Enterprise customer loss prediction method and system based on big data

Abstract

The invention discloses an enterprise customer loss prediction method and system based on big data. The method comprises the steps of data acquisition, data preprocessing, customer feature aggregation, clustering processing and enterprise customer loss prediction. The invention relates to the technical field of enterprise customer loss prediction, in particular to an enterprise customer loss prediction method and system based on big data, and the method comprises the steps: obtaining customer behavior data through data collection; a data preprocessing method of data cleaning, feature coding and data normalization is adopted; a method of aggregating customer features by adopting a self-organizing neural network model and combining with a K-means clustering model to perform preliminary classification is adopted, so that more representative aggregated features are obtained and classified, and the accuracy of the model is improved; according to the method for predicting enterprise customer loss by adopting the attention graph convolutional network model, the prediction precision and reliability of the model are improved by constructing graph structure data based on customer interaction records and dynamically adjusting the weight in combination with an attention mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of enterprise customer churn prediction, and in particular to an enterprise customer churn prediction method and system based on big data. Background Art

[0002] Enterprise customer churn prediction is to predict the probability that a customer may leave the enterprise through data analysis and modeling technology. This prediction uses information such as customers' historical behavior, purchasing patterns and interaction records to identify churn risks; it helps enterprises identify customers who may churn in advance, so as to take appropriate retention measures, reduce revenue losses, improve customer satisfaction, and thus enhance corporate competitiveness.

[0003] However, traditional enterprise customer churn prediction methods have the technical problem of relying on single customer behavior data for analysis, which makes it difficult to deeply explore the overall characteristics of enterprise customer churn and is easily affected by environmental fluctuations, resulting in low model prediction accuracy; traditional enterprise customer churn prediction methods have the technical problem of only considering the behavior data of individual customers and ignoring the correlation between enterprise customers, and are unable to fully capture the potential churn risks brought about by customer correlation, thereby reducing the accuracy and reliability of the model. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method and system for predicting enterprise customer churn based on big data. The traditional enterprise customer churn prediction method relies on single customer behavior data for analysis, is difficult to deeply explore the overall characteristics of enterprise customer churn and is easily affected by environmental fluctuations, resulting in low model prediction accuracy. The present invention creatively adopts a self-organizing neural network model to perform a customer feature aggregation method, aggregates a single feature that is difficult to accurately reflect customer churn, obtains more representative aggregated features, and improves the accuracy of the model; the traditional enterprise customer churn prediction method only considers the behavior data of individual customers, but ignores the correlation between enterprise customers, and cannot fully capture the potential churn risk brought by customer correlation, which reduces the accuracy and reliability of the model. The present invention creatively adopts an attention graph convolutional network model to predict enterprise customer churn, and improves the prediction accuracy and reliability of the model by constructing graph structure data based on customer interaction records and dynamically adjusting weights in combination with the attention mechanism.

[0005] The technical solution adopted by the present invention is as follows: The method for predicting enterprise customer churn based on big data provided by the present invention comprises the following steps:

[0006] Step S1: data collection;

[0007] Step S2: data preprocessing;

[0008] Step S3: customer feature aggregation;

[0009] Step S4: clustering processing;

[0010] Step S5: Enterprise customer churn prediction.

[0011] Furthermore, in step S1, the data collection is used to collect customer behavior data required for enterprise customer churn prediction, specifically to obtain a customer behavior data set through collection, and the customer behavior data set specifically includes basic customer information, transaction records, customer interaction records, customer product usage data, customer feedback data and customer activity data.

[0012] Furthermore, in step S2, the data preprocessing is used to preprocess the customer behavior data, and specifically includes the following steps:

[0013] Step S21: data cleaning, which is used to clean the customer behavior data, specifically removing missing values ​​and duplicate values ​​from the customer behavior data set to obtain a preliminary processed data set;

[0014] Step S22: feature encoding, used to encode the preliminary processed data, specifically, using a one-hot encoding method to encode the preliminary processed data set to obtain an encoded data set;

[0015] Step S23: data normalization, which is used to normalize the coded data, specifically using a minimum-maximum normalization method to normalize the coded data set to obtain a customer feature set;

[0016] Step S24: preprocessing, specifically, performing data preprocessing on the customer behavior data set through the data cleaning, feature encoding and data normalization to obtain a customer feature set.

[0017] Further, in step S3, the customer feature aggregation is used to aggregate customer features, specifically, the customer feature set is processed using a self-organizing neural network model to obtain an aggregated customer churn feature set;

[0018] The customer behavior feature aggregation specifically includes the following steps:

[0019] Step S31: Initialize weights, which are used to randomly initialize the self-organizing neural network weight vector, as shown below:

[0020] ;

[0021] In the formula, represents the weight vector of the i-th node, Represents the random initialization of the weight vector of the i-th node from a uniform distribution [0,1] ;

[0022] Step S32: Find the optimal node. The optimal node is specifically the point with the smallest Euclidean distance to the input feature. The formula used to calculate the Euclidean distance is as follows:

[0023] ;

[0024] In the formula, represents the Euclidean distance function, represents the jth feature of the customer, represents the Euclidean distance;

[0025] Step S33: Update the optimal node weight, the formula used is as follows:

[0026] ;

[0027] In the formula, It represents the neighborhood function between the optimal node and the lth node corresponding to the jth feature of the customer, Indicates the position vector of the optimal node corresponding to the jth feature of the customer, represents the position vector of the lth node, represents the initial neighborhood width, represents a constant used to control the rate at which the domain width decreases, t represents the time step, Indicates the weight vector of the optimal node corresponding to the jth feature of the customer at the t+1th time step, represents the weight vector of the optimal node corresponding to the jth feature of the customer at the tth time step, represents the learning rate function;

[0028] Step S34: constructing aggregated customer churn features, the steps include:

[0029] Step S341: Calculate the minimum quantization error, using the following formula:

[0030] ;

[0031] In the formula, represents the minimum quantization error of the i-th feature of the customer, represents the i-th feature of the customer, The weight vector representing the optimal node corresponding to the i-th feature of the customer;

[0032] Step S342: local weighted regression denoising, the formula used is as follows:

[0033] ;

[0034] In the formula, represents the influence weight of the minimum quantization error of the customer's j-th feature on the minimum quantization error of the customer's ith feature, represents the Gaussian kernel function, represents the minimum quantization error of the jth feature of a customer, represents the window width of the minimum quantization error of the i-th feature of the customer, represents the minimum quantization error of the i-th feature of the customer after denoising, The total number of characteristics representing the customer;

[0035] Step S343: constructing aggregated customer churn features, specifically, using the minimum quantization error of the denoised customer features as a vector element to construct aggregated customer churn features to obtain an aggregated customer churn feature set.

[0036] Further, in step S4, the clustering process is used to classify the current customer churn, specifically, to construct a K-means clustering model to perform preliminary classification on the aggregated customer churn feature set to obtain a current churn risk data set, wherein the preliminary classification is specifically based on the current customer churn risk and is divided into five categories, including primary churn risk, secondary churn risk, tertiary churn risk, quaternary churn risk and quinary churn risk, and used as data labels, the higher the level, the greater the possibility of customer churn;

[0037] The clustering process specifically includes the following steps:

[0038] Step S41: Initialize the cluster center, specifically randomly select K data points as the initial cluster center, where K represents the number of classification types:

[0039] Step S42: Allocating clusters, the steps include:

[0040] Step S421: Calculate the distance from the data point to the cluster center using the following formula:

[0041] ;

[0042] In the formula, represents the distance from the hth data point to the kth cluster center, represents the hth data point, represents the kth cluster center;

[0043] Step S422, assigning data points to clusters, the formula used is as follows:

[0044] ;

[0045] In the formula, represents the cluster label of the h-th data point, Represents the distance from the hth data point to the kth cluster center The smallest value of k;

[0046] Step S43: Update the cluster center. The formula used is as follows:

[0047] ;

[0048] In the formula, represents the updated cluster center, Represents the set of all data points in the kth cluster;

[0049] Step S44: iterative update, specifically, continuously iteratively updating the K-means clustering model until the change in the cluster center is less than a threshold or reaches a maximum number of iterations;

[0050] Step S45: preliminary classification, specifically, constructing the K-means clustering model through the initialization of cluster centers, the allocation of clusters, the updating of cluster centers and the iterative updating, and performing preliminary classification on the aggregated customer churn feature set to obtain the current churn risk data set.

[0051] Further, in step S5, the enterprise customer churn prediction is specifically to construct an attention graph convolutional network model, perform enterprise customer churn prediction based on the current churn risk data set, obtain customer churn prediction reference data, and consider customer churn risk based on the customer churn prediction reference data;

[0052] The enterprise customer churn prediction specifically includes the following steps:

[0053] Step S51: constructing graph structure data, specifically constructing graph nodes and edges, wherein the graph nodes represent customers, and the feature of each graph node is the cluster label of the K-means clustering model, and connecting graph nodes with the same K-means clustering model cluster label and customer interaction records to obtain graph structure data;

[0054] Step S52: construct a graph convolution layer, the formula used is as follows:

[0055] ;

[0056] In the formula, represents the node feature matrix of the v+1th graph convolutional layer, represents the node feature matrix of the vth graph convolutional layer, represents the ReLU activation function, represents the normalized adjacency matrix, Represents the weight matrix of the v+1th graph convolutional layer;

[0057] Step S53: constructing a graph attention layer, the steps include:

[0058] Step S531: Calculate the attention coefficient using the following formula:

[0059] ;

[0060] In the formula, Representation Node For Node The attention coefficient, represents the ReLU activation function with leakage, Av represents the attention vector, T represents the transposition operation, Am represents the attention weight, Representation Node The characteristic vector of Representation Node The eigenvector of

[0061] Step S532: Normalize the attention coefficient. The formula used is as follows:

[0062] ;

[0063] In the formula, Representation Node For Node The normalized attention coefficient of node The set of neighbor nodes of Representation Node For Node The attention coefficient;

[0064] Step S533: Update node features, using the following formula:

[0065] ;

[0066] In the formula, Representation Node The updated feature vector,

[0067] Step S54: Calculate the model output using the following formula:

[0068] ;

[0069] In the formula, y represents the customer churn prediction reference data output by the model, represents the softmax function, represents the output weight matrix, represents the input features of the output layer, represents the output bias term;

[0070] Step S55: Design a loss function, the formula used is as follows:

[0071] ;

[0072] In the formula, Loss represents the loss function value, N represents the total number of samples, and M represents the total number of categories. Indicates that sample n belongs to the true label of category m, Represents the model's predicted probability that sample n belongs to category m;

[0073] Step S56: predicting customer churn, specifically constructing the attention graph convolutional network model through the construction graph structure data, the construction graph convolutional layer, the construction graph attention layer, the calculation model output and the design loss function, and predicting enterprise customer churn based on the current churn risk data set, and obtaining customer churn prediction reference data to comprehensively consider customer churn risk.

[0074] The enterprise customer churn prediction system based on big data provided by the present invention comprises a data acquisition module, a data preprocessing module, a customer feature aggregation module, a clustering processing module and an enterprise customer churn prediction module;

[0075] The data collection module is used for data collection, obtains a customer behavior data set through data collection, and sends the customer behavior data set to the data preprocessing module;

[0076] The data preprocessing module is used for data preprocessing, obtains a customer feature set through data preprocessing, and sends the customer feature set to the customer feature aggregation module;

[0077] The customer feature aggregation module is used for customer feature aggregation, and obtains an aggregated customer churn feature set by adopting a self-organizing neural network model to aggregate customer features, and sends the aggregated customer churn feature set to the clustering processing module;

[0078] The cluster processing module is used to perform cluster processing, obtain the current churn risk data set by using the K-means clustering model for preliminary classification, and send the current churn risk data set to the enterprise customer churn prediction module;

[0079] The enterprise customer churn prediction module is used for enterprise customer churn prediction, and obtains customer churn prediction reference data by adopting an attention graph convolutional network model to perform enterprise customer churn prediction.

[0080] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0081] (1) In view of the technical problems that traditional enterprise customer churn prediction methods rely on single customer behavior data for analysis, are difficult to deeply explore the overall characteristics of enterprise customer churn, and are easily affected by environmental fluctuations, resulting in low model prediction accuracy, this solution creatively uses a self-organizing neural network model to aggregate customer features, aggregate single features that are difficult to accurately reflect customer churn, obtain more representative aggregate features, and improve the accuracy of the model.

[0082] (2) In view of the technical problem that traditional enterprise customer churn prediction methods only consider the behavioral data of individual customers, but ignore the correlation between enterprise customers, and cannot fully capture the potential churn risk brought by customer correlation, which reduces the accuracy and reliability of the model, this solution creatively adopts the attention graph convolutional network model to predict enterprise customer churn. By constructing graph structure data based on customer interaction records and dynamically adjusting weights in combination with the attention mechanism, the prediction accuracy and reliability of the model are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 A schematic diagram of the process of the enterprise customer churn prediction method based on big data provided by the present invention;

[0084] Figure 2 A schematic diagram of the modules of the enterprise customer churn prediction system based on big data provided by the present invention;

[0085] Figure 3 This is a flow chart of customer feature aggregation in step S3;

[0086] Figure 4 This is a schematic diagram of the clustering process in step S4;

[0087] Figure 5 This is a flow chart of enterprise customer churn prediction in step S5.

[0088] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0089] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0090] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0091] Example 1, see Figure 1 The technical solution adopted by the present invention is as follows: The enterprise customer churn prediction method based on big data provided by the present invention comprises the following steps:

[0092] Step S1: data collection;

[0093] Step S2: data preprocessing;

[0094] Step S3: customer feature aggregation;

[0095] Step S4: clustering processing;

[0096] Step S5: Enterprise customer churn prediction.

[0097] Example 2, see Figure 1 and Figure 2 In step S1, the data collection is used to collect the customer behavior data required for enterprise customer churn prediction, specifically to obtain a customer behavior data set through collection. The customer behavior data set specifically includes basic customer information, transaction records, customer interaction records, customer product usage data, customer feedback data and customer activity data.

[0098] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S2, the data preprocessing is used to preprocess the customer behavior data, and specifically includes the following steps:

[0099] Step S21: data cleaning, which is used to clean the customer behavior data, specifically removing missing values ​​and duplicate values ​​from the customer behavior data set to obtain a preliminary processed data set;

[0100] Step S22: feature encoding, used to encode the preliminary processed data, specifically, using a one-hot encoding method to encode the preliminary processed data set to obtain an encoded data set;

[0101] Step S23: data normalization, which is used to normalize the coded data, specifically using a minimum-maximum normalization method to normalize the coded data set to obtain a customer feature set;

[0102] Step S24: preprocessing, specifically, performing data preprocessing on the customer behavior data set through the data cleaning, feature encoding and data normalization to obtain a customer feature set.

[0103] Example 4, see Figure 1 , Figure 2 and Figure 3 , this embodiment is based on the above embodiment, in step S3, the customer feature aggregation is used to aggregate customer features, specifically, a self-organizing neural network model is used to process the customer feature set to obtain an aggregated customer churn feature set;

[0104] The customer behavior feature aggregation specifically includes the following steps:

[0105] Step S31: Initialize weights, which are used to randomly initialize the self-organizing neural network weight vector, as shown below:

[0106] ;

[0107] In the formula, represents the weight vector of the i-th node, Represents the random initialization of the weight vector of the i-th node from a uniform distribution [0,1] ;

[0108] Step S32: Find the optimal node. The optimal node is specifically the point with the smallest Euclidean distance to the input feature. The formula used to calculate the Euclidean distance is as follows:

[0109] ;

[0110] In the formula, represents the Euclidean distance function, represents the jth feature of the customer, represents the Euclidean distance;

[0111] Step S33: Update the optimal node weight, the formula used is as follows:

[0112] ;

[0113] In the formula, It represents the neighborhood function between the optimal node and the lth node corresponding to the jth feature of the customer, Indicates the position vector of the optimal node corresponding to the jth feature of the customer, represents the position vector of the lth node, represents the initial neighborhood width, represents a constant used to control the rate at which the domain width decreases, t represents the time step, Indicates the weight vector of the optimal node corresponding to the jth feature of the customer at the t+1th time step, represents the weight vector of the optimal node corresponding to the jth feature of the customer at the tth time step, represents the learning rate function;

[0114] Step S34: constructing aggregated customer churn features, the steps include:

[0115] Step S341: Calculate the minimum quantization error, using the following formula:

[0116] ;

[0117] In the formula, represents the minimum quantization error of the i-th feature of the customer, represents the i-th feature of the customer, The weight vector representing the optimal node corresponding to the i-th feature of the customer;

[0118] Step S342: local weighted regression denoising, the formula used is as follows:

[0119] ;

[0120] In the formula, represents the influence weight of the minimum quantization error of the customer's j-th feature on the minimum quantization error of the customer's ith feature, represents the Gaussian kernel function, represents the minimum quantization error of the jth feature of a customer, represents the window width of the minimum quantization error of the i-th feature of the customer, represents the minimum quantization error of the i-th feature of the customer after denoising, The total number of characteristics representing the customer;

[0121] Step S343: constructing aggregated customer churn features, specifically, using the minimum quantization error of the denoised customer features as a vector element to construct an aggregated customer churn feature to obtain an aggregated customer churn feature set.

[0122] By performing the above operations, this solution creatively uses a self-organizing neural network model to perform a customer feature aggregation method, aggregating single features that are difficult to accurately reflect customer churn, to obtain more representative aggregate features, thereby improving the accuracy of the model.

[0123] Example 5, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S4, the clustering process is used to classify the current customer churn, specifically to construct a K-means clustering model to perform preliminary classification on the aggregated customer churn feature set to obtain a current churn risk data set. The preliminary classification is specifically based on the current customer churn risk and is divided into five categories, including primary churn risk, secondary churn risk, tertiary churn risk, quaternary churn risk and fifth churn risk, which are used as data labels. The higher the level, the greater the possibility of customer churn.

[0124] The clustering process specifically includes the following steps:

[0125] Step S41: Initialize the cluster center, specifically randomly select K data points as the initial cluster center, where K represents the number of classification types:

[0126] Step S42: Allocating clusters, the steps include:

[0127] Step S421: Calculate the distance from the data point to the cluster center using the following formula:

[0128] ;

[0129] In the formula, represents the distance from the hth data point to the kth cluster center, represents the hth data point, represents the kth cluster center;

[0130] Step S422, assigning data points to clusters, the formula used is as follows:

[0131] ;

[0132] In the formula, represents the cluster label of the h-th data point, Represents the distance from the hth data point to the kth cluster center The smallest value of k;

[0133] Step S43: Update the cluster center. The formula used is as follows:

[0134] ;

[0135] In the formula, represents the updated cluster center, Represents the set of all data points in the kth cluster;

[0136] Step S44: iterative update, specifically, continuously iteratively updating the K-means clustering model until the change in the cluster center is less than a threshold or reaches a maximum number of iterations;

[0137] Step S45: preliminary classification, specifically, constructing the K-means clustering model through the initialization of cluster centers, the allocation of clusters, the updating of cluster centers and the iterative updating, and performing preliminary classification on the aggregated customer churn feature set to obtain the current churn risk data set.

[0138] Example 6, see Figure 1 , Figure 2 and Figure 5 , this embodiment is based on the above embodiment. In step S5, the enterprise customer churn prediction is specifically to construct an attention graph convolutional network model, perform enterprise customer churn prediction based on the current churn risk data set, obtain customer churn prediction reference data, and consider customer churn risk based on the customer churn prediction reference data;

[0139] The enterprise customer churn prediction specifically includes the following steps:

[0140] Step S51: constructing graph structure data, specifically constructing graph nodes and edges, wherein the graph nodes represent customers, and the feature of each graph node is the cluster label of the K-means clustering model, and connecting graph nodes with the same K-means clustering model cluster label and customer interaction records to obtain graph structure data;

[0141] Step S52: construct a graph convolution layer, the formula used is as follows:

[0142] ;

[0143] In the formula, represents the node feature matrix of the v+1th graph convolutional layer, represents the node feature matrix of the vth graph convolutional layer, represents the ReLU activation function, represents the normalized adjacency matrix, Represents the weight matrix of the v+1th graph convolutional layer;

[0144] Step S53: constructing a graph attention layer, the steps include:

[0145] Step S531: Calculate the attention coefficient using the following formula:

[0146] ;

[0147] In the formula, Representation Node For Node The attention coefficient, represents the ReLU activation function with leakage, Av represents the attention vector, T represents the transposition operation, Am represents the attention weight, Representation Node The characteristic vector of Representation Node The eigenvector of

[0148] Step S532: Normalize the attention coefficient. The formula used is as follows:

[0149] ;

[0150] In the formula, Representation Node For Node The normalized attention coefficient of node The set of neighbor nodes of Representation Node For Node The attention coefficient;

[0151] Step S533: Update node features, using the following formula:

[0152] ;

[0153] In the formula, Representation Node The updated feature vector,

[0154] Step S54: Calculate the model output using the following formula:

[0155] ;

[0156] In the formula, y represents the customer churn prediction reference data output by the model, represents the softmax function, represents the output weight matrix, represents the input features of the output layer, represents the output bias term;

[0157] Step S55: Design a loss function, the formula used is as follows:

[0158] ;

[0159] In the formula, Loss represents the loss function value, N represents the total number of samples, and M represents the total number of categories. Indicates that sample n belongs to the true label of category m, Represents the model's predicted probability that sample n belongs to category m;

[0160] Step S56: predicting customer churn, specifically constructing the attention graph convolutional network model through the construction graph structure data, the construction graph convolutional layer, the construction graph attention layer, the calculation model output and the design loss function, and predicting enterprise customer churn based on the current churn risk data set, and obtaining customer churn prediction reference data to comprehensively consider customer churn risk.

[0161] By performing the above operations, this solution creatively adopts the attention graph convolutional network model to predict enterprise customer churn. By constructing graph structure data based on customer interaction records and dynamically adjusting weights in combination with the attention mechanism, the prediction accuracy and reliability of the model are improved.

[0162] Embodiment 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and the enterprise customer churn prediction system based on big data provided by the present invention includes a data acquisition module, a data preprocessing module, a customer feature aggregation module, a clustering processing module and an enterprise customer churn prediction module;

[0163] The data collection module is used for data collection, obtains a customer behavior data set through data collection, and sends the customer behavior data set to the data preprocessing module;

[0164] The data preprocessing module is used for data preprocessing, obtains a customer feature set through data preprocessing, and sends the customer feature set to the customer feature aggregation module;

[0165] The customer feature aggregation module is used for customer feature aggregation, and obtains an aggregated customer churn feature set by adopting a self-organizing neural network model to aggregate customer features, and sends the aggregated customer churn feature set to the clustering processing module;

[0166] The cluster processing module is used to perform cluster processing, obtain the current churn risk data set by using the K-means clustering model for preliminary classification, and send the current churn risk data set to the enterprise customer churn prediction module;

[0167] The enterprise customer churn prediction module is used for enterprise customer churn prediction, and obtains customer churn prediction reference data by adopting an attention graph convolutional network model to perform enterprise customer churn prediction.

[0168] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0169] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0170] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for predicting enterprise customer churn based on big data, characterized by: The method comprises the following steps: Step S1: data collection; Step S2: Data preprocessing, data cleaning, feature encoding and data normalization of the collected customer behavior data to obtain a customer feature set; Step S3: customer feature aggregation, specifically, using a self-organizing neural network model to aggregate customer features to obtain an aggregated customer churn feature set; Step S4: Clustering processing, specifically using the K-means clustering model for preliminary classification to obtain the current loss risk data set; Step S5: Enterprise customer churn prediction, specifically using an attention graph convolutional network model to predict enterprise customer churn and obtain customer churn prediction reference data.

2. The method for predicting enterprise customer churn based on big data according to claim 1, characterized in that: In step S5, the enterprise customer churn prediction is specifically to construct an attention graph convolutional network model, perform enterprise customer churn prediction based on the current churn risk data set, obtain customer churn prediction reference data, and consider customer churn risk based on the customer churn prediction reference data; The enterprise customer churn prediction specifically includes the following steps: Step S51: constructing graph structure data, specifically constructing graph nodes and edges, wherein the graph nodes represent customers, and the feature of each graph node is the cluster label of the K-means clustering model, and connecting graph nodes with the same K-means clustering model cluster label and customer interaction records to obtain graph structure data; Step S52: construct a graph convolution layer, the formula used is as follows: ； In the formula, represents the node feature matrix of the v+1th graph convolutional layer, represents the node feature matrix of the vth graph convolutional layer, represents the ReLU activation function, represents the normalized adjacency matrix, Represents the weight matrix of the v+1th graph convolutional layer; Step S53: constructing a graph attention layer, the steps include: Step S531: Calculate the attention coefficient using the following formula: ； In the formula, Representation Node For Node The attention coefficient, represents the ReLU activation function with leakage, Av represents the attention vector, T represents the transposition operation, Am represents the attention weight, Representation Node The characteristic vector of Representation Node The eigenvector of Step S532: Normalize the attention coefficient. The formula used is as follows: ； In the formula, Representation Node For Node The normalized attention coefficient of node The set of neighbor nodes of Representation Node For Node The attention coefficient; Step S533: Update node features, using the following formula: ； In the formula, Representation Node The updated feature vector, Step S54: Calculate the model output using the following formula: ； In the formula, y represents the customer churn prediction reference data output by the model, represents the softmax function, represents the output weight matrix, represents the input features of the output layer, represents the output bias term; Step S55: Design a loss function, the formula used is as follows: ； In the formula, Loss represents the loss function value, N represents the total number of samples, and M represents the total number of categories. Indicates that sample n belongs to the true label of category m, Represents the model's predicted probability that sample n belongs to category m; Step S56: predicting customer churn, specifically constructing the attention graph convolutional network model through the construction graph structure data, the construction graph convolutional layer, the construction graph attention layer, the calculation model output and the design loss function, and predicting enterprise customer churn based on the current churn risk data set, and obtaining customer churn prediction reference data to comprehensively consider customer churn risk.

3. The method for predicting enterprise customer churn based on big data according to claim 1, characterized in that: In step S3, the customer feature aggregation is used to aggregate customer features, specifically, the customer feature set is processed using a self-organizing neural network model to obtain an aggregated customer churn feature set; The customer behavior feature aggregation specifically includes the following steps: Step S31: Initialize weights, which are used to randomly initialize the self-organizing neural network weight vector, as shown below: ； In the formula, represents the weight vector of the i-th node, Represents the random initialization of the weight vector of the i-th node from a uniform distribution [0,1] ; Step S32: Find the optimal node. The optimal node is specifically the point with the smallest Euclidean distance to the input feature. The formula used to calculate the Euclidean distance is as follows: ； In the formula, represents the Euclidean distance function, represents the jth feature of the customer, represents the Euclidean distance; Step S33: Update the optimal node weight, the formula used is as follows: ； In the formula, It represents the neighborhood function between the optimal node and the lth node corresponding to the jth feature of the customer, Indicates the position vector of the optimal node corresponding to the jth feature of the customer, represents the position vector of the lth node, represents the initial neighborhood width, represents a constant used to control the rate at which the domain width decreases, t represents the time step, Indicates the weight vector of the optimal node corresponding to the jth feature of the customer at the t+1th time step, represents the weight vector of the optimal node corresponding to the jth feature of the customer at the tth time step, represents the learning rate function; Step S34: constructing aggregated customer churn features, the steps include: Step S341: Calculate the minimum quantization error, using the following formula: ； In the formula, represents the minimum quantization error of the i-th feature of the customer, represents the i-th feature of the customer, The weight vector representing the optimal node corresponding to the i-th feature of the customer; Step S342: local weighted regression denoising, the formula used is as follows: ； In the formula, represents the influence weight of the minimum quantization error of the customer's j-th feature on the minimum quantization error of the customer's ith feature, represents the Gaussian kernel function, represents the minimum quantization error of the jth feature of a customer, represents the window width of the minimum quantization error of the i-th feature of the customer, represents the minimum quantization error of the i-th feature of the customer after denoising, The total number of characteristics representing the customer; Step S343: constructing aggregated customer churn features, specifically, using the minimum quantization error of the denoised customer features as a vector element to construct an aggregated customer churn feature to obtain an aggregated customer churn feature set.

4. The method for predicting enterprise customer churn based on big data according to claim 1, characterized in that: In step S4, the clustering process is used to classify the current customer churn, specifically, to construct a K-means clustering model to perform preliminary classification on the aggregated customer churn feature set to obtain a current churn risk data set. The preliminary classification is specifically based on the current customer churn risk and is divided into five categories, including primary churn risk, secondary churn risk, tertiary churn risk, quaternary churn risk and quinary churn risk, which are used as data labels. The higher the level, the greater the possibility of customer churn. The clustering process specifically includes the following steps: Step S41: Initialize the cluster center, specifically randomly select K data points as the initial cluster center, where K represents the number of classification types: Step S42: Allocating clusters, the steps include: Step S421: Calculate the distance from the data point to the cluster center using the following formula: ； In the formula, represents the distance from the hth data point to the kth cluster center, represents the hth data point, represents the kth cluster center; Step S422, assigning data points to clusters, the formula used is as follows: ； In the formula, represents the cluster label of the h-th data point, Represents the distance from the hth data point to the kth cluster center The smallest value of k; Step S43: Update the cluster center. The formula used is as follows: ； In the formula, represents the updated cluster center, Represents the set of all data points in the kth cluster; Step S44: iterative update, specifically, continuously iteratively updating the K-means clustering model until the change in the cluster center is less than a threshold or reaches a maximum number of iterations; Step S45: preliminary classification, specifically, constructing the K-means clustering model through the initialization of cluster centers, the allocation of clusters, the updating of cluster centers and the iterative updating, and performing preliminary classification on the aggregated customer churn feature set to obtain the current churn risk data set.

5. The method for predicting enterprise customer churn based on big data according to claim 1, characterized in that: In step S1, the data collection is used to collect customer behavior data required for enterprise customer churn prediction, specifically to obtain a customer behavior data set through collection. The customer behavior data set specifically includes basic customer information, transaction records, customer interaction records, customer product usage data, customer feedback data and customer activity data.

6. The method for predicting enterprise customer churn based on big data according to claim 1, characterized in that: In step S2, the data preprocessing is used to preprocess the customer behavior data, and specifically includes the following steps: Step S21: data cleaning, which is used to clean the customer behavior data, specifically removing missing values ​​and duplicate values ​​from the customer behavior data set to obtain a preliminary processed data set; Step S22: feature encoding, used to encode the preliminary processed data, specifically, using a one-hot encoding method to encode the preliminary processed data set to obtain an encoded data set; Step S23: data normalization, which is used to normalize the coded data, specifically using a minimum-maximum normalization method to normalize the coded data set to obtain a customer feature set; Step S24: preprocessing, specifically, performing data preprocessing on the customer behavior data set through the data cleaning, feature encoding and data normalization to obtain a customer feature set.

7. A system for predicting enterprise customer churn based on big data, used to implement the method for predicting enterprise customer churn based on big data as claimed in any one of claims 1 to 6, characterized in that: It includes data collection module, data preprocessing module, customer feature aggregation module, clustering processing module and enterprise customer churn prediction module.

8. The enterprise customer churn prediction system based on big data according to claim 7 is characterized by: The data collection module is used for data collection, obtains a customer behavior data set through data collection, and sends the customer behavior data set to the data preprocessing module; The data preprocessing module is used for data preprocessing, obtains a customer feature set through data preprocessing, and sends the customer feature set to the customer feature aggregation module; The customer feature aggregation module is used for customer feature aggregation, and obtains an aggregated customer churn feature set by adopting a self-organizing neural network model to aggregate customer features, and sends the aggregated customer churn feature set to the clustering processing module; The cluster processing module is used to perform cluster processing, obtain the current churn risk data set by using the K-means clustering model for preliminary classification, and send the current churn risk data set to the enterprise customer churn prediction module; The enterprise customer churn prediction module is used for enterprise customer churn prediction, and obtains customer churn prediction reference data by adopting an attention graph convolutional network model to perform enterprise customer churn prediction.

Images