Customer data management method and system

Abstract

The invention discloses a client data management method and system, and relates to the technical field of client data management, and the method comprises the steps: collecting client data from a plurality of data sources, carrying out the preprocessing, and extracting the time sequence features of the client data; constructing a customer loss model to calculate the loss probability of customers, classifying customer groups according to the loss probability of the customers, and implementing corresponding measures according to a classification result; and constructing a visual interface to display the result of the loss probability of the customer in real time, and storing, collecting and analyzing the generated customer data. The method comprises the following steps: acquiring customer data from multiple data sources, preprocessing the customer data, and extracting time sequence features of the customer data; according to the method, the customer loss model is constructed to calculate the loss probability of the customer, the customer loss prediction accuracy is improved, the definition and accuracy of data are improved, and the problem of processing unstructured and time-varying data in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of customer data management, and in particular to a customer data management method and system. Background Art

[0002] With the rapid development of big data technology and artificial intelligence technology, enterprises have gradually realized the importance of effectively managing and analyzing customer data to enhance market competitiveness. Customer data management technology has gradually developed from the initial simple storage and query to a complex system that comprehensively utilizes multi-source data, covering customer behavior tracking, analysis and prediction. Existing customer data management methods usually rely on simple database query and statistical methods. Although they can support marketing strategies and customer relationship management to a certain extent, with the diversification of customer needs and the increase in data volume, traditional technologies are unable to cope with data integration, preprocessing and effective analysis. In particular, with the diversification of interactive platforms, customer data presents characteristics such as unstructured and highly time-varying. Existing technologies often lack accuracy and flexibility in processing these complex data, and are unable to fully explore the potential information behind the data, resulting in inaccurate customer churn predictions, which in turn affects the company's market decisions.

[0003] Existing customer data management technologies have many deficiencies. Data preprocessing methods are not perfect. Conventional denoising and standardization methods have difficulty processing customer data of different dimensions and different data sources, resulting in low data quality. Traditional customer churn prediction models often use simpler statistical methods such as linear regression or logistic regression. These models cannot capture the nonlinear relationship of customer behavior, resulting in low accuracy of churn prediction. Summary of the invention

[0004] In view of the problems existing in the above-mentioned existing customer data management methods and systems, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is that the data preprocessing method is not perfect enough. Conventional denoising and standardization methods are difficult to process customer data of different dimensions and different data sources, resulting in low data quality. Traditional customer churn prediction models often use simpler statistical methods such as linear regression or logistic regression. These models cannot capture the nonlinear relationship of customer behavior, resulting in low accuracy of churn prediction.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a customer data management method, which includes collecting customer data from multiple data sources and preprocessing it, extracting time series characteristics of the customer data; building a customer churn model to calculate the customer churn probability, classifying customer groups according to the customer churn probability and implementing corresponding measures according to the classification results; building a visual interface to display the results of the customer churn probability in real time, and storing the customer data generated by the collection and analysis.

[0007] As a preferred solution of the customer data management method of the present invention, wherein: the collecting of customer data from multiple data sources and preprocessing thereof refers to using a database query tool to collect customer transaction data from a transaction database, including transaction time, collecting customer interaction data through a platform's click log recording system, including customer click times and page display times, and collecting customer feedback rating data through the enterprise's feedback platform;

[0008] Use the quartile method (IQR) to identify and remove outliers in customer data;

[0009] De-duplicate customer data using unique identifiers;

[0010] Use Z-score standardization to standardize customer data

[0011] Use wavelet transform to decompose the annotated customer data and get the wavelet transform coefficient W ψ (f)(a,b);

[0012] Calculate the wavelet transform coefficients W ψ (f) Signal energy E(a,b) of (a,b);

[0013] Based on the calculated signal energy E(a,b), the wavelet transform coefficients W ψ (f)(a,b) is normalized to obtain the normalized time-frequency component The formula is:

[0014]

[0015] Use inverse wavelet transform to transform the normalized time-frequency components Restore to the denoised time series signal and obtain the denoised customer data

[0016] As a preferred solution of the customer data management method of the present invention, wherein: the time series feature of extracting customer data refers to using the purchase frequency formula to calculate the customer's purchase frequency F at time t pur (t);

[0017] Use the click-through rate formula to calculate the customer's click-through rate F at time t CTR (t);

[0018] Calculate the average customer satisfaction score F at time t feed (t);

[0019] The purchase frequency F will be calculated pur (t), click rate F CTR (t) and the average score of feedback satisfaction F feed (t), construct the time series feature F(t), the formula is:

[0020] F(t)=[F pur (t),F CTR (t),F feed (t)],

[0021] The time series feature F(t) is normalized using the minimum-maximum normalization method.

[0022] As a preferred solution of the customer data management method of the present invention, wherein: the construction of the customer churn model to calculate the customer churn probability refers to calculating the LSTM hidden state by combining the convolution operation, and the formula is:

[0023]

[0024] where h t is the hidden state at time t, σ is the tanh activation function, h t-1 is the hidden state at time t-1, W h is the weight matrix of the LSTM hidden state, F'(t) is the normalized time series feature at time t, F'(ti) is the normalized time series feature at time ti, b h is the bias term of the LSTM hidden state, k is the size of the time window, ω i is the convolution kernel, f(·) is the ReLU activation function;

[0025] Use the nonlinear function tanh to calculate the attention score g t , the formula is:

[0026] g t =tanh(W g ·h t +b g ),

[0027] Where W g is the weight matrix of the attention mechanism, b g is the bias term of the attention mechanism;

[0028] Use the attention mechanism to calculate the attention weight α at time t t , the formula is:

[0029]

[0030] Where T is the total time to collect customer data, g k is the attention score at time k;

[0031] Use the weighted sum formula to combine the LSTM hidden state and the attention weight to calculate the weighted sum o;

[0032] Based on the weighted sum o, a customer churn probability model is constructed, and the formula is:

[0033]

[0034] Where P churn is the probability of customer churn, W o is the weight matrix of the output layer, b o is the bias term of the output layer, η is the sigmoid activation function;

[0035] Bring the normalized time series features into the customer churn probability model to obtain the probability of customer churn P churn .

[0036] As a preferred solution of the customer data management method of the present invention, wherein: the classification of customer groups according to the probability of customer churn and implementation of corresponding measures according to the classification results refers to collecting customer data with historical tags to set a judgment threshold;

[0037] The probability of customer churn is P churn Compared with the judgment threshold, if the probability of customer churn P churn If the probability of customer loss is greater than or equal to the judgment threshold, it is judged as a high-risk customer and an active retention strategy is adopted, including providing one-to-one exclusive customer service and sending warm reminders. churn When the amount is less than the judgment threshold, the customer is judged as a low-risk customer and incentive measures are adopted, including providing limited-time rewards and pushing regular discount information.

[0038] As a preferred solution of the customer data management method described in the present invention, wherein: the result of constructing a visual interface to display the customer churn probability in real time refers to using a front-end framework to develop a visual interface, using a data visualization tool to display the customer churn probability, performing a review operation on verified user requests, and displaying corresponding measures on the interface, and using charts and text boxes to display the customer churn probability change record in real time.

[0039] As a preferred solution of the customer data management method described in the present invention, wherein: the storage of customer data generated by collection and analysis refers to storing the collected customer data and the customer churn probability generated by the analysis in a database, sorting the database in chronological order and marking corresponding tags, synchronously backing up the collected customer data and the customer churn probability generated by the analysis in the cloud, and regularly performing integrity checks on the backup data.

[0040] Another object of the present invention is to provide a customer data management system, which comprises:

[0041] The collection and extraction module is used to collect customer data from various data sources and pre-process it to extract the time series characteristics of the customer data;

[0042] Build a classification module to build a customer churn model to calculate the customer churn probability, classify customer groups according to the customer churn probability, and implement corresponding measures based on the classification results;

[0043] The visualization storage module is used to build a visualization interface to display the results of customer churn probability in real time and store the customer data generated by collection and analysis.

[0044] A computer device comprises: a memory and a processor; the memory stores a computer program, and the processor implements the steps of the above-mentioned customer data management method when executing the computer program.

[0045] A computer-readable storage medium stores a computer program, which implements the steps of the above-mentioned customer data management method when executed by a processor.

[0046] The beneficial effects of the present invention are as follows: the present invention collects customer data from multiple data sources and performs preprocessing to extract the time series characteristics of the customer data; a customer churn model is constructed to calculate the customer churn probability, thereby improving the accuracy of customer churn prediction, improving the clarity and accuracy of data, and solving the problem of processing unstructured and time-varying data in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0048] Figure 1 A flowchart of the customer data management method.

[0049] Figure 2 This is a structural diagram of the customer data management system. DETAILED DESCRIPTION

[0050] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0051] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0052] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0053] Example 1, reference Figure 1 , which is the first embodiment of the present invention, and provides a customer data management method, the customer data management method comprising:

[0054] S1. Collect customer data from multiple data sources and preprocess them to extract the time series features of customer data;

[0055] Specifically, collecting customer data from multiple data sources and preprocessing it means using database query tools to collect customer transaction data from the transaction database, including transaction time, collecting customer interaction data through the platform's click log recording system, including customer clicks and page display times, and collecting customer feedback rating data through the company's feedback platform;

[0056] Use the quartile method (IQR) to identify and remove outliers in customer data;

[0057] De-duplicate customer data using unique identifiers;

[0058] Use Z-score standardization to standardize customer data

[0059] Use wavelet transform to decompose the annotated customer data. The formula is:

[0060]

[0061] Where W ψ(f)(a,b) is the wavelet transform coefficient of the customer data at scale a and time position b, ∞ and -∞ are the upper and lower limits of the integral respectively, ψ(x) is the mother wavelet function, which is used to decompose and reconstruct the signal, a is the scale of the wavelet transform, which determines the resolution of the wavelet in frequency, b is the time position of the wavelet transform, which determines the localization of the signal in the time dimension, and f(x) is the customer data at the time variable x;

[0062] The mother wavelet function ψ(r) is formulated as follows:

[0063]

[0064] Where r is Result;

[0065] Calculate the wavelet transform coefficients W ψ The signal energy E(a,b) of (f)(a,b) is:

[0066]

[0067] Based on the calculated signal energy E(a,b), the wavelet transform coefficients W ψ (f)(a,b) is normalized to obtain the normalized time-frequency component The formula is:

[0068]

[0069] Signal energy normalization processing automatically adjusts the denoising strength of the signal and normalizes it according to the energy size of each signal component. Signal components with large energy are retained first, and noise components with small energy are weakened. The adaptive denoising method avoids the inaccuracy of the preset threshold in the fixed threshold method and is the best method for handling dynamic and complex noise environments. Through energy normalization, the model can dynamically judge the energy of each part of the signal, retain high-energy signal components, and retain low-energy but important detail signals through energy normalization processing, effectively avoiding the damage to signal details by traditional threshold denoising methods. It is a more reasonable and accurate choice. Energy normalization processing automatically adapts to different types of signals and noise environments by analyzing the energy distribution of the signal. It no longer relies on manually setting thresholds or customizing for specific types of signals. This method is universal and can handle a variety of For signals and complex noise environments, energy normalization processing dynamically normalizes the signal through its own energy characteristics. There is no need to manually set the threshold, which eliminates the difficulty of threshold selection and can stably cope with the noise removal needs in different signal environments. The automated noise processing method is more robust, ensuring the adaptability and accuracy of noise processing. Signal energy normalization processing can automatically adjust the denoising intensity according to the energy of the signal at different time frequencies, avoiding the problem of poor denoising caused by inaccurate threshold setting, ensuring that high-energy signal components are retained and low-energy noise is naturally weakened. There is no need to manually set the threshold. The energy normalization method performs adaptive denoising through the energy characteristics of the signal, which is suitable for different signals and dynamic noise environments, and automatically adjusts the denoising intensity, making the denoising effect smoother and more stable, especially in high-noise or complex signal scenarios.

[0070] Use inverse wavelet transform to transform the normalized time-frequency components Restored to the denoised time series signal, the formula is:

[0071]

[0072] in is the denoised customer data, A is the scale range, which controls the maximum scale of signal decomposition, and B is the time position range, which controls the localization of the mother wavelet on the time axis.

[0073] The diversity of data sources ensures a comprehensive portrait of customers, which can not only capture customers' purchasing behavior, but also understand their interactive behavior and feedback attitude. This data collection method can provide a rich source of information for subsequent data analysis and churn prediction, and improve the accuracy of the model. By using the IQR method to identify and delete these outliers, the quality of the data can be significantly improved, noise interference can be reduced, and subsequent analysis can be more accurate and reliable. Deduplication through unique identifiers ensures that each customer occupies only one position in the data set. This operation not only ensures data consistency, but also avoids the interference of duplicate data on the analysis results, improves the authenticity of the data, and standardized processing helps improve machine learning. The convergence speed of the model can be improved, and some dimensions can be prevented from affecting the model training due to excessive or small values. Through the inverse wavelet transform, the time-frequency components can be restored to time domain signals, making the denoised data more suitable for subsequent modeling and analysis. The wavelet transform makes data processing more flexible and can deal with the non-stationary problems in customer behavior. The signal energy of the wavelet transform coefficients is calculated to capture the fluctuation range of customer data in different time periods. Through normalization processing, the data between different time periods and different customers are scaled to ensure that these data can be input into the model under the same conditions, eliminating the influence of individual differences and data magnitude, so that the model can make predictions more fairly and accurately.

[0074] Furthermore, extracting the time series features of customer data means using the purchase frequency formula to calculate the customer's purchase frequency F at time t pur (t), the formula is:

[0075]

[0076] Where n t is the number of purchases made by the customer at time t, and T is the total number of days in the time period;

[0077] Use the click-through rate formula to calculate the customer's click-through rate F at time t CTR (t), the formula is:

[0078]

[0079] where q t is the number of clicks by the customer at time t, Q t is the number of page impressions for the customer at time t;

[0080] Calculate the average customer satisfaction score F at time t feed (t), the formula is:

[0081]

[0082] Where R iis the customer's rating in the i-th feedback, m t The total number of times the customer's feedback rating is given at time t;

[0083] The purchase frequency F will be calculated pur (t), click rate F CTR (t) and the average score of feedback satisfaction F feed (t), construct the time series feature F(t), the formula is:

[0084]

[0085] The time series feature F(t) is normalized using the minimum-maximum normalization method.

[0086] The calculation of purchase frequency can identify the difference between active users and potential churn users, and provide accurate feature input for subsequent models. The dynamic changes in purchase frequency can also help discover trends in customer behavior. For example, a gradual decrease in customer purchase behavior may indicate churn risk. Click rate, as a time series feature, can reveal fluctuations in customer behavior. For example, a sharp drop in click rate in a short period of time may mean a decline in customer interest. It can provide accurate data input for the model and help improve the accuracy of customer churn prediction. The calculation of feedback satisfaction can identify dissatisfied customers or customers who are about to churn, because a significant drop in feedback satisfaction is usually strongly correlated with customer churn. Feedback satisfaction, as a time series feature, provides the model with a favorable reference for the customer's emotional state. Combining these features, the model can identify periodic changes or abnormal fluctuations in behavior. The construction of time series features helps to fully reflect the customer's behavior trajectory and provides a more accurate data basis for subsequent churn prediction. The minimum-maximum normalization process avoids the bias caused by the large values ​​of some features during the training process, so that all features can equally affect the output of the model, improving the stability and prediction ability of the model.

[0087] S2. Build a customer churn model to calculate the customer churn probability, classify customer groups according to the customer churn probability, and implement corresponding measures based on the classification results;

[0088] Specifically, building a customer churn model to calculate the customer churn probability means calculating the LSTM hidden state by combining convolution operations. The formula is:

[0089]

[0090] where h t is the hidden state at time t, σ is the tanh activation function, h t-1 is the hidden state at time t-1, W his the weight matrix of the LSTM hidden state, F'(t) is the normalized time series feature at time t, F'(ti) is the normalized time series feature at time ti, b h is the bias term of the LSTM hidden state, k is the size of the time window, ω i is the convolution kernel, f(·) is the ReLU activation function;

[0091] LSTM can capture long-term dependencies through its memory units, but it cannot accurately capture local, rapidly changing short-term behavioral fluctuations. Local features are particularly important in customer churn prediction, because customer short-term behaviors such as sudden purchases or clicks may directly affect the risk of churn. Convolution operations are good at extracting local information, especially in image recognition and processing time series. They can scan local data through convolution kernels to capture subtle changes. Adding convolution operations can supplement the shortcomings of LSTM, so that the model can capture long-term behavioral characteristics and flexibly respond to short-term fluctuations. Other RNN structures are similar to LSTM and can capture dependencies in long time series, but are not sensitive to local, short-term behavioral changes. If only LSTM or other RNN methods are used, it is difficult for the model to accurately capture short-term fluctuations in customer behavior that are critical to churn prediction. Simple smoothing or mean processing only focuses on the overall trend. , unable to respond effectively to local, instantaneous behavioral changes, which will cause important short-term changes to be ignored. By introducing convolution kernel operations in LSTM, the model can capture long-term dependencies and short-term fluctuations at the same time, thereby improving the ability to process customer data, especially for churn prediction. The convolution kernel can identify short-term behavioral anomalies, thereby improving the accuracy of prediction. The convolution operation extracts local features of the previous k time periods through a local sliding window, which can effectively handle drastic changes in customer behavior in a short period of time, allowing the model to respond quickly to sudden customer behavior and improve the model's sensitivity to critical moments. The convolution operation reduces the computational complexity and number of parameters through a parameter sharing mechanism, which can not only ensure efficient capture of short-term changes, but also reduce the complexity of the model. Compared with a simple fully connected layer or directly modeling each time independently, the introduction of convolution can make the model more concise and efficient when dealing with short-term changes.

[0092] The tanh activation function σ is formulated as:

[0093]

[0094] Where p is W h ·[h t-1 ,F'(t)]+b h Result;

[0095] The ReLU activation function f(·) is:

[0096] f(u)=max(0,u),

[0097] Where u is ω i *Result of F(ti);

[0098] Use the nonlinear function tanh to calculate the attention score g t , the formula is:

[0099] g t =tanh(W g ·h t +b g ),

[0100] Where W g is the weight matrix of the attention mechanism, b g is the bias term of the attention mechanism;

[0101] Use the attention mechanism to calculate the attention weight α at time t t , the formula is:

[0102]

[0103] Where T is the total time to collect customer data, g k is the attention score at time k;

[0104] The introduction of square terms and fractional smoothing terms strengthens the capture of high-weight time and suppresses the impact of unimportant time on the results. The optimized formula allows the model to be more sensitive to key moments in customer behavior while preventing the extreme impact of a single time. By squaring the attention score, the weight of important time is further amplified, making these times account for a larger proportion in the final decision of the model, which enables the model to focus on key behaviors in churn prediction. The fractional smoothing term effectively controls the impact of extreme weights. The smoothing term prevents the attention weight from rising or falling sharply, ensuring the stability of the model prediction results, especially when the difference between key time and secondary time is large. The standard The softmax only performs linear normalization on the attention score, which is unable to amplify the weight of important time in complex scenarios and control the extreme value of the weight. Fractional smoothing is used to adjust the size of the attention score, which is more flexible and targeted than the simple regularization method. Especially when processing complex behavioral time series data, the optimized formula combines nonlinearity and fractional smoothing to make the attention weight allocation more flexible and intelligent. It can adjust the weight according to the importance of each time, so that important time is amplified and minor time is suppressed. This enhanced discrimination makes the model more accurate in processing customer data and improves the accuracy and performance of the overall prediction.

[0105] Use the weighted sum formula to combine the LSTM hidden state and the attention weight to calculate the weighted sum o, the formula is:

[0106]

[0107] Based on the weighted sum o, a customer churn probability model is constructed, and the formula is:

[0108]

[0109] Where P churn is the probability of customer churn, W o is the weight matrix of the output layer, b o is the bias term of the output layer, η is the sigmoid activation function;

[0110] By squaring the weights of all times and introducing them into the denominator for smoothing, we ensure that the weight changes will not be too large. Other methods such as direct regularization cannot achieve the same effect, because the square sum not only adjusts the size of attention, but also effectively balances the differences between weights to prevent the weights at a certain time from dominating. Without introducing smoothing, the standard weighted sum formula is easily dominated by some extreme weights. Directly using softmax or normalizing the attention score method cannot provide sufficient control to avoid extreme values ​​affecting the model's prediction results. Although traditional L2 regularization can control weights to a certain extent, it lacks specificity. In contrast, fractional smoothing can more effectively adjust by processing the square sum of attention weights. The fluctuation of attention weights ensures smooth output results. The fluctuation range of weights is effectively controlled by introducing fractional smoothing terms to prevent the attention weights at a single time from excessively dominating the final results. This improvement ensures that the model prediction results are more stable and reasonable, especially when extreme scores appear at certain times. While ensuring that high-weighted times are given sufficient importance, their excessive impact on the overall results is suppressed. This control is relatively rare in existing technologies and greatly improves the performance of the model when facing extreme attention weights. The attention weights are controlled through fractional smoothing, and the final probability output is ensured to be in a reasonable range. In contrast, other methods cannot take into account both smoothness and output probability at the same time, especially when the attention weights fluctuate greatly, which can easily lead to loss of control or inaccurate predictions.

[0111] The sigmoid activation function η is formulated as:

[0112]

[0113] Where w is Result;

[0114] Bring the normalized time series features into the customer churn probability model to obtain the probability of customer churn P churn .

[0115] The LSTM hidden state itself is used to memorize the customer's historical behavior, while the convolution operation can capture subtle changes in the data, especially the pattern of local behavior. Through this combination, the present invention can deeply mine customer data, effectively improve the accuracy of churn prediction, and improve the model's ability to capture long-term dependencies. At the same time, local features are extracted through convolution operations, further enriching the input information. The tanh function can map the input to between [-1,1] to help the model process complex nonlinear relationships. The ReLU activation function is used for nonlinear processing after the convolution operation, which can effectively alleviate the gradient vanishing problem and has good sparse activation characteristics, making the model more efficient when processing high-dimensional data. By combining tanh and The ReLU activation function can handle complex time series features while improving the computational efficiency and learning ability of the model. The attention mechanism ensures that the model can focus on the customer behavior that contributes most to the prediction results by calculating the attention weights at each time point, making the model more flexible and intelligent, and can dynamically adjust the weights to capture key features in customer behavior and improve prediction accuracy. Through this weighted method, the model can better balance the behavioral characteristics at different time points, and the weighted sum gives the model a clearer understanding of the time dependency of customer behavior. The weighted sum improves the comprehensive understanding of historical behavior, and the sigmoid function outputs a clear and definite probability of churn, providing companies with a reliable basis for customer management.

[0116] Further, classifying customer groups according to their churn probability and implementing corresponding measures according to the classification results refers to collecting customer data with historical tags to set judgment thresholds;

[0117] The probability of customer churn is P churn Compared with the judgment threshold, if the probability of customer churn P churn If the probability of customer loss is greater than or equal to the judgment threshold, it is judged as a high-risk customer and an active retention strategy is adopted, including providing one-to-one exclusive customer service and sending warm reminders. churn When the amount is less than the judgment threshold, the customer is judged as a low-risk customer and incentive measures are adopted, including providing limited-time rewards and pushing regular discount information.

[0118] The classification step can help companies accurately divide customer groups and take different management measures for high-risk and low-risk customers, so that companies can adjust resource allocation in a targeted manner and avoid a one-size-fits-all management approach for all customers, thereby improving customer management efficiency. One-to-one exclusive customer service can provide customers with more personalized and timely services, solve problems that customers may encounter, and respond to their needs quickly. Sending reminders reminds customers of their account activities or unfinished transactions through soft communication methods, prompting them to re-participate in platform activities, which can significantly reduce the probability of loss of high-risk customers, extend the customer life cycle, and bring higher customer value to the company. Limited-time rewards create a sense of urgency, prompting customers to increase their purchasing behavior or participate in activities in a short period of time, thereby increasing customers' purchase frequency. Pushing regular discount information maintains continuous interaction between customers and the platform, maintains long-term customer loyalty, and continuously maintains customer activity, further consolidating customer loyalty to the platform. It allows companies to dynamically adjust strategies based on the actual situation of customers and further optimize customer experience.

[0119] S3, build a visual interface to display the results of customer churn probability in real time, and store the customer data generated by collection and analysis;

[0120] Specifically, building a visual interface to display the results of customer churn probability in real time means using a front-end framework to develop a visual interface, using data visualization tools to display the probability of customer churn, reviewing verified user requests, and displaying corresponding measures on the interface, and using charts and text boxes to display customer churn probability change records in real time.

[0121] It enables developers to quickly build highly interactive and high-performance visual interfaces. Through the front-end framework, developers can easily integrate data visualization tools and ensure the responsiveness and stability of the interface. Customer managers are provided with a smooth and intuitive platform that enables them to view customer churn risks in real time while ensuring the maintainability and scalability of the system. By using chart tools such as line charts, pie charts or bar charts, users can clearly see how the churn risk of customer groups changes over time or specific behaviors. Intuitive chart displays can help corporate decision makers quickly identify high-risk customer groups and formulate corresponding retention strategies, so that customer churn data is no longer an abstract number, but is presented in a visual form, greatly improving the efficiency and accuracy of data interpretation. The user authority management mechanism not only protects the privacy and security of customer data, but also ensures that users of different levels can only access data related to their responsibilities, enhances data security and access control, and ensures that sensitive customer information is not accessed by unauthorized personnel at will. At the same time, it also makes the internal customer management process of the enterprise more standardized and transparent. Real-time display can help enterprises understand the changing trend of customer churn risks in a timely manner, provide a basis for rapid adjustment and response, and make customer management more timely and proactive. Enterprises can make more accurate decisions based on real-time data, thereby effectively reducing customer churn rates.

[0122] Furthermore, storing the customer data generated by collection and analysis means storing the collected customer data and the customer churn probability generated by the analysis in a database, sorting the database in chronological order and marking corresponding tags, synchronously backing up the collected customer data and the customer churn probability generated by the analysis in the cloud, and regularly performing integrity checks on the backup data.

[0123] Storing customer data and churn probability in a unified database not only ensures centralized data management, but also supports subsequent data retrieval and analysis. Unified data storage can provide enterprises with a comprehensive customer portrait and provide basic support for subsequent customer classification, churn prediction and behavior analysis. By sorting customer data in chronological order, enterprises can clearly see changes in customer behavior over time, thereby capturing potential churn signals. Time sorting can help enterprises more accurately understand the time dependency of customer behavior and enhance the depth and accuracy of data analysis. By marking each piece of data with a corresponding label, such as behavior type, time, customer group, etc., enterprises can quickly filter out customer data with specific characteristics. Labeled management makes data classification more flexible and efficient, and can support multi-dimensional data analysis and rapid retrieval, improving the data utilization efficiency of enterprises. By synchronously backing up customer data and churn probability to the cloud, enterprises can prevent data loss caused by hardware failure, database damage or other unforeseen accidents, greatly enhancing data security and reliability, and ensuring business continuity in the event of a data disaster.

[0124] Example 2, reference Figure 2 , which is a second embodiment of the present invention, and which is different from the previous embodiment, provides a customer data management system, including:

[0125] The collection and extraction module is used to collect customer data from various data sources and pre-process it to extract the time series characteristics of the customer data;

[0126] Build a classification module to build a customer churn model to calculate the customer churn probability, classify customer groups according to the customer churn probability, and implement corresponding measures based on the classification results;

[0127] The visualization storage module is used to build a visualization interface to display the results of customer churn probability in real time and store the customer data generated by collection and analysis.

[0128] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0129] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0130] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0131] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

Claims

1. A method for managing customer data, Features: include, Collect customer data from multiple data sources and pre-process it to extract the time series features of customer data; Build a customer churn model to calculate the customer churn probability, classify customer groups according to the customer churn probability and implement corresponding measures based on the classification results; Build a visual interface to display the results of customer churn probability in real time, and store the customer data generated by collection and analysis.

2. The customer data management method according to claim 1, Features: The said collecting customer data from multiple data sources and preprocessing means using a database query tool to collect customer transaction data from a transaction database, including transaction time, collecting customer interaction data through the platform's click log recording system, including customer click times and page display times, and collecting customer feedback rating data through the enterprise's feedback platform; Use the quartile method (IQR) to identify and remove outliers in customer data; De-duplicate customer data using unique identifiers; Use Z-score standardization to standardize customer data Use wavelet transform to decompose the annotated customer data and get the wavelet transform coefficient W ψ (f)(a,b); Calculate the wavelet transform coefficients W ψ (f) Signal energy E(a,b) of (a,b); Based on the calculated signal energy E(a,b), the wavelet transform coefficients W ψ (f)(a,b) is normalized to obtain the normalized time-frequency component The formula is: Use inverse wavelet transform to transform the normalized time-frequency components Restore to the denoised time series signal and obtain the denoised customer data 3. The customer data management method according to claim 2, Features: The time series feature of extracting customer data refers to using the purchase frequency formula to calculate the customer's purchase frequency F at time t pur (t); Use the click-through rate formula to calculate the customer's click-through rate F at time t CTR (t); Calculate the average customer satisfaction score F at time t feed (t); The purchase frequency F will be calculated pur (t), click rate F CTR (t) and the average score of feedback satisfaction F feed (t), construct the time series feature F(t), the formula is: The time series feature F(t) is normalized using the minimum-maximum normalization method.

4. The customer data management method according to claim 3, Features: The construction of the customer churn model to calculate the customer churn probability refers to calculating the LSTM hidden state by combining the convolution operation, and the formula is: where h t is the hidden state at time t, σ is the tanh activation function, h t-1 is the hidden state at time t-1, W h is the weight matrix of the LSTM hidden state, F'(t) is the normalized time series feature at time t, F'(ti) is the normalized time series feature at time ti, b h is the bias term of the LSTM hidden state, k is the size of the time window, ω i is the convolution kernel, f(·) is the ReLU activation function; Use the nonlinear function tanh to calculate the attention score g t , the formula is: g t =tanh(W g ·h t +b g ), Where W g is the weight matrix of the attention mechanism, b g is the bias term of the attention mechanism; Use the attention mechanism to calculate the attention weight α at time t t , the formula is: Where T is the total time to collect customer data, g k is the attention score at time k; Use the weighted sum formula to combine the LSTM hidden state and the attention weight to calculate the weighted sum o; Based on the weighted sum o, a customer churn probability model is constructed, and the formula is: Where P churn is the probability of customer churn, W o is the weight matrix of the output layer, b o is the bias term of the output layer, η is the sigmoid activation function; Bring the normalized time series features into the customer churn probability model to obtain the probability of customer churn P churn .

5. The customer data management method according to claim 4, Features: The classification of customer groups according to the probability of customer churn and implementation of corresponding measures according to the classification results refers to collecting customer data with historical tags and setting judgment thresholds; The probability of customer churn is P churn Compared with the judgment threshold, if the probability of customer churn P churn If the probability of customer loss is greater than or equal to the judgment threshold, it is judged as a high-risk customer and an active retention strategy is adopted, including providing one-to-one exclusive customer service and sending warm reminders. churn When the amount is less than the judgment threshold, the customer is judged as a low-risk customer and incentive measures are adopted, including providing limited-time rewards and pushing regular discount information.

6. The customer data management method according to claim 5, Features: The construction of a visualization interface to display the result of the customer churn probability in real time refers to using a front-end framework to develop a visualization interface, using data visualization tools to display the customer churn probability, reviewing verified user requests, and displaying corresponding measures on the interface, and using charts and text boxes to display customer churn probability change records in real time.

7. The customer data management method according to claim 6, Features: The storing of customer data generated by collection and analysis refers to storing the collected customer data and the customer churn probability generated by the analysis in a database, sorting the database in chronological order and marking corresponding tags, synchronously backing up the collected customer data and the customer churn probability generated by the analysis in the cloud, and regularly performing integrity checks on the backup data.

8. A customer data management system based on the customer data management method according to any one of claims 1 to 7, Features: include, The collection and extraction module is used to collect customer data from various data sources and pre-process it to extract the time series characteristics of the customer data; Build a classification module to build a customer churn model to calculate the customer churn probability, classify customer groups according to the customer churn probability, and implement corresponding measures based on the classification results; The visualization storage module is used to build a visualization interface to display the results of customer churn probability in real time and store the customer data generated by collection and analysis.

9. A computer device, include: A memory and a processor; the memory stores a computer program, wherein the processor implements the steps of the customer data management method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, Features: When the computer program is executed by a processor, the steps of the customer data management method according to any one of claims 1 to 7 are implemented.

Images