, it may recommend simple loyalty enhancement activities.

[0106] It is understandable that the churn rate prediction module not only makes predictions based on members' historical behavior data, but also makes dynamic adjustments based on real-time behavior data, thereby improving the accuracy of churn predictions.

[0107] 4. Recommendation module In this embodiment, the recommendation module recommends the most suitable promotion activity template according to the churn probability and the characteristics of the member group.

[0108] The recommendation module will automatically match and recommend activity templates suitable for each member based on their churn probability. For example, members with a high churn probability may be recommended activities with large discounts or additional gifts, while members with a medium churn probability may be recommended activities with point rewards or limited-time discounts.

[0109] It should be noted that the recommendation process of this module adopts a multi-objective optimization algorithm to ensure that while increasing the member participation rate, it does not waste excessive platform resources and maximizes benefits.

[0110] 5. Adjustment module In one possible implementation, the adjustment module makes feedback adjustments based on the real-time effects of the activity. By monitoring the participation and actual benefits of the promotion activity, the adjustment module can optimize the recommendation strategy in real time to cope with changes in the market environment and dynamic changes in member behavior.

[0111] For example, if a campaign template has a low participation rate after recommendation, the system will adjust the campaign parameters (such as discount strength, number of gifts, etc.) or change the recommendation strategy (such as recommending more targeted discounts) based on members' feedback and interaction data.

[0112] It is understandable that the feedback adjustment mechanism of this module can enable the system to achieve self-learning and evolution, and over time, it can continuously improve the accuracy and effectiveness of activities, ultimately minimizing churn and maximizing platform benefits.

[0113] In this embodiment, the workflow of the system can be summarized as the following steps: The behavior data collection module collects members’ real-time behavior data; The data preprocessing module cleans and standardizes the behavioral data; The churn rate prediction module predicts the churn probability of each member by building a model and classifies the members; The recommendation module recommends suitable promotion activity templates based on member groups and their churn probability; The adjustment module optimizes the recommendation strategy and adjusts the activity parameters according to the real-time feedback of the activity.

[0114] Through the collaborative work of these modules, the system can efficiently predict member churn, recommend activities and optimize results, thereby achieving precise marketing and efficient use of resources on the platform.

[0115] As an option, the modules in the above system can be deployed through a distributed architecture to cope with the processing needs of large-scale member data. Each module can run independently and cooperate with each other through efficient data transmission and communication mechanisms.

[0116] It should be noted that the machine learning model used in the churn rate prediction module can be selected according to actual needs. For example, algorithms such as decision trees, support vector machines, random forests, neural networks, etc. can be used, or integrated learning methods can be used to further improve the accuracy of prediction.

[0117] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A promotion template calling method based on member churn rate prediction, characterized in that: The following steps are involved: Collecting and preprocessing the behavior data of multiple members, wherein the behavior data includes members' purchase records, browsing history, and social interactions; Based on the preprocessed data, a churn rate prediction model is established, wherein the churn rate prediction model is used to calculate the churn probability of each member; Classifying members according to the churn probability, wherein the classification includes high churn probability members, medium churn probability members and low churn probability members; According to the classification results, intelligently recommend promotion activity templates suitable for each member group, the promotion activity templates including more-recommend-more-send activities, volume-reaching reward activities and daily ranking activities; Automatically configure personalized promotion activities for each type of member based on the recommended promotion activity template; Provide real-time feedback and adjust the recommendation strategy based on the implementation effect of the promotion activities.

2. The promotion template calling method based on member churn rate prediction according to claim 1 is characterized in that: The churn rate prediction model is constructed by Bayesian reasoning, and the Bayesian reasoning model uses the following formula to predict the churn rate: Among them, P(Y|X) represents the probability of churn, P(X|Y) is the probability of the behavioral feature in the churn state, P(Y) is the prior probability of churn, and P(X) is the prior probability of the behavioral feature.

3. The promotion template calling method based on member churn rate prediction according to claim 1 is characterized in that: The step of classifying members according to the churn probability comprises: Members with high churn probability: The churn probability is greater than the first set threshold; Members with medium churn probability: The churn probability is between the first and second thresholds. Members with low churn probability: The churn probability is less than the set second threshold.

4. The promotion template calling method based on member churn rate prediction according to claim 1 is characterized in that: The promotion activity templates include the following types: More promotion, more gifts: rewards based on members’ purchase frequency or activity; Volume Reward Activities: rewards for members with high frequency of consumption; Daily ranking activity: Generate rankings and award rankings based on member participation or purchasing behavior.

5. The promotion template calling method based on member churn rate prediction according to claim 1 is characterized in that: The step of intelligently recommending promotion activity templates adapted to each member group is performed by a multi-objective optimization method, and the optimization objectives include but are not limited to: Maximize the benefits of member activities; Minimize the risk of losing members; By solving multi-objective optimization problems, the optimal activity template recommendation is obtained.

6. The promotion template calling method based on member churn rate prediction according to claim 5 is characterized in that: The multi-objective optimization model is solved by the following objective function: Among them, P i is the probability of loss of the i-th member, L i (x) is the member’s loss when he / she does not participate in the recommendation activity, E i (x) is the benefit after members accept the recommendation activity, λ1 and λ2 are weight coefficients, and N is the total number of members.

7. The promotion template calling method based on member churn rate prediction according to claim 1 is characterized in that: The recommendation strategy of the recommendation activity template is optimized through game theory analysis, and the game theory analysis is modeled based on the member's response utility function, and the utility function includes: U i (x)=α·R i (x)-β·C i (x) Among them, U i (x) represents the overall utility of the i-th member for the promotion activity template, R i (x) represents the income of the i-th member after responding to the recommendation activity, C i (x) represents the cost of the ith member participating in the promotion activity, and α and β are weight coefficients.

8. The promotion template calling method based on member churn rate prediction according to claim 1 is characterized in that: The real-time feedback mechanism is optimized through optimal control theory, and the recommendation strategy is adjusted in real time to cope with changes in member behavior. The optimal control problem is solved by the following objective function: Among them, P t is the probability of churn at time t, L t (u(t)) is the loss of not participating in the activity at time t, E t (u(t)) is the benefit after participating in the activity at time t, u(t) is the control variable of the recommended activity template at time t, T is the recommendation period, and λ is the weight coefficient.

9. A promotion template calling system based on member churn rate prediction, characterized in that: include: Behavior data collection module, used to collect members' behavior data; A data preprocessing module, used for cleaning, feature selection and feature normalization of the member behavior data; A churn rate prediction module, used to predict the probability of member churn and classify member groups based on the member data; Recommendation module, used to recommend suitable promotion activity templates based on churn probability and member groups; The adjustment module is used to make feedback adjustments and optimize the recommendation strategy based on the real-time effects of the activity.

10. The promotion template calling system based on member churn rate prediction according to claim 9, characterized in that: The churn rate prediction module uses Bayesian reasoning method to make predictions, and the recommendation module provides optimal activity template recommendations based on multi-objective optimization method and game theory analysis.

Images