Marketing promotion method based on member loss prediction

Abstract

The invention relates to the field of digital marketing, and discloses a member loss prediction-based marketing promotion method, which comprises the following steps of: 1, acquiring behavior data, transaction data, social interaction data and advertisement putting data of members; step 2, preprocessing the data, including data cleaning, feature selection and data standardization; 3, establishing a member loss prediction model, and predicting whether the members are lost or not; 4, establishing a promotion channel effect prediction model, and evaluating the effect of each promotion channel; 5, calculating the weight value of each promotion channel, wherein the weight value reflects the cost performance of the promotion channel; and step 6, sorting the promotion channels based on the weight values. Through accurate analysis based on member loss prediction, corresponding delivery budget can be allocated to each promotion channel, and reasonable allocation of resources of different channels is ensured, so that promotion activities can be more accurately matched with target audiences, and resource waste is avoided.

Description

Technical Field

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

[0002] With the booming development of e-commerce and the Internet industry, membership marketing has become an important means for businesses to improve customer loyalty, increase sales and expand market share. In order to improve marketing effectiveness, businesses need to attract more potential customers and maintain existing customers through precise marketing and promotion methods. However, how to effectively identify potential lost members, optimize promotion channels and reasonably allocate marketing resources are still major challenges facing the current marketing and promotion field. Existing traditional marketing methods mostly rely on manual design and experience judgment, and cannot make full use of big data and artificial intelligence technologies, resulting in inefficiency and waste of resources in many marketing activities.

[0003] At present, there are some technical solutions on the market that predict churn and optimize marketing strategies based on member behavior data. Traditional churn prediction methods mostly rely on historical data analysis, user tag classification, and some simple statistical models, such as logistic regression and decision trees. These methods predict whether members are likely to churn by analyzing data such as members' purchase frequency, activity, browsing behavior, and adjust marketing strategies based on this. At the same time, some technical solutions have also begun to try to combine multi-source data (such as social media, user interaction, etc.) to optimize the selection of promotion channels and budget allocation. However, these existing technologies still have certain limitations. They cannot accurately predict the impact of different promotion channels on member churn, and it is difficult to automatically adjust strategies based on real-time data, and they cannot meet the rapidly changing market needs.

[0004] Existing technical solutions mainly rely on static data analysis and manual intervention to predict member churn and adjust promotion strategies, but these methods have the following problems: Insufficient evaluation of promotion channel effectiveness: Among the existing solutions, most only focus on member churn prediction and ignore the comprehensive evaluation of the effectiveness of different promotion channels, resulting in a lack of accuracy in the allocation of promotion resources and an inability to ensure the optimization of marketing activities.

[0005] Lack of real-time dynamic adjustment mechanism: Existing methods usually make decisions based on historical data and fail to adjust promotion strategies in real time according to member behavior and market changes, resulting in slow response of marketing activities in the face of rapidly changing market environments. Summary of the invention

[0006] In view of the shortcomings of the existing technology, the present invention provides a marketing promotion method based on member churn prediction, which solves the problem that the existing methods usually make decisions based on historical data and fail to adjust the promotion strategy in real time according to member behavior and market changes, resulting in slow response of marketing activities in the face of rapidly changing market environments.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a marketing promotion method based on member churn prediction, comprising: Step 1: Collect members’ behavior data, transaction data, social interaction data, and advertising data; Step 2: preprocessing the data, including data cleaning, feature selection and data standardization; Step 3: Establish a member churn prediction model to predict whether members will churn; Step 4: Establish a promotion channel effect prediction model to evaluate the effect of each promotion channel; Step 5: Calculate the weight value of each promotion channel, where the weight value reflects the cost performance of the promotion channel; Step 6: sorting the promotion channels based on the weight values; Step 7: Adjust the promotion strategy based on the ranking results and optimize the promotion activities by adjusting the budget for different channels; Step 8. Monitor the implementation of the promotion activities and adjust the strategy based on actual feedback results.

[0008] Preferably, the member churn prediction model is established by a deep neural network or an XGBoost model, and the churn prediction formula is: Among them, P 流失 (i) is the probability of loss of the i-th member, σ is the Sigmoid function, w j is the weight of the jth feature, X ij is the value of the i-th member on the j-th feature, and b is the bias term.

[0009] Preferably, the promotion channel effect prediction model is established using a regression model, and the regression formula is: y=β0+β1X1+β2X2+…+β n X n Among them, y is the effect index of the promotion channel, X1, X2,…, X n are the characteristics of the promotion channel, β0,β1,…,β n are the coefficients of the regression model.

[0010] Preferably, the promotion budget allocation formula is: Budgetj =W j ×Total Budget Among them, the budget j is the promotion budget for the jth channel, and the total budget is the budget for the entire marketing campaign.

[0011] Preferably, the optimization promotion strategy is performed based on an A / B test or a multi-armed bandit algorithm, and the A / B test selects the optimal strategy by comparing the effects of different promotion strategies.

[0012] Preferably, the multi-armed bandit algorithm is used to dynamically adjust the delivery ratio of each promotion channel to balance exploration and utilization, ensuring that the delivery strategy is adjusted according to historical data and real-time feedback.

[0013] Preferably, the monitoring step includes real-time monitoring of the promotion effect of each channel, and adjusting the delivery strategy according to indicators such as churn rate and conversion rate.

[0014] Preferably, the weight value W of the promotion channel j The calculation formula is: Among them, W j is the weight value of the jth promotion channel, P 流失 is the predicted probability of member churn corresponding to the jth channel, C 转化率 is the conversion rate of the channel, and the cost j The promotion cost of this channel.

[0015] Preferably, the method further comprises regularly updating the member churn prediction model and the promotion channel effect prediction model to cope with changes in the market and user behavior.

[0016] Preferably, the method adjusts the promotion strategy based on user behavior data and market feedback.

[0017] The present invention provides a marketing promotion method based on member churn prediction. It has the following beneficial effects: 1. Through accurate analysis based on member churn prediction, the present invention can allocate corresponding delivery budgets to various promotion channels, ensuring that resources of different channels are reasonably allocated, so that promotion activities can match target audiences more accurately and avoid waste of resources.

[0018] 2. The present invention automatically predicts member loss and promotion channel effects by using deep learning and regression analysis technologies, and adjusts promotion budgets and strategies based on the results, thereby reducing manual intervention and improving the efficiency and response speed of promotion activities. 3. The present invention can effectively avoid the waste of resources in inefficient channels by predicting and accurately matching the effects of various promotion channels, so that marketing expenses can be used more accurately and effectively, thereby reducing marketing costs, monitoring and feedback mechanisms, and constantly adjusting and optimizing promotion strategies to cope with market changes and changes in user behavior, thereby having strong adaptability and flexibility and being able to adjust in time to cope with different marketing needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The figure is a flow chart of the method 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] Embodiment 1: Please see attached Figure 1 The embodiment of the present invention provides a marketing promotion method based on member churn prediction, comprising: Step 1: Collect members’ behavior data, transaction data, social interaction data, and advertising data; Step 2: Preprocess the data, including data cleaning, feature selection and data standardization; Step 3: Establish a member churn prediction model to predict whether members will churn; Step 4: Establish a promotion channel effect prediction model to evaluate the effect of each promotion channel; Step 5: Calculate the weight value of each promotion channel, which reflects the cost-effectiveness of the promotion channel; Step 6: Sort each promotion channel based on the weight value; Step 7: Adjust the promotion strategy based on the ranking results and optimize the promotion activities by adjusting the budget for different channels; Step 8. Monitor the implementation of the promotion activities and adjust the strategy based on actual feedback results.

[0022] The member churn prediction model is established through a deep neural network or XGBoost model. The churn prediction formula is: Among them, P 流失 (i) is the probability of loss of the i-th member, σ is the Sigmoid function, w j is the weight of the jth feature, X ijis the value of the i-th member on the j-th feature, and b is the bias term.

[0023] The promotion channel effect prediction model is established using a regression model, and the regression formula is: y=β0+β1X1+β2X2+…+β n X n Among them, y is the effect index of the promotion channel, X1, X2,…, X n are the characteristics of the promotion channel, β0,β1,…,β n are the coefficients of the regression model.

[0024] The promotion budget allocation formula is: Budget j =W j ×Total Budget Among them, the budget j is the promotion budget for the jth channel, and the total budget is the budget for the entire marketing campaign.

[0025] The optimization promotion strategy is based on A / B testing or multi-armed bandit algorithm. A / B testing compares the effects of different promotion strategies and selects the best strategy.

[0026] The multi-armed bandit algorithm is used to dynamically adjust the delivery ratio of each promotion channel to balance exploration and utilization, ensuring that the delivery strategy is adjusted based on historical data and real-time feedback.

[0027] The monitoring step includes real-time monitoring of the promotion effect of each channel, and adjusting the delivery strategy based on indicators such as churn rate and conversion rate.

[0028] The weight value W of the promotion channel j The calculation formula is: Among them, W j is the weight value of the jth promotion channel, P 流失 is the predicted probability of member churn corresponding to the jth channel, C 转化率 is the conversion rate of the channel, and the cost j The promotion cost of this channel.

[0029] The method also includes regularly updating the member churn prediction model and the promotion channel effect prediction model to cope with changes in the market and user behavior.

[0030] Method: Adjust promotion strategy based on user behavior data and market feedback.

[0031] In one embodiment, step 1: multi-source data collection This method first obtains a large amount of multi-source data through the data acquisition module. The data sources include but are not limited to: Basic member information: member registration information, account level, historical purchase behavior, member points, etc.

[0032] Member behavior data: members’ browsing, clicking, and searching records on the platform, product click-through rates, page dwell time, etc.

[0033] Transaction data: members’ historical purchase records, order frequency, consumption amount, types of purchased goods, average transaction amount, etc.

[0034] Social interaction data: members’ social interaction behaviors, such as comments, likes, shares, joining groups, etc.

[0035] Advertising data: advertising data for each promotion channel, including the number of ad impressions, clicks, conversion rate, delivery cost, etc.

[0036] User feedback data: User feedback data collected through questionnaires, customer service records, etc., helps to better understand members' needs and satisfaction.

[0037] All data should be collected while ensuring privacy and security, and updated in real time.

[0038] Step 2: Data Preprocessing The collected raw data usually contains missing values, outliers, and duplicate data, which need to be cleaned and preprocessed for subsequent use. Preprocessing includes the following steps: Missing value processing: processing through interpolation methods (such as mean filling, median filling) or deleting missing data records.

[0039] Outlier detection: Use statistical methods (such as Z-Score, IQR) to detect and correct abnormal data to ensure data quality.

[0040] Data standardization: Standardize numerical data so that data of different dimensions have the same scale. A commonly used method is Z-score standardization: Among them, X is the original data, μ is the mean, and σ is the standard deviation.

[0041] Feature engineering: Extract meaningful features from raw data, for example, calculate purchase frequency, consumption amount, average purchase cycle, etc. based on members' transaction data. The L1 regularization (Lasso) algorithm can be used to select the most relevant features.

[0042] Step 3: Construction of member churn prediction model The present invention uses machine learning technology to build a member churn prediction model. According to the data preprocessing results in step 2, select appropriate features and use algorithms such as deep neural network (DNN) or random forest (Random Forest) to perform churn prediction modeling. The steps for building a churn prediction model are as follows: Dataset division: Divide the dataset into training set and test set. Usually 70% of the data is used for training and 30% for validation.

[0043] Model selection: You can choose machine learning algorithms such as deep neural networks, decision trees, support vector machines (SVM), etc. Deep neural networks usually include an input layer, multiple hidden layers, and an output layer.

[0044] Model training: Backpropagation algorithm is used for training. The loss function usually uses the cross-entropy loss function: Among them, y i is the true label of the i-th sample, p i is the predicted churn probability of the sample.

[0045] Model evaluation: AUC, precision, recall and other indicators are used to evaluate the prediction performance of the model.

[0046] Step 4: Construction of promotion channel effect prediction model The effect of promotion channels needs to be estimated through prediction models. The specific approach is to establish a promotion channel effect prediction model based on historical advertising data and basic channel characteristics (such as click-through rate, conversion rate, delivery cost, etc.). We use a linear regression model, and the prediction formula is as follows: y j =β0+β1x1+β2x2+…+β n x n Among them, y j is the effect score of the jth promotion channel, x1,x2,…,x n are the characteristics of the channel, β0,β1,…,β n is the regression coefficient. The regression coefficient is estimated by the least square method to predict the promotion effect.

[0047] Step 5: Calculate the promotion channel weight Based on the member churn prediction results and the effect prediction values ​​of each promotion channel, the weight value of each promotion channel is calculated. The weight calculation formula is as follows: Among them, W jis the weight value of the jth promotion channel, P 流失 is the predicted probability of member churn corresponding to the jth channel, C 转化率 is the conversion rate of the channel, and the cost j is the promotion cost of this channel. The weight value reflects the cost-effectiveness of each channel. A channel with a high weight is more likely to bring a higher conversion rate or lower cost.

[0048] Step 6: Promotion budget allocation and optimization According to the weight value calculated in step 5, allocate the corresponding budget to each promotion channel. The specific budget allocation formula is: budget j =W j ×Total Budget Adjust the amount of money invested in each promotion channel based on the channel's budget and weight to ensure maximum promotion effect.

[0049] In addition, A / B testing and multi-armed bandit algorithms are used to optimize promotion strategies. In A / B testing, members are randomly divided into groups and different channel combinations are used for experiments to evaluate the effects of different strategies and select the best strategy. In the multi-armed bandit algorithm, the budget of each channel is dynamically adjusted in order to obtain the best long-term return.

[0050] Step 7: Real-time monitoring and feedback During the marketing activities, the effects of various promotion channels are monitored in real time, and key indicators such as conversion rate, click-through rate, ROI, etc. are collected in a timely manner. Based on real-time data feedback, the budget and delivery strategy of each channel are dynamically adjusted to ensure that the promotion activities are always in an efficient state.

[0051] Step 8: Model Update and Maintenance In order to ensure the adaptability and long-term effectiveness of the model, the churn prediction model and the promotion channel effect prediction model need to be updated regularly. The update process includes retraining with the latest data and adjusting model hyperparameters.

[0052] Comparative Example 1: Improvement and experimental verification of the "promotion channel weight calculation" step; Original steps description: The present invention adopts the promotion channel weight calculation formula: Among them, W j is the weight value of the jth promotion channel, P 流失 is the predicted probability of member churn corresponding to the jth channel, C 转化率 is the conversion rate of the channel, and the cost j The promotion cost of this channel.

[0053] This formula comprehensively considers the risk of member loss, the effect of promotion channels and the cost of delivery, and provides a quantitative basis for budget allocation. The following will gradually improve the formula, conduct experimental verification, and prove the comprehensive optimality of the solution of the present invention in performance and practical application.

[0054] Improvement plan 1: Introduce click-through rate (CTR) feature.

[0055] Improvements: Try to introduce click-through rate (CTR) to make the weight calculation formula more comprehensively reflect the channel effect: Experimental steps: Data preparation: Collect the following data for each promotion channel: P 流失 : Calculated by the churn prediction model.

[0056] C 转化率 : The actual conversion rate of the promotion channel.

[0057] CTR j : The click-through rate of the ad.

[0058] cost j : The advertising cost of this channel.

[0059] Weight calculation: Use the improved formula to calculate the weight value of each channel.

[0060] Model evaluation: Verify the performance of the improvement plan by comprehensively evaluating the click-through rate, conversion rate and ROI (return on investment) of the promotion activities.

[0061] Table 1 Analysis and conclusion: Performance degradation: After introducing CTR, the F1-Score of the model decreases slightly, indicating that CTR is highly correlated with conversion rate and the marginal contribution of the newly added features to the model is low.

[0062] Increased resource consumption: CTR calculation increases resource consumption by 15%, and has limited effect on optimizing promotion strategies.

[0063] Final conclusion: Improved solution 1 performs poorly in terms of resource consumption and performance improvement, and is inferior to the original solution of the present invention.

[0064] Comparative Example 2: Experimental steps: Data preparation and preprocessing: Collect at least 6 months of member behavior data from e-commerce platforms, social media, etc., including purchase records, browsing history, social interactions, etc.

[0065] Use Python and other data processing tools to preprocess the data, fill in missing values, remove duplicate data, and perform standardization.

[0066] Model training and evaluation: Train multiple machine learning models (logistic regression, random forest, support vector machine, etc.) using historical data and evaluate the accuracy of the models through cross-validation.

[0067] AUC (area under the curve) and F1-score were used as evaluation indicators to select the best model for churn prediction.

[0068] Weight calculation and optimization: Use the trained churn prediction model to output the churn probability of each member under each promotion channel.

[0069] Calculate the weight value of each channel, sort the channels according to the weight, and allocate budget.

[0070] Promotion strategy implementation and effect monitoring: Implement optimized promotion strategies, regularly monitor the conversion effects of various channels, and adjust budget allocation strategies.

[0071] Through real-time monitoring of member behavior data, we analyze promotion effects and adjust strategies in a timely manner to ensure maximum input-output ratio.

[0072] Experimental Results Through comparative experiments, the promotion method of Example 1 is compared with the traditional manual promotion strategy and other optimization schemes. The experimental results are shown in the following table: Table 2 Experimental analysis and conclusion: Accuracy and effect: The first embodiment is comparable to other solutions in terms of accuracy, precision, recall and other indicators, among which the F1-score is slightly higher than that of other improved solutions, indicating that it performs better in balancing precision and recall.

[0073] In practical applications, the improvement of accuracy and F1-score means that member churn prediction is more accurate, which can effectively optimize the promotion strategy and avoid over-delivery or waste of resources.

[0074] Computational complexity: Compared with Improvement Scheme 2 (multi-source data fusion) and Improvement Scheme 3 (deep learning model), Embodiment 1 has the lowest computational complexity and is suitable for real-time prediction of large-scale user data.

[0075] Embodiment 1 can complete data processing and prediction in a relatively short time, and is suitable for marketing activities that require a quick response.

[0076] Training time: The training time of Example 1 is the shortest, only 1 hour, which shows that the solution has higher computational efficiency and can quickly adapt to changing market demands.

[0077] Budget utilization: Since the promotion strategy has been refined and optimized, the budget utilization rate of Example 1 is 98.5%, which is much higher than other solutions, indicating that the promotion funds have been optimally allocated and the return on investment has been improved.

[0078] Conclusion of the best embodiment Taking all indicators into consideration, Example 1 performs best in terms of accuracy, computing efficiency, budget utilization, etc., proving to be the optimal example. Compared with other improved solutions, Example 1 is not only at the same level in terms of model prediction accuracy, but also has significant advantages in computing efficiency and resource utilization. Therefore, Example 1 is determined to be the optimal example of the present invention, and is suitable for marketing and promotion activities with high real-time requirements and large-scale user groups.

[0079] 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 marketing promotion method based on member churn prediction, characterized in that: include: Step 1: Collect members’ behavior data, transaction data, social interaction data, and advertising data; Step 2: preprocessing the data, including data cleaning, feature selection and data standardization; Step 3: Establish a member churn prediction model to predict whether members will churn; Step 4: Establish a promotion channel effect prediction model to evaluate the effect of each promotion channel; Step 5: Calculate the weight value of each promotion channel, where the weight value reflects the cost performance of the promotion channel; Step 6: sorting the promotion channels based on the weight values; Step 7: Adjust the promotion strategy based on the ranking results and optimize the promotion activities by adjusting the budget for different channels; Step 8. Monitor the implementation of the promotion activities and adjust the strategy based on actual feedback results.

2. The marketing promotion method based on member churn prediction according to claim 1, characterized in that: The member churn prediction model is established by a deep neural network or an XGBoost model, and the churn prediction formula is: Among them, P 流失 (i) is the probability of loss of the i-th member, σ is the Sigmoid function, w j is the weight of the jth feature, X ij is the value of the i-th member on the j-th feature, and b is the bias term.

3. The marketing promotion method based on member churn prediction according to claim 1 is characterized in that: The promotion channel effect prediction model is established using a regression model, and the regression formula is: y=β0+β1X1+β2X2+…+β n X n Among them, y is the effect index of the promotion channel, X1, X2,…, X n are the characteristics of the promotion channel, β0,β1,…,β n are the coefficients of the regression model.

4. The marketing promotion method based on member churn prediction according to claim 1 is characterized in that: The promotion budget allocation formula is: Budget j =W j ×Total Budget Among them, the budget j is the promotion budget for the jth channel, and the total budget is the budget for the entire marketing campaign.

5. The marketing promotion method based on member churn prediction according to claim 1 is characterized in that: The optimization promotion strategy is performed based on A / B testing or a multi-armed bandit algorithm. The A / B test selects the optimal strategy by comparing the effects of different promotion strategies.

6. The marketing promotion method based on member churn prediction according to claim 5 is characterized in that: The multi-armed bandit algorithm is used to dynamically adjust the delivery ratio of each promotion channel to balance exploration and utilization, ensuring that the delivery strategy is adjusted based on historical data and real-time feedback.

7. The marketing promotion method based on member churn prediction according to claim 1 is characterized in that: The monitoring step includes real-time monitoring of the promotion effect of each channel and adjusting the delivery strategy according to indicators such as churn rate and conversion rate.

8. The marketing promotion method based on member churn prediction according to claim 1, characterized in that: The weight value W of the promotion channel j The calculation formula is: Among them, W j is the weight value of the jth promotion channel, P 流失 is the predicted probability of member churn corresponding to the jth channel, C 转化率 is the conversion rate of the channel, and the cost j The promotion cost of this channel.

9. The marketing promotion method based on member churn prediction according to claim 1, characterized in that: The method also includes regularly updating the member churn prediction model and the promotion channel effect prediction model to cope with changes in the market and user behavior.

10. The marketing promotion method based on member churn prediction according to claim 1, characterized in that: The method adjusts the promotion strategy according to user behavior data and market feedback.

Images